readme: add instructions to build with MLX

The 100MB mlx.metallib file significantly increased the app bundle size,
causing Apple's notarization service to timeout with the previous 10m limit.

.github/ISSUE_TEMPLATE/10_bug_report.yml

@@ -13,7 +13,7 @@ body:
     id: logs
     attributes:
       label: Relevant log output
-      description: Please copy and paste any relevant log output. See [Troubleshooting Guide](https://github.com/ollama/ollama/blob/main/docs/troubleshooting.md#how-to-troubleshoot-issues) for details.
+      description: Please copy and paste any relevant log output. See [Troubleshooting Guide](https://github.com/ollama/ollama/blob/main/docs/troubleshooting.mdx#how-to-troubleshoot-issues) for details.
       render: shell
     validations:
       required: false

.github/workflows/release.yaml

@@ -372,13 +372,17 @@ jobs:
           outputs: type=local,dest=dist/${{ matrix.os }}-${{ matrix.arch }}
           cache-from: type=registry,ref=${{ vars.DOCKER_REPO }}:latest
           cache-to: type=inline
+      - name: Deduplicate CUDA libraries
+        run: |
+          ./scripts/deduplicate_cuda_libs.sh dist/${{ matrix.os }}-${{ matrix.arch }}
       - run: |
           for COMPONENT in bin/* lib/ollama/*; do
             case "$COMPONENT" in
-              bin/ollama)                echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
+              bin/ollama*)               echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
               lib/ollama/*.so*)          echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
               lib/ollama/cuda_v*)        echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
               lib/ollama/vulkan*)        echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
+              lib/ollama/mlx*)           echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
               lib/ollama/cuda_jetpack5)  echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}-jetpack5.tar.in ;;
               lib/ollama/cuda_jetpack6)  echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}-jetpack6.tar.in ;;
               lib/ollama/rocm)           echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}-rocm.tar.in ;;

CMakeLists.txt

@@ -48,9 +48,10 @@ if((CMAKE_OSX_ARCHITECTURES AND NOT CMAKE_OSX_ARCHITECTURES MATCHES "arm64")
     set(GGML_CPU_ALL_VARIANTS ON)
 endif()
 
-if (CMAKE_OSX_ARCHITECTURES MATCHES "x86_64")
+if(APPLE)
     set(CMAKE_BUILD_RPATH "@loader_path")
     set(CMAKE_INSTALL_RPATH "@loader_path")
+    set(CMAKE_BUILD_WITH_INSTALL_RPATH ON)
 endif()
 
 set(OLLAMA_BUILD_DIR ${CMAKE_BINARY_DIR}/lib/ollama)
@@ -189,13 +190,21 @@ if(MLX_ENGINE)
     install(TARGETS mlx mlxc
         RUNTIME_DEPENDENCIES
             DIRECTORIES ${CUDAToolkit_BIN_DIR} ${CUDAToolkit_BIN_DIR}/x64 ${CUDAToolkit_LIBRARY_DIR}
-            PRE_INCLUDE_REGEXES cublas cublasLt cudart nvrtc cudnn nccl
+            PRE_INCLUDE_REGEXES cublas cublasLt cudart nvrtc nvrtc-builtins cudnn nccl openblas gfortran
             PRE_EXCLUDE_REGEXES ".*"
         RUNTIME DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT MLX
         LIBRARY DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT MLX
         FRAMEWORK DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT MLX
     )
 
+    # Install the Metal library for macOS arm64 (must be colocated with the binary)
+    # Metal backend is only built for arm64, not x86_64
+    if(APPLE AND CMAKE_SYSTEM_PROCESSOR STREQUAL "arm64")
+        install(FILES ${CMAKE_BINARY_DIR}/_deps/mlx-build/mlx/backend/metal/kernels/mlx.metallib
+            DESTINATION ${OLLAMA_INSTALL_DIR}
+            COMPONENT MLX)
+    endif()
+
     # Manually install cudart and cublas since they might not be picked up as direct dependencies
     if(CUDAToolkit_FOUND)
         file(GLOB CUDART_LIBS

Dockerfile

@@ -161,6 +161,9 @@ ARG GOFLAGS="'-ldflags=-w -s'"
 ENV CGO_ENABLED=1
 ARG CGO_CFLAGS
 ARG CGO_CXXFLAGS
+RUN mkdir -p dist/bin
+RUN --mount=type=cache,target=/root/.cache/go-build \
+    go build -tags mlx -trimpath -buildmode=pie -o dist/bin/ollama-mlx .
 
 FROM base AS build
 WORKDIR /go/src/github.com/ollama/ollama
@@ -182,6 +185,7 @@ COPY --from=cuda-12 dist/lib/ollama /lib/ollama/
 COPY --from=cuda-13 dist/lib/ollama /lib/ollama/
 COPY --from=vulkan  dist/lib/ollama  /lib/ollama/
 COPY --from=mlx     /go/src/github.com/ollama/ollama/dist/lib/ollama /lib/ollama/
+COPY --from=mlx     /go/src/github.com/ollama/ollama/dist/bin/ /bin/
 
 FROM --platform=linux/arm64 scratch AS arm64
 # COPY --from=cuda-11 dist/lib/ollama/ /lib/ollama/

README.md

@@ -48,7 +48,7 @@ ollama run gemma3
 
 ## Model library
 
-Ollama supports a list of models available on [ollama.com/library](https://ollama.com/library 'ollama model library')
+Ollama supports a list of models available on [ollama.com/library](https://ollama.com/library "ollama model library")
 
 Here are some example models that can be downloaded:
 
@@ -79,7 +79,7 @@ Here are some example models that can be downloaded:
 | Code Llama         | 7B         | 3.8GB | `ollama run codellama`           |
 | Llama 2 Uncensored | 7B         | 3.8GB | `ollama run llama2-uncensored`   |
 | LLaVA              | 7B         | 4.5GB | `ollama run llava`               |
-| Granite-3.3         | 8B         | 4.9GB | `ollama run granite3.3`          |
+| Granite-3.3        | 8B         | 4.9GB | `ollama run granite3.3`          |
 
 > [!NOTE]
 > You should have at least 8 GB of RAM available to run the 7B models, 16 GB to run the 13B models, and 32 GB to run the 33B models.
@@ -260,6 +260,38 @@ Finally, in a separate shell, run a model:
 ./ollama run llama3.2
 ```
 
+## Building with MLX (experimental)
+
+First build the MLX libraries:
+
+```shell
+cmake --preset MLX
+cmake --build --preset MLX --parallel
+cmake --install build --component MLX
+```
+
+Next, build the `ollama-mlx` binary, which is a separate build of the Ollama runtime with MLX support enabled (needs to be in the same directory as `ollama`):
+
+```shell
+go build -tags mlx -o ollama-mlx .
+```
+
+Finally, start the server:
+
+```
+./ollama serve
+```
+
+### Building MLX with CUDA
+
+When building with CUDA, use the preset "MLX CUDA 13" or "MLX CUDA 12" to enable CUDA with default architectures:
+
+```shell
+cmake --preset 'MLX CUDA 13'
+cmake --build --preset 'MLX CUDA 13' --parallel
+cmake --install build --component MLX
+```
+
 ## REST API
 
 Ollama has a REST API for running and managing models.
@@ -421,7 +453,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [AppFlowy](https://github.com/AppFlowy-IO/AppFlowy) (AI collaborative workspace with Ollama, cross-platform and self-hostable)
 - [Lumina](https://github.com/cushydigit/lumina.git) (A lightweight, minimal React.js frontend for interacting with Ollama servers)
 - [Tiny Notepad](https://pypi.org/project/tiny-notepad) (A lightweight, notepad-like interface to chat with ollama available on PyPI)
-- [macLlama (macOS native)](https://github.com/hellotunamayo/macLlama) (A native macOS GUI application for interacting with Ollama models, featuring a chat interface.) 
+- [macLlama (macOS native)](https://github.com/hellotunamayo/macLlama) (A native macOS GUI application for interacting with Ollama models, featuring a chat interface.)
 - [GPTranslate](https://github.com/philberndt/GPTranslate) (A fast and lightweight, AI powered desktop translation application written with Rust and Tauri. Features real-time translation with OpenAI/Azure/Ollama.)
 - [ollama launcher](https://github.com/NGC13009/ollama-launcher) (A launcher for Ollama, aiming to provide users with convenient functions such as ollama server launching, management, or configuration.)
 - [ai-hub](https://github.com/Aj-Seven/ai-hub) (AI Hub supports multiple models via API keys and Chat support via Ollama API.)
@@ -493,7 +525,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 ### Database
 
 - [pgai](https://github.com/timescale/pgai) - PostgreSQL as a vector database (Create and search embeddings from Ollama models using pgvector)
-   - [Get started guide](https://github.com/timescale/pgai/blob/main/docs/vectorizer-quick-start.md)
+  - [Get started guide](https://github.com/timescale/pgai/blob/main/docs/vectorizer-quick-start.md)
 - [MindsDB](https://github.com/mindsdb/mindsdb/blob/staging/mindsdb/integrations/handlers/ollama_handler/README.md) (Connects Ollama models with nearly 200 data platforms and apps)
 - [chromem-go](https://github.com/philippgille/chromem-go/blob/v0.5.0/embed_ollama.go) with [example](https://github.com/philippgille/chromem-go/tree/v0.5.0/examples/rag-wikipedia-ollama)
 - [Kangaroo](https://github.com/dbkangaroo/kangaroo) (AI-powered SQL client and admin tool for popular databases)
@@ -636,6 +668,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [llama.cpp](https://github.com/ggml-org/llama.cpp) project founded by Georgi Gerganov.
 
 ### Observability
+
 - [Opik](https://www.comet.com/docs/opik/cookbook/ollama) is an open-source platform to debug, evaluate, and monitor your LLM applications, RAG systems, and agentic workflows with comprehensive tracing, automated evaluations, and production-ready dashboards. Opik supports native integration to Ollama.
 - [Lunary](https://lunary.ai/docs/integrations/ollama) is the leading open-source LLM observability platform. It provides a variety of enterprise-grade features such as real-time analytics, prompt templates management, PII masking, and comprehensive agent tracing.
 - [OpenLIT](https://github.com/openlit/openlit) is an OpenTelemetry-native tool for monitoring Ollama Applications & GPUs using traces and metrics.
@@ -644,4 +677,5 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [MLflow Tracing](https://mlflow.org/docs/latest/llms/tracing/index.html#automatic-tracing) is an open source LLM observability tool with a convenient API to log and visualize traces, making it easy to debug and evaluate GenAI applications.
 
 ### Security
+
 - [Ollama Fortress](https://github.com/ParisNeo/ollama_proxy_server)

api/client.go

@@ -165,7 +165,7 @@ func (c *Client) do(ctx context.Context, method, path string, reqData, respData
 	return nil
 }
 
-const maxBufferSize = 512 * format.KiloByte
+const maxBufferSize = 8 * format.MegaByte
 
 func (c *Client) stream(ctx context.Context, method, path string, data any, fn func([]byte) error) error {
 	var buf io.Reader

cmd/cmd.go

@@ -100,7 +100,8 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 		if filename == "" {
 			// No Modelfile found - check if current directory is an image gen model
 			if imagegen.IsTensorModelDir(".") {
-				return imagegenclient.CreateModel(args[0], ".", p)
+				quantize, _ := cmd.Flags().GetString("quantize")
+				return imagegenclient.CreateModel(args[0], ".", quantize, p)
 			}
 			reader = strings.NewReader("FROM .\n")
 		} else {
@@ -464,14 +465,6 @@ func RunHandler(cmd *cobra.Command, args []string) error {
 
 	name := args[0]
 
-	// Check if this is a known image generation model (skip Show/Pull)
-	if imagegen.HasTensorLayers(name) {
-		if opts.Prompt == "" && !interactive {
-			return errors.New("image generation models require a prompt. Usage: ollama run " + name + " \"your prompt here\"")
-		}
-		return imagegen.RunCLI(cmd, name, opts.Prompt, interactive, opts.KeepAlive)
-	}
-
 	info, err := func() (*api.ShowResponse, error) {
 		showReq := &api.ShowRequest{Name: name}
 		info, err := client.Show(cmd.Context(), showReq)
@@ -533,9 +526,18 @@ func RunHandler(cmd *cobra.Command, args []string) error {
 		return generateEmbedding(cmd, name, opts.Prompt, opts.KeepAlive, truncate, dimensions)
 	}
 
+	// Check if this is an image generation model
+	if slices.Contains(info.Capabilities, model.CapabilityImageGeneration) {
+		if opts.Prompt == "" && !interactive {
+			return errors.New("image generation models require a prompt. Usage: ollama run " + name + " \"your prompt here\"")
+		}
+		return imagegen.RunCLI(cmd, name, opts.Prompt, interactive, opts.KeepAlive)
+	}
+
 	// Check for experimental flag
 	isExperimental, _ := cmd.Flags().GetBool("experimental")
 	yoloMode, _ := cmd.Flags().GetBool("experimental-yolo")
+	enableWebsearch, _ := cmd.Flags().GetBool("experimental-websearch")
 
 	if interactive {
 		if err := loadOrUnloadModel(cmd, &opts); err != nil {
@@ -565,7 +567,7 @@ func RunHandler(cmd *cobra.Command, args []string) error {
 
 		// Use experimental agent loop with tools
 		if isExperimental {
-			return xcmd.GenerateInteractive(cmd, opts.Model, opts.WordWrap, opts.Options, opts.Think, opts.HideThinking, opts.KeepAlive, yoloMode)
+			return xcmd.GenerateInteractive(cmd, opts.Model, opts.WordWrap, opts.Options, opts.Think, opts.HideThinking, opts.KeepAlive, yoloMode, enableWebsearch)
 		}
 
 		return generateInteractive(cmd, opts)
@@ -671,7 +673,11 @@ func PushHandler(cmd *cobra.Command, args []string) error {
 
 			bar, ok := bars[resp.Digest]
 			if !ok {
-				bar = progress.NewBar(fmt.Sprintf("pushing %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
+				msg := resp.Status
+				if msg == "" {
+					msg = fmt.Sprintf("pushing %s...", resp.Digest[7:19])
+				}
+				bar = progress.NewBar(msg, resp.Total, resp.Completed)
 				bars[resp.Digest] = bar
 				p.Add(resp.Digest, bar)
 			}
@@ -837,11 +843,6 @@ func DeleteHandler(cmd *cobra.Command, args []string) error {
 }
 
 func ShowHandler(cmd *cobra.Command, args []string) error {
-	// Check if this is an image generation model
-	if imagegen.HasTensorLayers(args[0]) {
-		return imagegen.Show(args[0], os.Stdout)
-	}
-
 	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		return err
@@ -1786,6 +1787,7 @@ func NewCLI() *cobra.Command {
 	runCmd.Flags().Int("dimensions", 0, "Truncate output embeddings to specified dimension (embedding models only)")
 	runCmd.Flags().Bool("experimental", false, "Enable experimental agent loop with tools")
 	runCmd.Flags().Bool("experimental-yolo", false, "Skip all tool approval prompts (use with caution)")
+	runCmd.Flags().Bool("experimental-websearch", false, "Enable web search tool in experimental mode")
 
 	// Image generation flags (width, height, steps, seed, etc.)
 	imagegen.RegisterFlags(runCmd)

cmd/cmd_test.go

@@ -1547,6 +1547,79 @@ func TestRunOptions_Copy_ThinkValueVariants(t *testing.T) {
 	}
 }
 
+func TestShowInfoImageGen(t *testing.T) {
+	var b bytes.Buffer
+	err := showInfo(&api.ShowResponse{
+		Details: api.ModelDetails{
+			Family:            "ZImagePipeline",
+			ParameterSize:     "10.3B",
+			QuantizationLevel: "FP8",
+		},
+		Capabilities: []model.Capability{model.CapabilityImageGeneration},
+		Requires:     "0.14.0",
+	}, false, &b)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	expect := "  Model\n" +
+		"    architecture    ZImagePipeline    \n" +
+		"    parameters      10.3B             \n" +
+		"    quantization    FP8               \n" +
+		"    requires        0.14.0            \n" +
+		"\n" +
+		"  Capabilities\n" +
+		"    image    \n" +
+		"\n"
+	if diff := cmp.Diff(expect, b.String()); diff != "" {
+		t.Errorf("unexpected output (-want +got):\n%s", diff)
+	}
+}
+
+func TestPushProgressMessage(t *testing.T) {
+	tests := []struct {
+		name    string
+		status  string
+		digest  string
+		wantMsg string
+	}{
+		{
+			name:    "uses status when provided",
+			status:  "uploading model",
+			digest:  "sha256:abc123456789def",
+			wantMsg: "uploading model",
+		},
+		{
+			name:    "falls back to digest when status empty",
+			status:  "",
+			digest:  "sha256:abc123456789def",
+			wantMsg: "pushing abc123456789...",
+		},
+		{
+			name:    "handles short digest gracefully",
+			status:  "",
+			digest:  "sha256:abc",
+			wantMsg: "pushing sha256:abc...",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			msg := tt.status
+			if msg == "" {
+				if len(tt.digest) >= 19 {
+					msg = fmt.Sprintf("pushing %s...", tt.digest[7:19])
+				} else {
+					msg = fmt.Sprintf("pushing %s...", tt.digest)
+				}
+			}
+			if msg != tt.wantMsg {
+				t.Errorf("got %q, want %q", msg, tt.wantMsg)
+			}
+		})
+	}
+}
+
 func TestRunOptions_Copy_Independence(t *testing.T) {
 	// Test that modifications to original don't affect copy
 	originalThink := &api.ThinkValue{Value: "original"}

docs/troubleshooting.md

@@ -1,3 +0,0 @@
-# Troubleshooting
-
-For troubleshooting, see [https://docs.ollama.com/troubleshooting](https://docs.ollama.com/troubleshooting)

middleware/anthropic.go

@@ -118,6 +118,9 @@ func AnthropicMessagesMiddleware() gin.HandlerFunc {
 			return
 		}
 
+		// Set think to nil when being used with Anthropic API to connect to tools like claude code
+		c.Set("relax_thinking", true)
+
 		var b bytes.Buffer
 		if err := json.NewEncoder(&b).Encode(chatReq); err != nil {
 			c.AbortWithStatusJSON(http.StatusInternalServerError, anthropic.NewError(http.StatusInternalServerError, err.Error()))

middleware/anthropic_test.go

@@ -582,3 +582,26 @@ func TestAnthropicWriter_ErrorFromRoutes(t *testing.T) {
 		})
 	}
 }
+
+func TestAnthropicMessagesMiddleware_SetsRelaxThinkingFlag(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	var flagSet bool
+	router := gin.New()
+	router.Use(AnthropicMessagesMiddleware())
+	router.POST("/v1/messages", func(c *gin.Context) {
+		_, flagSet = c.Get("relax_thinking")
+		c.Status(http.StatusOK)
+	})
+
+	body := `{"model": "test-model", "max_tokens": 100, "messages": [{"role": "user", "content": "Hi"}]}`
+	req, _ := http.NewRequest(http.MethodPost, "/v1/messages", strings.NewReader(body))
+	req.Header.Set("Content-Type", "application/json")
+
+	resp := httptest.NewRecorder()
+	router.ServeHTTP(resp, req)
+
+	if !flagSet {
+		t.Error("expected relax_thinking flag to be set in context")
+	}
+}

scripts/build_darwin.sh

@@ -73,7 +73,7 @@ _build_darwin() {
             MLX_CGO_CFLAGS="-O3 -I$(pwd)/$BUILD_DIR/_deps/mlx-c-src -mmacosx-version-min=14.0"
             MLX_CGO_LDFLAGS="-L$(pwd)/$BUILD_DIR/lib/ollama -lmlxc -lmlx -Wl,-rpath,@executable_path -lc++ -framework Metal -framework Foundation -framework Accelerate -mmacosx-version-min=14.0"
         fi
-        GOOS=darwin GOARCH=$ARCH CGO_ENABLED=1 CGO_CFLAGS="$MLX_CGO_CFLAGS" CGO_LDFLAGS="$MLX_CGO_LDFLAGS" go build -tags mlx -o $INSTALL_PREFIX/imagegen ./x/imagegen/cmd/engine
+        GOOS=darwin GOARCH=$ARCH CGO_ENABLED=1 CGO_CFLAGS="$MLX_CGO_CFLAGS" CGO_LDFLAGS="$MLX_CGO_LDFLAGS" go build -tags mlx -o $INSTALL_PREFIX/ollama-mlx .
         GOOS=darwin GOARCH=$ARCH CGO_ENABLED=1 go build -o $INSTALL_PREFIX .
     done
 }
@@ -82,19 +82,19 @@ _sign_darwin() {
     status "Creating universal binary..."
     mkdir -p dist/darwin
     lipo -create -output dist/darwin/ollama dist/darwin-*/ollama
-    lipo -create -output dist/darwin/imagegen dist/darwin-*/imagegen
+    lipo -create -output dist/darwin/ollama-mlx dist/darwin-*/ollama-mlx
     chmod +x dist/darwin/ollama
-    chmod +x dist/darwin/imagegen
+    chmod +x dist/darwin/ollama-mlx
 
     if [ -n "$APPLE_IDENTITY" ]; then
-        for F in dist/darwin/ollama dist/darwin-*/lib/ollama/* dist/darwin/imagegen; do
+        for F in dist/darwin/ollama dist/darwin-*/lib/ollama/* dist/darwin/ollama-mlx; do
             codesign -f --timestamp -s "$APPLE_IDENTITY" --identifier ai.ollama.ollama --options=runtime $F
         done
 
         # create a temporary zip for notarization
         TEMP=$(mktemp -u).zip
         ditto -c -k --keepParent dist/darwin/ollama "$TEMP"
-        xcrun notarytool submit "$TEMP" --wait --timeout 10m --apple-id $APPLE_ID --password $APPLE_PASSWORD --team-id $APPLE_TEAM_ID
+        xcrun notarytool submit "$TEMP" --wait --timeout 20m --apple-id $APPLE_ID --password $APPLE_PASSWORD --team-id $APPLE_TEAM_ID
         rm -f "$TEMP"
     fi
 
@@ -154,23 +154,25 @@ _build_macapp() {
     mkdir -p dist/Ollama.app/Contents/Resources
     if [ -d dist/darwin-amd64 ]; then
         lipo -create -output dist/Ollama.app/Contents/Resources/ollama dist/darwin-amd64/ollama dist/darwin-arm64/ollama
-        lipo -create -output dist/Ollama.app/Contents/Resources/imagegen dist/darwin-amd64/imagegen dist/darwin-arm64/imagegen
+        lipo -create -output dist/Ollama.app/Contents/Resources/ollama-mlx dist/darwin-amd64/ollama-mlx dist/darwin-arm64/ollama-mlx
         for F in dist/darwin-amd64/lib/ollama/*mlx*.dylib ; do
             lipo -create -output dist/darwin/$(basename $F) $F dist/darwin-arm64/lib/ollama/$(basename $F)
         done
         cp dist/darwin-*/lib/ollama/*.so dist/darwin-*/lib/ollama/*.dylib dist/Ollama.app/Contents/Resources/
         cp dist/darwin/*.dylib dist/Ollama.app/Contents/Resources/
+        # Copy MLX metallib (architecture-independent, just use arm64 version)
+        cp dist/darwin-arm64/lib/ollama/*.metallib dist/Ollama.app/Contents/Resources/ 2>/dev/null || true
     else
         cp -a dist/darwin/ollama dist/Ollama.app/Contents/Resources/ollama
         cp dist/darwin/*.so dist/darwin/*.dylib dist/Ollama.app/Contents/Resources/
     fi
-    cp -a dist/darwin/imagegen dist/Ollama.app/Contents/Resources/imagegen
+    cp -a dist/darwin/ollama-mlx dist/Ollama.app/Contents/Resources/ollama-mlx
     chmod a+x dist/Ollama.app/Contents/Resources/ollama
 
     # Sign
     if [ -n "$APPLE_IDENTITY" ]; then
         codesign -f --timestamp -s "$APPLE_IDENTITY" --identifier ai.ollama.ollama --options=runtime dist/Ollama.app/Contents/Resources/ollama
-        for lib in dist/Ollama.app/Contents/Resources/*.so dist/Ollama.app/Contents/Resources/*.dylib dist/Ollama.app/Contents/Resources/imagegen ; do
+        for lib in dist/Ollama.app/Contents/Resources/*.so dist/Ollama.app/Contents/Resources/*.dylib dist/Ollama.app/Contents/Resources/*.metallib dist/Ollama.app/Contents/Resources/ollama-mlx ; do
             codesign -f --timestamp -s "$APPLE_IDENTITY" --identifier ai.ollama.ollama --options=runtime ${lib}
         done
         codesign -f --timestamp -s "$APPLE_IDENTITY" --identifier com.electron.ollama --deep --options=runtime dist/Ollama.app
@@ -178,11 +180,11 @@ _build_macapp() {
 
     rm -f dist/Ollama-darwin.zip
     ditto -c -k --keepParent dist/Ollama.app dist/Ollama-darwin.zip
-    (cd dist/Ollama.app/Contents/Resources/; tar -cf - ollama imagegen *.so *.dylib) | gzip -9vc > dist/ollama-darwin.tgz
+    (cd dist/Ollama.app/Contents/Resources/; tar -cf - ollama ollama-mlx *.so *.dylib *.metallib 2>/dev/null) | gzip -9vc > dist/ollama-darwin.tgz
 
     # Notarize and Staple
     if [ -n "$APPLE_IDENTITY" ]; then
-        $(xcrun -f notarytool) submit dist/Ollama-darwin.zip --wait --timeout 10m --apple-id "$APPLE_ID" --password "$APPLE_PASSWORD" --team-id "$APPLE_TEAM_ID"
+        $(xcrun -f notarytool) submit dist/Ollama-darwin.zip --wait --timeout 20m --apple-id "$APPLE_ID" --password "$APPLE_PASSWORD" --team-id "$APPLE_TEAM_ID"
         rm -f dist/Ollama-darwin.zip
         $(xcrun -f stapler) staple dist/Ollama.app
         ditto -c -k --keepParent dist/Ollama.app dist/Ollama-darwin.zip
@@ -206,7 +208,7 @@ _build_macapp() {
         rm -f dist/rw*.dmg
 
         codesign -f --timestamp -s "$APPLE_IDENTITY" --identifier ai.ollama.ollama --options=runtime dist/Ollama.dmg
-        $(xcrun -f notarytool) submit dist/Ollama.dmg --wait --timeout 10m --apple-id "$APPLE_ID" --password "$APPLE_PASSWORD" --team-id "$APPLE_TEAM_ID"
+        $(xcrun -f notarytool) submit dist/Ollama.dmg --wait --timeout 20m --apple-id "$APPLE_ID" --password "$APPLE_PASSWORD" --team-id "$APPLE_TEAM_ID"
         $(xcrun -f stapler) staple dist/Ollama.dmg
     else
         echo "WARNING: Code signing disabled, this bundle will not work for upgrade testing"

scripts/build_linux.sh

@@ -48,53 +48,12 @@ if echo $PLATFORM | grep "amd64" > /dev/null; then
         .
 fi
 
-# Deduplicate CUDA libraries across mlx_* and cuda_* directories
-deduplicate_cuda_libs() {
-    local base_dir="$1"
-    echo "Deduplicating CUDA libraries in ${base_dir}..."
-
-    # Find all mlx_cuda_* directories
-    for mlx_dir in "${base_dir}"/lib/ollama/mlx_cuda_*; do
-        [ -d "${mlx_dir}" ] || continue
-
-        # Extract CUDA version (e.g., v12, v13)
-        cuda_version=$(basename "${mlx_dir}" | sed 's/mlx_cuda_//')
-        cuda_dir="${base_dir}/lib/ollama/cuda_${cuda_version}"
-
-        # Skip if corresponding cuda_* directory doesn't exist
-        [ -d "${cuda_dir}" ] || continue
-
-        echo "  Checking ${mlx_dir} against ${cuda_dir}..."
-
-        # Find all .so* files in mlx directory
-        find "${mlx_dir}" -type f -name "*.so*" | while read mlx_file; do
-            filename=$(basename "${mlx_file}")
-            cuda_file="${cuda_dir}/${filename}"
-
-            # Skip if file doesn't exist in cuda directory
-            [ -f "${cuda_file}" ] || continue
-
-            # Compare checksums
-            mlx_sum=$(sha256sum "${mlx_file}" | awk '{print $1}')
-            cuda_sum=$(sha256sum "${cuda_file}" | awk '{print $1}')
-
-            if [ "${mlx_sum}" = "${cuda_sum}" ]; then
-                echo "    Deduplicating ${filename}"
-                # Calculate relative path from mlx_dir to cuda_dir
-                rel_path="../cuda_${cuda_version}/${filename}"
-                rm -f "${mlx_file}"
-                ln -s "${rel_path}" "${mlx_file}"
-            fi
-        done
-    done
-}
-
 # Run deduplication for each platform output directory
 if echo $PLATFORM | grep "," > /dev/null ; then
-    deduplicate_cuda_libs "./dist/linux_amd64"
-    deduplicate_cuda_libs "./dist/linux_arm64"
+    $(dirname $0)/deduplicate_cuda_libs.sh "./dist/linux_amd64"
+    $(dirname $0)/deduplicate_cuda_libs.sh "./dist/linux_arm64"
 elif echo $PLATFORM | grep "amd64\|arm64" > /dev/null ; then
-    deduplicate_cuda_libs "./dist"
+    $(dirname $0)/deduplicate_cuda_libs.sh "./dist"
 fi
 
 # buildx behavior changes for single vs. multiplatform

scripts/deduplicate_cuda_libs.sh

@@ -0,0 +1,60 @@
+#!/bin/sh
+#
+# Deduplicate CUDA libraries across mlx_* and cuda_* directories
+# This script finds identical .so* files in mlx_cuda_* directories that exist
+# in corresponding cuda_* directories and replaces them with symlinks.
+#
+
+set -eu
+
+if [ $# -eq 0 ]; then
+    echo "ERROR: No directory specified" >&2
+    echo "Usage: $0 <base_directory>" >&2
+    exit 1
+fi
+
+base_dir="$1"
+
+if [ ! -d "${base_dir}" ]; then
+    echo "ERROR: Directory ${base_dir} does not exist" >&2
+    exit 1
+fi
+
+echo "Deduplicating CUDA libraries in ${base_dir}..."
+
+# Find all mlx_cuda_* directories
+for mlx_dir in "${base_dir}"/lib/ollama/mlx_cuda_*; do
+    [ -d "${mlx_dir}" ] || continue
+
+    # Extract CUDA version (e.g., v12, v13)
+    cuda_version=$(basename "${mlx_dir}" | sed 's/mlx_cuda_//')
+    cuda_dir="${base_dir}/lib/ollama/cuda_${cuda_version}"
+
+    # Skip if corresponding cuda_* directory doesn't exist
+    [ -d "${cuda_dir}" ] || continue
+
+    echo "  Checking ${mlx_dir} against ${cuda_dir}..."
+
+    # Find all .so* files in mlx directory
+    find "${mlx_dir}" -type f -name "*.so*" | while read mlx_file; do
+        filename=$(basename "${mlx_file}")
+        cuda_file="${cuda_dir}/${filename}"
+
+        # Skip if file doesn't exist in cuda directory
+        [ -f "${cuda_file}" ] || continue
+
+        # Compare checksums
+        mlx_sum=$(sha256sum "${mlx_file}" | awk '{print $1}')
+        cuda_sum=$(sha256sum "${cuda_file}" | awk '{print $1}')
+
+        if [ "${mlx_sum}" = "${cuda_sum}" ]; then
+            echo "    Deduplicating ${filename}"
+            # Calculate relative path from mlx_dir to cuda_dir
+            rel_path="../cuda_${cuda_version}/${filename}"
+            rm -f "${mlx_file}"
+            ln -s "${rel_path}" "${mlx_file}"
+        fi
+    done
+done
+
+echo "Deduplication complete"

server/manifest.go

@@ -47,16 +47,40 @@ func (m *Manifest) Remove() error {
 }
 
 func (m *Manifest) RemoveLayers() error {
-	for _, layer := range append(m.Layers, m.Config) {
-		if layer.Digest != "" {
-			if err := layer.Remove(); errors.Is(err, os.ErrNotExist) {
-				slog.Debug("layer does not exist", "digest", layer.Digest)
-			} else if err != nil {
-				return err
+	ms, err := Manifests(true)
+	if err != nil {
+		return err
+	}
+
+	// Build set of digests still in use by other manifests
+	inUse := make(map[string]struct{})
+	for _, other := range ms {
+		for _, layer := range append(other.Layers, other.Config) {
+			if layer.Digest != "" {
+				inUse[layer.Digest] = struct{}{}
 			}
 		}
 	}
 
+	// Remove layers not used by any other manifest
+	for _, layer := range append(m.Layers, m.Config) {
+		if layer.Digest == "" {
+			continue
+		}
+		if _, used := inUse[layer.Digest]; used {
+			continue
+		}
+		blob, err := GetBlobsPath(layer.Digest)
+		if err != nil {
+			return err
+		}
+		if err := os.Remove(blob); errors.Is(err, os.ErrNotExist) {
+			slog.Debug("layer does not exist", "digest", layer.Digest)
+		} else if err != nil {
+			return err
+		}
+	}
+
 	return nil
 }
 

server/routes.go

@@ -1124,6 +1124,15 @@ func GetModelInfo(req api.ShowRequest) (*api.ShowResponse, error) {
 		QuantizationLevel: m.Config.FileType,
 	}
 
+	// For image generation models, populate details from imagegen package
+	if slices.Contains(m.Capabilities(), model.CapabilityImageGeneration) {
+		if info, err := imagegen.GetModelInfo(name.String()); err == nil {
+			modelDetails.Family = info.Architecture
+			modelDetails.ParameterSize = format.HumanNumber(uint64(info.ParameterCount))
+			modelDetails.QuantizationLevel = info.Quantization
+		}
+	}
+
 	if req.System != "" {
 		m.System = req.System
 	}
@@ -1206,6 +1215,10 @@ func GetModelInfo(req api.ShowRequest) (*api.ShowResponse, error) {
 		return resp, nil
 	}
 
+	if slices.Contains(m.Capabilities(), model.CapabilityImageGeneration) {
+		return resp, nil
+	}
+
 	kvData, tensors, err := getModelData(m.ModelPath, req.Verbose)
 	if err != nil {
 		return nil, err
@@ -2059,8 +2072,14 @@ func (s *Server) ChatHandler(c *gin.Context) {
 		}
 	} else {
 		if req.Think != nil && req.Think.Bool() {
-			c.JSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("%q does not support thinking", req.Model)})
-			return
+			// Set think to nil when being used with Anthropic API to connect to tools like claude code
+			if _, ok := c.Get("relax_thinking"); ok {
+				slog.Warn("model does not support thinking, relaxing thinking to nil", "model", req.Model)
+				req.Think = nil
+			} else {
+				c.JSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("%q does not support thinking", req.Model)})
+				return
+			}
 		}
 	}
 

x/README.md

@@ -1,24 +0,0 @@
-# Experimental Features 
-
-## MLX Backend
-
-We're working on a new experimental backend based on the [MLX project](https://github.com/ml-explore/mlx)
-
-Support is currently limited to MacOS and Linux with CUDA GPUs.  We're looking to add support for Windows CUDA soon, and other GPU vendors.  To build:
-
-```
-cmake --preset MLX
-cmake --build --preset MLX --parallel
-cmake --install --component MLX
-go build -tags mlx .
-```
-
-On linux, use the preset "MLX CUDA 13" or "MLX CUDA 12" to enable CUDA with the default Ollama NVIDIA GPU architectures enabled. 
-
-## Image Generation
-
-Based on the experimental MLX backend, we're working on adding imagegen support.  After running the cmake commands above:
-
-```
-go build -o imagegen ./x/imagegen/cmd/engine
-```

x/agent/approval.go

@@ -41,6 +41,7 @@ var optionLabels = []string{
 var toolDisplayNames = map[string]string{
 	"bash":       "Bash",
 	"web_search": "Web Search",
+	"web_fetch":  "Web Fetch",
 }
 
 // ToolDisplayName returns the human-readable display name for a tool.
@@ -565,6 +566,16 @@ func formatToolDisplay(toolName string, args map[string]any) string {
 		}
 	}
 
+	// For web fetch, show URL and internet notice
+	if toolName == "web_fetch" {
+		if url, ok := args["url"].(string); ok {
+			sb.WriteString(fmt.Sprintf("Tool: %s\n", displayName))
+			sb.WriteString(fmt.Sprintf("URL: %s\n", url))
+			sb.WriteString("Uses internet via ollama.com")
+			return sb.String()
+		}
+	}
+
 	// Generic display
 	sb.WriteString(fmt.Sprintf("Tool: %s", displayName))
 	if len(args) > 0 {
@@ -1017,6 +1028,16 @@ func FormatApprovalResult(toolName string, args map[string]any, result ApprovalR
 		}
 	}
 
+	if toolName == "web_fetch" {
+		if url, ok := args["url"].(string); ok {
+			// Truncate long URLs
+			if len(url) > 50 {
+				url = url[:47] + "..."
+			}
+			return fmt.Sprintf("\033[1m%s:\033[0m %s: %s", label, displayName, url)
+		}
+	}
+
 	return fmt.Sprintf("\033[1m%s:\033[0m %s", label, displayName)
 }
 

x/cmd/run.go

@@ -9,6 +9,7 @@ import (
 	"net/url"
 	"os"
 	"os/signal"
+	"slices"
 	"strings"
 	"syscall"
 	"time"
@@ -24,6 +25,14 @@ import (
 	"github.com/ollama/ollama/x/tools"
 )
 
+// MultilineState tracks the state of multiline input
+type MultilineState int
+
+const (
+	MultilineNone MultilineState = iota
+	MultilineSystem
+)
+
 // Tool output capping constants
 const (
 	// localModelTokenLimit is the token limit for local models (smaller context).
@@ -130,6 +139,7 @@ type RunOptions struct {
 	KeepAlive    *api.Duration
 	Think        *api.ThinkValue
 	HideThinking bool
+	Verbose      bool
 
 	// Agent fields (managed externally for session persistence)
 	Tools    *tools.Registry
@@ -178,6 +188,7 @@ func Chat(ctx context.Context, opts RunOptions) (*api.Message, error) {
 	var thinkTagClosed bool = false
 	var pendingToolCalls []api.ToolCall
 	var consecutiveErrors int // Track consecutive 500 errors for retry limit
+	var latest api.ChatResponse
 
 	role := "assistant"
 	messages := opts.Messages
@@ -187,6 +198,7 @@ func Chat(ctx context.Context, opts RunOptions) (*api.Message, error) {
 			p.StopAndClear()
 		}
 
+		latest = response
 		role = response.Message.Role
 		if response.Message.Thinking != "" && !opts.HideThinking {
 			if !thinkTagOpened {
@@ -483,6 +495,10 @@ func Chat(ctx context.Context, opts RunOptions) (*api.Message, error) {
 		fmt.Println()
 	}
 
+	if opts.Verbose {
+		latest.Summary()
+	}
+
 	return &api.Message{Role: role, Thinking: thinkingContent.String(), Content: fullResponse.String()}, nil
 }
 
@@ -634,7 +650,8 @@ func checkModelCapabilities(ctx context.Context, modelName string) (supportsTool
 // GenerateInteractive runs an interactive agent session.
 // This is called from cmd.go when --experimental flag is set.
 // If yoloMode is true, all tool approvals are skipped.
-func GenerateInteractive(cmd *cobra.Command, modelName string, wordWrap bool, options map[string]any, think *api.ThinkValue, hideThinking bool, keepAlive *api.Duration, yoloMode bool) error {
+// If enableWebsearch is true, the web search tool is registered.
+func GenerateInteractive(cmd *cobra.Command, modelName string, wordWrap bool, options map[string]any, think *api.ThinkValue, hideThinking bool, keepAlive *api.Duration, yoloMode bool, enableWebsearch bool) error {
 	scanner, err := readline.New(readline.Prompt{
 		Prompt:         ">>> ",
 		AltPrompt:      "... ",
@@ -660,6 +677,12 @@ func GenerateInteractive(cmd *cobra.Command, modelName string, wordWrap bool, op
 	if supportsTools {
 		toolRegistry = tools.DefaultRegistry()
 
+		// Register web search and web fetch tools if enabled via flag
+		if enableWebsearch {
+			toolRegistry.RegisterWebSearch()
+			toolRegistry.RegisterWebFetch()
+		}
+
 		if toolRegistry.Has("bash") {
 			fmt.Fprintln(os.Stderr)
 			fmt.Fprintln(os.Stderr, "This experimental version of Ollama has the \033[1mbash\033[0m tool enabled.")
@@ -667,6 +690,11 @@ func GenerateInteractive(cmd *cobra.Command, modelName string, wordWrap bool, op
 			fmt.Fprintln(os.Stderr)
 		}
 
+		if toolRegistry.Has("web_search") || toolRegistry.Has("web_fetch") {
+			fmt.Fprintln(os.Stderr, "The \033[1mWeb Search\033[0m and \033[1mWeb Fetch\033[0m tools are enabled. Models can search and fetch web content via ollama.com.")
+			fmt.Fprintln(os.Stderr)
+		}
+
 		if yoloMode {
 			fmt.Fprintf(os.Stderr, "\033[1mwarning:\033[0m yolo mode - all tool approvals will be skipped\n")
 		}
@@ -677,6 +705,9 @@ func GenerateInteractive(cmd *cobra.Command, modelName string, wordWrap bool, op
 
 	var messages []api.Message
 	var sb strings.Builder
+	var format string
+	var system string
+	var multiline MultilineState = MultilineNone
 
 	for {
 		line, err := scanner.Readline()
@@ -688,13 +719,39 @@ func GenerateInteractive(cmd *cobra.Command, modelName string, wordWrap bool, op
 			if line == "" {
 				fmt.Println("\nUse Ctrl + d or /bye to exit.")
 			}
+			scanner.Prompt.UseAlt = false
 			sb.Reset()
+			multiline = MultilineNone
 			continue
 		case err != nil:
 			return err
 		}
 
 		switch {
+		case multiline != MultilineNone:
+			// check if there's a multiline terminating string
+			before, ok := strings.CutSuffix(line, `"""`)
+			sb.WriteString(before)
+			if !ok {
+				fmt.Fprintln(&sb)
+				continue
+			}
+
+			switch multiline {
+			case MultilineSystem:
+				system = sb.String()
+				newMessage := api.Message{Role: "system", Content: system}
+				if len(messages) > 0 && messages[len(messages)-1].Role == "system" {
+					messages[len(messages)-1] = newMessage
+				} else {
+					messages = append(messages, newMessage)
+				}
+				fmt.Println("Set system message.")
+				sb.Reset()
+			}
+
+			multiline = MultilineNone
+			scanner.Prompt.UseAlt = false
 		case strings.HasPrefix(line, "/exit"), strings.HasPrefix(line, "/bye"):
 			return nil
 		case strings.HasPrefix(line, "/clear"):
@@ -707,6 +764,10 @@ func GenerateInteractive(cmd *cobra.Command, modelName string, wordWrap bool, op
 			continue
 		case strings.HasPrefix(line, "/help"), strings.HasPrefix(line, "/?"):
 			fmt.Fprintln(os.Stderr, "Available Commands:")
+			fmt.Fprintln(os.Stderr, "  /set            Set session variables")
+			fmt.Fprintln(os.Stderr, "  /show           Show model information")
+			fmt.Fprintln(os.Stderr, "  /load           Load a different model")
+			fmt.Fprintln(os.Stderr, "  /save           Save session as a model")
 			fmt.Fprintln(os.Stderr, "  /tools          Show available tools and approvals")
 			fmt.Fprintln(os.Stderr, "  /clear          Clear session context and approvals")
 			fmt.Fprintln(os.Stderr, "  /bye            Exit")
@@ -716,6 +777,303 @@ func GenerateInteractive(cmd *cobra.Command, modelName string, wordWrap bool, op
 			fmt.Fprintln(os.Stderr, "  Ctrl+O          Expand last tool output")
 			fmt.Fprintln(os.Stderr, "")
 			continue
+		case strings.HasPrefix(line, "/set"):
+			args := strings.Fields(line)
+			if len(args) > 1 {
+				switch args[1] {
+				case "history":
+					scanner.HistoryEnable()
+				case "nohistory":
+					scanner.HistoryDisable()
+				case "wordwrap":
+					wordWrap = true
+					fmt.Println("Set 'wordwrap' mode.")
+				case "nowordwrap":
+					wordWrap = false
+					fmt.Println("Set 'nowordwrap' mode.")
+				case "verbose":
+					if err := cmd.Flags().Set("verbose", "true"); err != nil {
+						return err
+					}
+					fmt.Println("Set 'verbose' mode.")
+				case "quiet":
+					if err := cmd.Flags().Set("verbose", "false"); err != nil {
+						return err
+					}
+					fmt.Println("Set 'quiet' mode.")
+				case "think":
+					thinkValue := api.ThinkValue{Value: true}
+					var maybeLevel string
+					if len(args) > 2 {
+						maybeLevel = args[2]
+					}
+					if maybeLevel != "" {
+						thinkValue.Value = maybeLevel
+					}
+					think = &thinkValue
+					// Check if model supports thinking
+					if client, err := api.ClientFromEnvironment(); err == nil {
+						if resp, err := client.Show(cmd.Context(), &api.ShowRequest{Model: modelName}); err == nil {
+							if !slices.Contains(resp.Capabilities, model.CapabilityThinking) {
+								fmt.Fprintf(os.Stderr, "warning: model %q does not support thinking output\n", modelName)
+							}
+						}
+					}
+					if maybeLevel != "" {
+						fmt.Printf("Set 'think' mode to '%s'.\n", maybeLevel)
+					} else {
+						fmt.Println("Set 'think' mode.")
+					}
+				case "nothink":
+					think = &api.ThinkValue{Value: false}
+					// Check if model supports thinking
+					if client, err := api.ClientFromEnvironment(); err == nil {
+						if resp, err := client.Show(cmd.Context(), &api.ShowRequest{Model: modelName}); err == nil {
+							if !slices.Contains(resp.Capabilities, model.CapabilityThinking) {
+								fmt.Fprintf(os.Stderr, "warning: model %q does not support thinking output\n", modelName)
+							}
+						}
+					}
+					fmt.Println("Set 'nothink' mode.")
+				case "format":
+					if len(args) < 3 || args[2] != "json" {
+						fmt.Println("Invalid or missing format. For 'json' mode use '/set format json'")
+					} else {
+						format = args[2]
+						fmt.Printf("Set format to '%s' mode.\n", args[2])
+					}
+				case "noformat":
+					format = ""
+					fmt.Println("Disabled format.")
+				case "parameter":
+					if len(args) < 4 {
+						fmt.Println("Usage: /set parameter <name> <value>")
+						continue
+					}
+					params := args[3:]
+					fp, err := api.FormatParams(map[string][]string{args[2]: params})
+					if err != nil {
+						fmt.Printf("Couldn't set parameter: %q\n", err)
+						continue
+					}
+					fmt.Printf("Set parameter '%s' to '%s'\n", args[2], strings.Join(params, ", "))
+					options[args[2]] = fp[args[2]]
+				case "system":
+					if len(args) < 3 {
+						fmt.Println("Usage: /set system <message> or /set system \"\"\"<multi-line message>\"\"\"")
+						continue
+					}
+
+					multiline = MultilineSystem
+
+					line := strings.Join(args[2:], " ")
+					line, ok := strings.CutPrefix(line, `"""`)
+					if !ok {
+						multiline = MultilineNone
+					} else {
+						// only cut suffix if the line is multiline
+						line, ok = strings.CutSuffix(line, `"""`)
+						if ok {
+							multiline = MultilineNone
+						}
+					}
+
+					sb.WriteString(line)
+					if multiline != MultilineNone {
+						scanner.Prompt.UseAlt = true
+						continue
+					}
+
+					system = sb.String()
+					newMessage := api.Message{Role: "system", Content: sb.String()}
+					// Check if the slice is not empty and the last message is from 'system'
+					if len(messages) > 0 && messages[len(messages)-1].Role == "system" {
+						// Replace the last message
+						messages[len(messages)-1] = newMessage
+					} else {
+						messages = append(messages, newMessage)
+					}
+					fmt.Println("Set system message.")
+					sb.Reset()
+					continue
+				default:
+					fmt.Printf("Unknown command '/set %s'. Type /? for help\n", args[1])
+				}
+			} else {
+				fmt.Println("Usage: /set <parameter|system|history|format|wordwrap|think|verbose> [value]")
+			}
+			continue
+		case strings.HasPrefix(line, "/show"):
+			args := strings.Fields(line)
+			if len(args) > 1 {
+				client, err := api.ClientFromEnvironment()
+				if err != nil {
+					fmt.Println("error: couldn't connect to ollama server")
+					continue
+				}
+				req := &api.ShowRequest{
+					Name:    modelName,
+					Options: options,
+				}
+				resp, err := client.Show(cmd.Context(), req)
+				if err != nil {
+					fmt.Println("error: couldn't get model")
+					continue
+				}
+
+				switch args[1] {
+				case "info":
+					fmt.Fprintf(os.Stderr, "  Model\n")
+					fmt.Fprintf(os.Stderr, "    %-16s %s\n", "Name", modelName)
+					if resp.Details.Family != "" {
+						fmt.Fprintf(os.Stderr, "    %-16s %s\n", "Family", resp.Details.Family)
+					}
+					if resp.Details.ParameterSize != "" {
+						fmt.Fprintf(os.Stderr, "    %-16s %s\n", "Parameter Size", resp.Details.ParameterSize)
+					}
+					if resp.Details.QuantizationLevel != "" {
+						fmt.Fprintf(os.Stderr, "    %-16s %s\n", "Quantization", resp.Details.QuantizationLevel)
+					}
+					if len(resp.Capabilities) > 0 {
+						caps := make([]string, len(resp.Capabilities))
+						for i, c := range resp.Capabilities {
+							caps[i] = string(c)
+						}
+						fmt.Fprintf(os.Stderr, "    %-16s %s\n", "Capabilities", strings.Join(caps, ", "))
+					}
+					fmt.Fprintln(os.Stderr)
+				case "license":
+					if resp.License == "" {
+						fmt.Println("No license was specified for this model.")
+					} else {
+						fmt.Println(resp.License)
+					}
+				case "modelfile":
+					fmt.Println(resp.Modelfile)
+				case "parameters":
+					fmt.Println("Model defined parameters:")
+					if resp.Parameters == "" {
+						fmt.Println("  No additional parameters were specified.")
+					} else {
+						for _, l := range strings.Split(resp.Parameters, "\n") {
+							fmt.Printf("  %s\n", l)
+						}
+					}
+					if len(options) > 0 {
+						fmt.Println("\nUser defined parameters:")
+						for k, v := range options {
+							fmt.Printf("  %-30s %v\n", k, v)
+						}
+					}
+				case "system":
+					switch {
+					case system != "":
+						fmt.Println(system + "\n")
+					case resp.System != "":
+						fmt.Println(resp.System + "\n")
+					default:
+						fmt.Println("No system message was specified for this model.")
+					}
+				case "template":
+					if resp.Template != "" {
+						fmt.Println(resp.Template)
+					} else {
+						fmt.Println("No prompt template was specified for this model.")
+					}
+				default:
+					fmt.Printf("Unknown command '/show %s'. Type /? for help\n", args[1])
+				}
+			} else {
+				fmt.Println("Usage: /show <info|license|modelfile|parameters|system|template>")
+			}
+			continue
+		case strings.HasPrefix(line, "/load"):
+			args := strings.Fields(line)
+			if len(args) != 2 {
+				fmt.Println("Usage: /load <modelname>")
+				continue
+			}
+			newModelName := args[1]
+			fmt.Printf("Loading model '%s'\n", newModelName)
+
+			// Create progress spinner
+			p := progress.NewProgress(os.Stderr)
+			spinner := progress.NewSpinner("")
+			p.Add("", spinner)
+
+			// Get client
+			client, err := api.ClientFromEnvironment()
+			if err != nil {
+				p.StopAndClear()
+				fmt.Println("error: couldn't connect to ollama server")
+				continue
+			}
+
+			// Check if model exists and get its info
+			info, err := client.Show(cmd.Context(), &api.ShowRequest{Model: newModelName})
+			if err != nil {
+				p.StopAndClear()
+				if strings.Contains(err.Error(), "not found") {
+					fmt.Printf("Couldn't find model '%s'\n", newModelName)
+				} else {
+					fmt.Printf("error: %v\n", err)
+				}
+				continue
+			}
+
+			// For cloud models, no need to preload
+			if info.RemoteHost == "" {
+				// Preload the model by sending an empty generate request
+				req := &api.GenerateRequest{
+					Model: newModelName,
+					Think: think,
+				}
+				err = client.Generate(cmd.Context(), req, func(r api.GenerateResponse) error {
+					return nil
+				})
+				if err != nil {
+					p.StopAndClear()
+					if strings.Contains(err.Error(), "not found") {
+						fmt.Printf("Couldn't find model '%s'\n", newModelName)
+					} else if strings.Contains(err.Error(), "does not support thinking") {
+						fmt.Printf("error: %v\n", err)
+					} else {
+						fmt.Printf("error loading model: %v\n", err)
+					}
+					continue
+				}
+			}
+
+			p.StopAndClear()
+			modelName = newModelName
+			messages = []api.Message{}
+			approval.Reset()
+			continue
+		case strings.HasPrefix(line, "/save"):
+			args := strings.Fields(line)
+			if len(args) != 2 {
+				fmt.Println("Usage: /save <modelname>")
+				continue
+			}
+			client, err := api.ClientFromEnvironment()
+			if err != nil {
+				fmt.Println("error: couldn't connect to ollama server")
+				continue
+			}
+			req := &api.CreateRequest{
+				Model:      args[1],
+				From:       modelName,
+				Parameters: options,
+				Messages:   messages,
+			}
+			fn := func(resp api.ProgressResponse) error { return nil }
+			err = client.Create(cmd.Context(), req, fn)
+			if err != nil {
+				fmt.Printf("error: %v\n", err)
+				continue
+			}
+			fmt.Printf("Created new model '%s'\n", args[1])
+			continue
 		case strings.HasPrefix(line, "/"):
 			fmt.Printf("Unknown command '%s'. Type /? for help\n", strings.Fields(line)[0])
 			continue
@@ -723,14 +1081,16 @@ func GenerateInteractive(cmd *cobra.Command, modelName string, wordWrap bool, op
 			sb.WriteString(line)
 		}
 
-		if sb.Len() > 0 {
+		if sb.Len() > 0 && multiline == MultilineNone {
 			newMessage := api.Message{Role: "user", Content: sb.String()}
 			messages = append(messages, newMessage)
 
+			verbose, _ := cmd.Flags().GetBool("verbose")
 			opts := RunOptions{
 				Model:        modelName,
 				Messages:     messages,
 				WordWrap:     wordWrap,
+				Format:       format,
 				Options:      options,
 				Think:        think,
 				HideThinking: hideThinking,
@@ -738,6 +1098,7 @@ func GenerateInteractive(cmd *cobra.Command, modelName string, wordWrap bool, op
 				Tools:        toolRegistry,
 				Approval:     approval,
 				YoloMode:     yoloMode,
+				Verbose:      verbose,
 			}
 
 			assistant, err := Chat(cmd.Context(), opts)

x/imagegen/README.md

@@ -234,3 +234,17 @@ ollama create z-image
 3. Copy config files (*.json) as config layers
 4. Write manifest
 ```
+
+## FP8 Quantization
+
+Z-Image supports FP8 quantization to reduce memory usage by ~50% while maintaining image quality.
+
+### Usage
+
+```bash
+cd ./weights/Z-Image-Turbo
+ollama create z-image-fp8 --quantize fp8
+```
+
+This quantizes weights during import. The resulting model will be ~15GB instead of ~31GB.
+

x/imagegen/api/handler.go

@@ -1,10 +1,8 @@
 package api
 
 import (
-	"encoding/base64"
 	"fmt"
 	"net/http"
-	"os"
 	"strconv"
 	"strings"
 	"time"
@@ -101,10 +99,10 @@ func handleStreamingResponse(c *gin.Context, runner llm.LlamaServer, req llm.Com
 	c.Header("Cache-Control", "no-cache")
 	c.Header("Connection", "keep-alive")
 
-	var imagePath string
+	var imageBase64 string
 	err := runner.Completion(c.Request.Context(), req, func(resp llm.CompletionResponse) {
 		if resp.Done {
-			imagePath = extractPath(resp.Content)
+			imageBase64 = extractBase64(resp.Content)
 		} else {
 			progress := parseProgress(resp.Content)
 			if progress.Total > 0 {
@@ -118,14 +116,14 @@ func handleStreamingResponse(c *gin.Context, runner llm.LlamaServer, req llm.Com
 		return
 	}
 
-	c.SSEvent("done", buildResponse(imagePath, format))
+	c.SSEvent("done", buildResponse(imageBase64, format))
 }
 
 func handleNonStreamingResponse(c *gin.Context, runner llm.LlamaServer, req llm.CompletionRequest, format string) {
-	var imagePath string
+	var imageBase64 string
 	err := runner.Completion(c.Request.Context(), req, func(resp llm.CompletionResponse) {
 		if resp.Done {
-			imagePath = extractPath(resp.Content)
+			imageBase64 = extractBase64(resp.Content)
 		}
 	})
 	if err != nil {
@@ -133,7 +131,7 @@ func handleNonStreamingResponse(c *gin.Context, runner llm.LlamaServer, req llm.
 		return
 	}
 
-	c.JSON(http.StatusOK, buildResponse(imagePath, format))
+	c.JSON(http.StatusOK, buildResponse(imageBase64, format))
 }
 
 func parseSize(size string) (int32, int32) {
@@ -152,9 +150,9 @@ func parseSize(size string) (int32, int32) {
 	return int32(w), int32(h)
 }
 
-func extractPath(content string) string {
-	if idx := strings.Index(content, "Image saved to: "); idx >= 0 {
-		return strings.TrimSpace(content[idx+16:])
+func extractBase64(content string) string {
+	if strings.HasPrefix(content, "IMAGE_BASE64:") {
+		return content[13:]
 	}
 	return ""
 }
@@ -165,23 +163,21 @@ func parseProgress(content string) ImageProgressEvent {
 	return ImageProgressEvent{Step: step, Total: total}
 }
 
-func buildResponse(imagePath, format string) ImageGenerationResponse {
+func buildResponse(imageBase64, format string) ImageGenerationResponse {
 	resp := ImageGenerationResponse{
 		Created: time.Now().Unix(),
 		Data:    make([]ImageData, 1),
 	}
 
-	if imagePath == "" {
+	if imageBase64 == "" {
 		return resp
 	}
 
 	if format == "url" {
-		resp.Data[0].URL = "file://" + imagePath
+		// URL format not supported when using base64 transfer
+		resp.Data[0].B64JSON = imageBase64
 	} else {
-		data, err := os.ReadFile(imagePath)
-		if err == nil {
-			resp.Data[0].B64JSON = base64.StdEncoding.EncodeToString(data)
-		}
+		resp.Data[0].B64JSON = imageBase64
 	}
 
 	return resp

x/imagegen/cache/teacache.go

@@ -0,0 +1,197 @@
+//go:build mlx
+
+// Package cache provides caching mechanisms for diffusion model inference.
+package cache
+
+import (
+	"github.com/ollama/ollama/x/imagegen/mlx"
+)
+
+// TeaCache implements Timestep Embedding Aware Caching for diffusion models.
+// It caches the transformer output and reuses it when timestep values
+// are similar between consecutive steps.
+//
+// For CFG (classifier-free guidance), it caches pos and neg predictions
+// separately and always computes CFG fresh to avoid error amplification.
+//
+// Reference: "Timestep Embedding Tells: It's Time to Cache for Video Diffusion Model"
+// https://github.com/ali-vilab/TeaCache
+type TeaCache struct {
+	// Cached transformer output from last computed step (non-CFG mode)
+	cachedOutput *mlx.Array
+
+	// Cached CFG outputs (pos and neg separately)
+	cachedPosOutput *mlx.Array
+	cachedNegOutput *mlx.Array
+
+	// Previous timestep value for difference calculation
+	prevTimestep float32
+
+	// Accumulated difference for rescaling
+	accumulatedDiff float32
+
+	// Configuration
+	threshold      float32 // Threshold for recomputation decision
+	rescaleFactor  float32 // Model-specific rescaling factor
+	skipEarlySteps int     // Number of early steps to never cache
+
+	// Statistics
+	cacheHits   int
+	cacheMisses int
+}
+
+// TeaCacheConfig holds configuration for TeaCache.
+type TeaCacheConfig struct {
+	// Threshold for recomputation. Lower = more cache hits, potential quality loss.
+	// Recommended: 0.05-0.15 for image models
+	Threshold float32
+
+	// Rescale factor to adjust timestep embedding differences.
+	// Model-specific, typically 1.0-2.0
+	RescaleFactor float32
+
+	// SkipEarlySteps: number of early steps to always compute (never cache).
+	// Set to 2-3 for CFG mode to preserve structure. 0 = no skipping.
+	SkipEarlySteps int
+}
+
+// DefaultTeaCacheConfig returns default configuration for TeaCache.
+func DefaultTeaCacheConfig() *TeaCacheConfig {
+	return &TeaCacheConfig{
+		Threshold:     0.1,
+		RescaleFactor: 1.0,
+	}
+}
+
+// NewTeaCache creates a new TeaCache instance.
+func NewTeaCache(cfg *TeaCacheConfig) *TeaCache {
+	if cfg == nil {
+		cfg = DefaultTeaCacheConfig()
+	}
+	return &TeaCache{
+		threshold:      cfg.Threshold,
+		rescaleFactor:  cfg.RescaleFactor,
+		skipEarlySteps: cfg.SkipEarlySteps,
+	}
+}
+
+// ShouldCompute determines if we should compute the full forward pass
+// or reuse the cached output based on timestep similarity.
+//
+// Algorithm:
+// 1. First step always computes
+// 2. Subsequent steps compare |currTimestep - prevTimestep| * rescaleFactor
+// 3. If accumulated difference > threshold, compute new output
+// 4. Otherwise, reuse cached output
+func (tc *TeaCache) ShouldCompute(step int, timestep float32) bool {
+	// Always compute early steps (critical for structure)
+	// Check both regular cache and CFG cache
+	hasCachedOutput := tc.cachedOutput != nil || tc.HasCFGCache()
+	if step < tc.skipEarlySteps || step == 0 || !hasCachedOutput {
+		return true
+	}
+
+	// Compute absolute difference between current and previous timestep
+	diff := timestep - tc.prevTimestep
+	if diff < 0 {
+		diff = -diff
+	}
+
+	// Apply rescaling factor
+	scaledDiff := diff * tc.rescaleFactor
+
+	// Accumulate difference (helps track drift over multiple cached steps)
+	tc.accumulatedDiff += scaledDiff
+
+	// Decision based on accumulated difference
+	if tc.accumulatedDiff > tc.threshold {
+		tc.accumulatedDiff = 0 // Reset accumulator
+		return true
+	}
+
+	return false
+}
+
+// UpdateCache stores the computed output for potential reuse (non-CFG mode).
+func (tc *TeaCache) UpdateCache(output *mlx.Array, timestep float32) {
+	// Free previous cached output
+	if tc.cachedOutput != nil {
+		tc.cachedOutput.Free()
+	}
+
+	// Store new cached values
+	tc.cachedOutput = output
+	tc.prevTimestep = timestep
+	tc.cacheMisses++
+}
+
+// UpdateCFGCache stores pos and neg outputs separately for CFG mode.
+// This allows CFG to be computed fresh each step, avoiding error amplification.
+func (tc *TeaCache) UpdateCFGCache(posOutput, negOutput *mlx.Array, timestep float32) {
+	// Free previous cached outputs
+	if tc.cachedPosOutput != nil {
+		tc.cachedPosOutput.Free()
+	}
+	if tc.cachedNegOutput != nil {
+		tc.cachedNegOutput.Free()
+	}
+
+	// Store new cached values
+	tc.cachedPosOutput = posOutput
+	tc.cachedNegOutput = negOutput
+	tc.prevTimestep = timestep
+	tc.cacheMisses++
+}
+
+// GetCached returns the cached output (non-CFG mode).
+func (tc *TeaCache) GetCached() *mlx.Array {
+	tc.cacheHits++
+	return tc.cachedOutput
+}
+
+// GetCFGCached returns cached pos and neg outputs for CFG mode.
+func (tc *TeaCache) GetCFGCached() (pos, neg *mlx.Array) {
+	tc.cacheHits++
+	return tc.cachedPosOutput, tc.cachedNegOutput
+}
+
+// HasCFGCache returns true if CFG cache is available.
+func (tc *TeaCache) HasCFGCache() bool {
+	return tc.cachedPosOutput != nil && tc.cachedNegOutput != nil
+}
+
+// Arrays returns all arrays that should be kept alive.
+func (tc *TeaCache) Arrays() []*mlx.Array {
+	var arrays []*mlx.Array
+	if tc.cachedOutput != nil {
+		arrays = append(arrays, tc.cachedOutput)
+	}
+	if tc.cachedPosOutput != nil {
+		arrays = append(arrays, tc.cachedPosOutput)
+	}
+	if tc.cachedNegOutput != nil {
+		arrays = append(arrays, tc.cachedNegOutput)
+	}
+	return arrays
+}
+
+// Stats returns cache hit/miss statistics.
+func (tc *TeaCache) Stats() (hits, misses int) {
+	return tc.cacheHits, tc.cacheMisses
+}
+
+// Free releases all cached arrays.
+func (tc *TeaCache) Free() {
+	if tc.cachedOutput != nil {
+		tc.cachedOutput.Free()
+		tc.cachedOutput = nil
+	}
+	if tc.cachedPosOutput != nil {
+		tc.cachedPosOutput.Free()
+		tc.cachedPosOutput = nil
+	}
+	if tc.cachedNegOutput != nil {
+		tc.cachedNegOutput.Free()
+		tc.cachedNegOutput = nil
+	}
+}

x/imagegen/cli.go

@@ -44,62 +44,64 @@ func DefaultOptions() ImageGenOptions {
 	}
 }
 
-// Show displays information about an image generation model.
-func Show(modelName string, w io.Writer) error {
-	manifest, err := LoadManifest(modelName)
-	if err != nil {
-		return fmt.Errorf("failed to load manifest: %w", err)
-	}
-
-	// Count total size
-	var totalSize int64
-	for _, layer := range manifest.Manifest.Layers {
-		if layer.MediaType == "application/vnd.ollama.image.tensor" {
-			totalSize += layer.Size
-		}
-	}
-
-	// Read model_index.json for architecture
-	var architecture string
-	if data, err := manifest.ReadConfig("model_index.json"); err == nil {
-		var index struct {
-			Architecture string `json:"architecture"`
-		}
-		if json.Unmarshal(data, &index) == nil {
-			architecture = index.Architecture
-		}
-	}
-
-	// Estimate parameter count from total size (assuming BF16 = 2 bytes per param)
-	paramCount := totalSize / 2
-	paramStr := formatParamCount(paramCount)
-
-	// Print Model info
-	fmt.Fprintln(w, "  Model")
-	if architecture != "" {
-		fmt.Fprintf(w, "    %-20s %s\n", "architecture", architecture)
-	}
-	fmt.Fprintf(w, "    %-20s %s\n", "parameters", paramStr)
-	fmt.Fprintf(w, "    %-20s %s\n", "quantization", "BF16")
-	fmt.Fprintln(w)
-
-	// Print Capabilities
-	fmt.Fprintln(w, "  Capabilities")
-	fmt.Fprintf(w, "    %s\n", "image")
-	fmt.Fprintln(w)
-
-	return nil
+// ModelInfo contains metadata about an image generation model.
+type ModelInfo struct {
+	Architecture   string
+	ParameterCount int64
+	Quantization   string
 }
 
-// formatParamCount formats parameter count as human-readable string.
-func formatParamCount(count int64) string {
-	if count >= 1_000_000_000 {
-		return fmt.Sprintf("%.1fB", float64(count)/1_000_000_000)
+// GetModelInfo returns metadata about an image generation model.
+func GetModelInfo(modelName string) (*ModelInfo, error) {
+	manifest, err := LoadManifest(modelName)
+	if err != nil {
+		return nil, fmt.Errorf("failed to load manifest: %w", err)
 	}
-	if count >= 1_000_000 {
-		return fmt.Sprintf("%.1fM", float64(count)/1_000_000)
+
+	info := &ModelInfo{}
+
+	// Read model_index.json for architecture, parameter count, and quantization
+	if data, err := manifest.ReadConfig("model_index.json"); err == nil {
+		var index struct {
+			Architecture   string `json:"architecture"`
+			ParameterCount int64  `json:"parameter_count"`
+			Quantization   string `json:"quantization"`
+		}
+		if json.Unmarshal(data, &index) == nil {
+			info.Architecture = index.Architecture
+			info.ParameterCount = index.ParameterCount
+			info.Quantization = index.Quantization
+		}
 	}
-	return fmt.Sprintf("%d", count)
+
+	// Fallback: detect quantization from tensor names if not in config
+	if info.Quantization == "" {
+		for _, layer := range manifest.Manifest.Layers {
+			if strings.HasSuffix(layer.Name, ".weight_scale") {
+				info.Quantization = "FP8"
+				break
+			}
+		}
+		if info.Quantization == "" {
+			info.Quantization = "BF16"
+		}
+	}
+
+	// Fallback: estimate parameter count if not in config
+	if info.ParameterCount == 0 {
+		var totalSize int64
+		for _, layer := range manifest.Manifest.Layers {
+			if layer.MediaType == "application/vnd.ollama.image.tensor" {
+				if !strings.HasSuffix(layer.Name, "_scale") && !strings.HasSuffix(layer.Name, "_qbias") {
+					totalSize += layer.Size
+				}
+			}
+		}
+		// Assume BF16 (2 bytes/param) as rough estimate
+		info.ParameterCount = totalSize / 2
+	}
+
+	return info, nil
 }
 
 // RegisterFlags adds image generation flags to the given command.
@@ -121,11 +123,6 @@ func RegisterFlags(cmd *cobra.Command) {
 // Returns true if it handled the request, false if the caller should continue with normal flow.
 // Supports flags: --width, --height, --steps, --seed, --negative
 func RunCLI(cmd *cobra.Command, name string, prompt string, interactive bool, keepAlive *api.Duration) error {
-	// Verify it's a valid image gen model
-	if ResolveModelName(name) == "" {
-		return fmt.Errorf("unknown image generation model: %s", name)
-	}
-
 	// Get options from flags (with env var defaults)
 	opts := DefaultOptions()
 	if cmd != nil && cmd.Flags() != nil {
@@ -183,8 +180,7 @@ func generateImageWithOptions(cmd *cobra.Command, modelName, prompt string, keep
 	p.Add("", spinner)
 
 	var stepBar *progress.StepBar
-	var imagePath string
-
+	var imageBase64 string
 	err = client.Generate(cmd.Context(), req, func(resp api.GenerateResponse) error {
 		content := resp.Response
 
@@ -203,11 +199,9 @@ func generateImageWithOptions(cmd *cobra.Command, modelName, prompt string, keep
 			return nil
 		}
 
-		// Handle final response with image path
-		if resp.Done && strings.Contains(content, "Image saved to:") {
-			if idx := strings.Index(content, "Image saved to: "); idx >= 0 {
-				imagePath = strings.TrimSpace(content[idx+16:])
-			}
+		// Handle final response with base64 image data
+		if resp.Done && strings.HasPrefix(content, "IMAGE_BASE64:") {
+			imageBase64 = content[13:]
 		}
 
 		return nil
@@ -218,9 +212,27 @@ func generateImageWithOptions(cmd *cobra.Command, modelName, prompt string, keep
 		return err
 	}
 
-	if imagePath != "" {
-		displayImageInTerminal(imagePath)
-		fmt.Printf("Image saved to: %s\n", imagePath)
+	if imageBase64 != "" {
+		// Decode base64 and save to CWD
+		imageData, err := base64.StdEncoding.DecodeString(imageBase64)
+		if err != nil {
+			return fmt.Errorf("failed to decode image: %w", err)
+		}
+
+		// Create filename from prompt
+		safeName := sanitizeFilename(prompt)
+		if len(safeName) > 50 {
+			safeName = safeName[:50]
+		}
+		timestamp := time.Now().Format("20060102-150405")
+		filename := fmt.Sprintf("%s-%s.png", safeName, timestamp)
+
+		if err := os.WriteFile(filename, imageData, 0o644); err != nil {
+			return fmt.Errorf("failed to save image: %w", err)
+		}
+
+		displayImageInTerminal(filename)
+		fmt.Printf("Image saved to: %s\n", filename)
 	}
 
 	return nil
@@ -306,7 +318,7 @@ func runInteractive(cmd *cobra.Command, modelName string, keepAlive *api.Duratio
 		p.Add("", spinner)
 
 		var stepBar *progress.StepBar
-		var imagePath string
+		var imageBase64 string
 
 		err = client.Generate(cmd.Context(), req, func(resp api.GenerateResponse) error {
 			content := resp.Response
@@ -326,11 +338,9 @@ func runInteractive(cmd *cobra.Command, modelName string, keepAlive *api.Duratio
 				return nil
 			}
 
-			// Handle final response with image path
-			if resp.Done && strings.Contains(content, "Image saved to:") {
-				if idx := strings.Index(content, "Image saved to: "); idx >= 0 {
-					imagePath = strings.TrimSpace(content[idx+16:])
-				}
+			// Handle final response with base64 image data
+			if resp.Done && strings.HasPrefix(content, "IMAGE_BASE64:") {
+				imageBase64 = content[13:]
 			}
 
 			return nil
@@ -342,25 +352,30 @@ func runInteractive(cmd *cobra.Command, modelName string, keepAlive *api.Duratio
 			continue
 		}
 
-		// Copy image to current directory with descriptive name
-		if imagePath != "" {
+		// Save image to current directory with descriptive name
+		if imageBase64 != "" {
+			// Decode base64 image data
+			imageData, err := base64.StdEncoding.DecodeString(imageBase64)
+			if err != nil {
+				fmt.Fprintf(os.Stderr, "Error decoding image: %v\n", err)
+				continue
+			}
+
 			// Create filename from prompt (sanitized)
 			safeName := sanitizeFilename(line)
 			if len(safeName) > 50 {
 				safeName = safeName[:50]
 			}
 			timestamp := time.Now().Format("20060102-150405")
-			newName := fmt.Sprintf("%s-%s.png", safeName, timestamp)
+			filename := fmt.Sprintf("%s-%s.png", safeName, timestamp)
 
-			// Copy file to CWD
-			if err := copyFile(imagePath, newName); err != nil {
-				fmt.Fprintf(os.Stderr, "Error saving to current directory: %v\n", err)
-				displayImageInTerminal(imagePath)
-				fmt.Printf("Image saved to: %s\n", imagePath)
-			} else {
-				displayImageInTerminal(newName)
-				fmt.Printf("Image saved to: %s\n", newName)
+			if err := os.WriteFile(filename, imageData, 0o644); err != nil {
+				fmt.Fprintf(os.Stderr, "Error saving image: %v\n", err)
+				continue
 			}
+
+			displayImageInTerminal(filename)
+			fmt.Printf("Image saved to: %s\n", filename)
 		}
 
 		fmt.Println()
@@ -381,24 +396,6 @@ func sanitizeFilename(s string) string {
 	return result.String()
 }
 
-// copyFile copies a file from src to dst.
-func copyFile(src, dst string) error {
-	sourceFile, err := os.Open(src)
-	if err != nil {
-		return err
-	}
-	defer sourceFile.Close()
-
-	destFile, err := os.Create(dst)
-	if err != nil {
-		return err
-	}
-	defer destFile.Close()
-
-	_, err = io.Copy(destFile, sourceFile)
-	return err
-}
-
 // printInteractiveHelp prints help for interactive mode commands.
 func printInteractiveHelp(opts ImageGenOptions) {
 	fmt.Fprintln(os.Stderr, "Commands:")
@@ -509,10 +506,7 @@ func displayImageInTerminal(imagePath string) bool {
 		// Send in chunks for large images
 		const chunkSize = 4096
 		for i := 0; i < len(encoded); i += chunkSize {
-			end := i + chunkSize
-			if end > len(encoded) {
-				end = len(encoded)
-			}
+			end := min(i+chunkSize, len(encoded))
 			chunk := encoded[i:end]
 
 			if i == 0 {

x/imagegen/client/create.go

@@ -29,9 +29,10 @@ const MinOllamaVersion = "0.14.0"
 
 // CreateModel imports a tensor-based model from a local directory.
 // This creates blobs and manifest directly on disk, bypassing the HTTP API.
+// If quantize is "fp8", weights will be quantized to mxfp8 format during import.
 //
 // TODO (jmorganca): Replace with API-based creation when promoted to production.
-func CreateModel(modelName, modelDir string, p *progress.Progress) error {
+func CreateModel(modelName, modelDir, quantize string, p *progress.Progress) error {
 	if !imagegen.IsTensorModelDir(modelDir) {
 		return fmt.Errorf("%s is not an image generation model directory (model_index.json not found)", modelDir)
 	}
@@ -58,18 +59,77 @@ func CreateModel(modelName, modelDir string, p *progress.Progress) error {
 
 	// Create tensor layer callback for individual tensors
 	// name is path-style: "component/tensor_name"
-	createTensorLayer := func(r io.Reader, name, dtype string, shape []int32) (imagegen.LayerInfo, error) {
+	// When quantize is true, returns multiple layers (weight + scales)
+	createTensorLayer := func(r io.Reader, name, dtype string, shape []int32, doQuantize bool) ([]imagegen.LayerInfo, error) {
+		if doQuantize {
+			// Check if quantization is supported
+			if !QuantizeSupported() {
+				return nil, fmt.Errorf("quantization requires MLX support")
+			}
+
+			// Quantize the tensor (affine mode returns weight, scales, qbiases)
+			qweightData, scalesData, qbiasData, _, _, _, err := quantizeTensor(r, name, dtype, shape)
+			if err != nil {
+				return nil, fmt.Errorf("failed to quantize %s: %w", name, err)
+			}
+
+			// Create layer for quantized weight
+			weightLayer, err := server.NewLayer(bytes.NewReader(qweightData), server.MediaTypeImageTensor)
+			if err != nil {
+				return nil, err
+			}
+
+			// Create layer for scales (use _scale suffix convention)
+			scalesLayer, err := server.NewLayer(bytes.NewReader(scalesData), server.MediaTypeImageTensor)
+			if err != nil {
+				return nil, err
+			}
+
+			layers := []imagegen.LayerInfo{
+				{
+					Digest:    weightLayer.Digest,
+					Size:      weightLayer.Size,
+					MediaType: weightLayer.MediaType,
+					Name:      name, // Keep original name for weight
+				},
+				{
+					Digest:    scalesLayer.Digest,
+					Size:      scalesLayer.Size,
+					MediaType: scalesLayer.MediaType,
+					Name:      name + "_scale", // Add _scale suffix
+				},
+			}
+
+			// Add qbiases layer if present (affine mode)
+			if qbiasData != nil {
+				qbiasLayer, err := server.NewLayer(bytes.NewReader(qbiasData), server.MediaTypeImageTensor)
+				if err != nil {
+					return nil, err
+				}
+				layers = append(layers, imagegen.LayerInfo{
+					Digest:    qbiasLayer.Digest,
+					Size:      qbiasLayer.Size,
+					MediaType: qbiasLayer.MediaType,
+					Name:      name + "_qbias", // Add _qbias suffix
+				})
+			}
+
+			return layers, nil
+		}
+
+		// Non-quantized path: just create a single layer
 		layer, err := server.NewLayer(r, server.MediaTypeImageTensor)
 		if err != nil {
-			return imagegen.LayerInfo{}, err
+			return nil, err
 		}
-		layer.Name = name
 
-		return imagegen.LayerInfo{
-			Digest:    layer.Digest,
-			Size:      layer.Size,
-			MediaType: layer.MediaType,
-			Name:      name,
+		return []imagegen.LayerInfo{
+			{
+				Digest:    layer.Digest,
+				Size:      layer.Size,
+				MediaType: layer.MediaType,
+				Name:      name,
+			},
 		}, nil
 	}
 
@@ -119,7 +179,7 @@ func CreateModel(modelName, modelDir string, p *progress.Progress) error {
 		p.Add("imagegen", spinner)
 	}
 
-	err := imagegen.CreateModel(modelName, modelDir, createLayer, createTensorLayer, writeManifest, progressFn)
+	err := imagegen.CreateModel(modelName, modelDir, quantize, createLayer, createTensorLayer, writeManifest, progressFn)
 	spinner.Stop()
 	if err != nil {
 		return err

x/imagegen/client/quantize.go

@@ -0,0 +1,120 @@
+//go:build mlx
+
+package client
+
+import (
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+
+	"github.com/ollama/ollama/x/imagegen/mlx"
+)
+
+// quantizeTensor loads a tensor from safetensors format, quantizes it to affine int8,
+// and returns safetensors data for the quantized weights, scales, and biases.
+// Uses MLX's native SaveSafetensors to ensure correct dtype handling (especially uint32 for quantized weights).
+func quantizeTensor(r io.Reader, name, dtype string, shape []int32) (qweightData, scalesData, qbiasData []byte, qweightShape, scalesShape, qbiasShape []int32, err error) {
+	tmpDir := ensureTempDir()
+
+	// Read safetensors data to a temp file (LoadSafetensorsNative needs a path)
+	tmpFile, err := os.CreateTemp(tmpDir, "quant-input-*.safetensors")
+	if err != nil {
+		return nil, nil, nil, nil, nil, nil, fmt.Errorf("failed to create temp file: %w", err)
+	}
+	tmpPath := tmpFile.Name()
+	defer os.Remove(tmpPath)
+
+	if _, err := io.Copy(tmpFile, r); err != nil {
+		tmpFile.Close()
+		return nil, nil, nil, nil, nil, nil, fmt.Errorf("failed to write temp file: %w", err)
+	}
+	tmpFile.Close()
+
+	// Load the tensor using MLX's native loader
+	st, err := mlx.LoadSafetensorsNative(tmpPath)
+	if err != nil {
+		return nil, nil, nil, nil, nil, nil, fmt.Errorf("failed to load safetensors: %w", err)
+	}
+	defer st.Free()
+
+	// Get the tensor (it's stored as "data" in our minimal safetensors format)
+	arr := st.Get("data")
+	if arr == nil {
+		return nil, nil, nil, nil, nil, nil, fmt.Errorf("tensor 'data' not found in safetensors")
+	}
+
+	// Convert to BFloat16 if needed (quantize expects float type)
+	if arr.Dtype() != mlx.DtypeBFloat16 && arr.Dtype() != mlx.DtypeFloat32 && arr.Dtype() != mlx.DtypeFloat16 {
+		arr = mlx.AsType(arr, mlx.DtypeBFloat16)
+		mlx.Eval(arr)
+	}
+
+	// Quantize with affine mode: group_size=32, bits=8
+	// Note: mxfp8 mode doesn't have matmul kernels in MLX, affine mode does
+	qweight, scales, qbiases := mlx.Quantize(arr, 32, 8, "affine")
+
+	// Eval and make contiguous for data access
+	qweight = mlx.Contiguous(qweight)
+	scales = mlx.Contiguous(scales)
+	if qbiases != nil {
+		qbiases = mlx.Contiguous(qbiases)
+		mlx.Eval(qweight, scales, qbiases)
+	} else {
+		mlx.Eval(qweight, scales)
+	}
+
+	// Get shapes
+	qweightShape = qweight.Shape()
+	scalesShape = scales.Shape()
+
+	// Save quantized weight using MLX's native safetensors (correctly handles uint32 dtype)
+	qweightPath := filepath.Join(tmpDir, "qweight.safetensors")
+	defer os.Remove(qweightPath)
+	if err := mlx.SaveSafetensors(qweightPath, map[string]*mlx.Array{"data": qweight}); err != nil {
+		return nil, nil, nil, nil, nil, nil, fmt.Errorf("failed to save quantized weight: %w", err)
+	}
+	qweightData, err = os.ReadFile(qweightPath)
+	if err != nil {
+		return nil, nil, nil, nil, nil, nil, fmt.Errorf("failed to read quantized weight: %w", err)
+	}
+
+	// Save scales using MLX's native safetensors
+	scalesPath := filepath.Join(tmpDir, "scales.safetensors")
+	defer os.Remove(scalesPath)
+	if err := mlx.SaveSafetensors(scalesPath, map[string]*mlx.Array{"data": scales}); err != nil {
+		return nil, nil, nil, nil, nil, nil, fmt.Errorf("failed to save scales: %w", err)
+	}
+	scalesData, err = os.ReadFile(scalesPath)
+	if err != nil {
+		return nil, nil, nil, nil, nil, nil, fmt.Errorf("failed to read scales: %w", err)
+	}
+
+	// Affine mode returns qbiases for zero-point offset
+	if qbiases != nil {
+		qbiasShape = qbiases.Shape()
+		qbiasPath := filepath.Join(tmpDir, "qbias.safetensors")
+		defer os.Remove(qbiasPath)
+		if err := mlx.SaveSafetensors(qbiasPath, map[string]*mlx.Array{"data": qbiases}); err != nil {
+			return nil, nil, nil, nil, nil, nil, fmt.Errorf("failed to save qbiases: %w", err)
+		}
+		qbiasData, err = os.ReadFile(qbiasPath)
+		if err != nil {
+			return nil, nil, nil, nil, nil, nil, fmt.Errorf("failed to read qbiases: %w", err)
+		}
+	}
+
+	return qweightData, scalesData, qbiasData, qweightShape, scalesShape, qbiasShape, nil
+}
+
+// QuantizeSupported returns true if quantization is supported (MLX build)
+func QuantizeSupported() bool {
+	return true
+}
+
+// ensureTempDir creates the temp directory for quantization if it doesn't exist
+func ensureTempDir() string {
+	tmpDir := filepath.Join(os.TempDir(), "ollama-quantize")
+	os.MkdirAll(tmpDir, 0755)
+	return tmpDir
+}

x/imagegen/client/quantize_stub.go

@@ -0,0 +1,18 @@
+//go:build !mlx
+
+package client
+
+import (
+	"fmt"
+	"io"
+)
+
+// quantizeTensor is not available without MLX
+func quantizeTensor(r io.Reader, name, dtype string, shape []int32) (qweightData, scalesData, qbiasData []byte, qweightShape, scalesShape, qbiasShape []int32, err error) {
+	return nil, nil, nil, nil, nil, nil, fmt.Errorf("quantization requires MLX support (build with mlx tag)")
+}
+
+// QuantizeSupported returns false when MLX is not available
+func QuantizeSupported() bool {
+	return false
+}

x/imagegen/cmd/engine/main.go

@@ -11,9 +11,11 @@ import (
 	"os"
 	"path/filepath"
 	"runtime/pprof"
+	"strings"
 
 	"github.com/ollama/ollama/x/imagegen/mlx"
 	"github.com/ollama/ollama/x/imagegen/models/gemma3"
+	"github.com/ollama/ollama/x/imagegen/models/glm_image"
 	"github.com/ollama/ollama/x/imagegen/models/gpt_oss"
 	"github.com/ollama/ollama/x/imagegen/models/llama"
 	"github.com/ollama/ollama/x/imagegen/models/qwen_image"
@@ -61,12 +63,16 @@ func main() {
 
 	// Legacy mode flags
 	zimageFlag := flag.Bool("zimage", false, "Z-Image generation")
+	glmImageFlag := flag.Bool("glm-image", false, "GLM-Image generation")
 	qwenImage := flag.Bool("qwen-image", false, "Qwen-Image text-to-image generation")
 	qwenImageEdit := flag.Bool("qwen-image-edit", false, "Qwen-Image-Edit image editing")
 	var inputImages stringSlice
 	flag.Var(&inputImages, "input-image", "Input image for image editing (can be specified multiple times)")
 	negativePrompt := flag.String("negative-prompt", "", "Negative prompt for CFG (empty = no CFG, matching Python)")
 	cfgScale := flag.Float64("cfg-scale", 4.0, "CFG scale for image editing")
+	teaCache := flag.Bool("teacache", false, "Enable TeaCache for faster inference")
+	teaCacheThreshold := flag.Float64("teacache-threshold", 0.1, "TeaCache threshold (lower = more aggressive caching)")
+	fusedQKV := flag.Bool("fused-qkv", false, "Enable fused QKV projection for faster attention")
 
 	flag.Parse()
 
@@ -99,13 +105,44 @@ func main() {
 		}
 		var img *mlx.Array
 		img, err = m.GenerateFromConfig(context.Background(), &zimage.GenerateConfig{
-			Prompt:      *prompt,
-			Width:       int32(*width),
-			Height:      int32(*height),
-			Steps:       *steps,
-			Seed:        *seed,
-			CapturePath: *gpuCapture,
-			LayerCache:  *layerCache,
+			Prompt:            *prompt,
+			NegativePrompt:    *negativePrompt,
+			CFGScale:          float32(*cfgScale),
+			Width:             int32(*width),
+			Height:            int32(*height),
+			Steps:             *steps,
+			Seed:              *seed,
+			CapturePath:       *gpuCapture,
+			TeaCache:          *teaCache,
+			TeaCacheThreshold: float32(*teaCacheThreshold),
+			FusedQKV:          *fusedQKV,
+		})
+		if err == nil {
+			err = saveImageArray(img, *out)
+		}
+	case *glmImageFlag:
+		m := &glm_image.Model{}
+		// Use LoadFromPath if model path looks like a directory, otherwise use Load (ollama manifest)
+		var loadErr error
+		if strings.HasPrefix(*modelPath, ".") || strings.HasPrefix(*modelPath, "/") {
+			loadErr = m.LoadFromPath(*modelPath)
+		} else {
+			loadErr = m.Load(*modelPath)
+		}
+		if loadErr != nil {
+			log.Fatal(loadErr)
+		}
+		var img *mlx.Array
+		img, err = m.GenerateFromConfig(context.Background(), &glm_image.GenerateConfig{
+			Prompt:          *prompt,
+			Width:           int32(*width),
+			Height:          int32(*height),
+			Steps:           *steps,
+			Seed:            *seed,
+			Temperature:     float32(*temperature),
+			TopP:            float32(*topP),
+			GuidanceScale:   float32(*cfgScale),
+			MaxVisualTokens: int32(*maxTokens),
 		})
 		if err == nil {
 			err = saveImageArray(img, *out)

x/imagegen/create.go

@@ -40,13 +40,15 @@ type ManifestWriter func(modelName string, config LayerInfo, layers []LayerInfo)
 
 // CreateModel imports an image generation model from a directory.
 // Stores each tensor as a separate blob for fine-grained deduplication.
+// If quantize is "fp8", linear weights in transformer/text_encoder are quantized to mxfp8 format.
 // Layer creation and manifest writing are done via callbacks to avoid import cycles.
-func CreateModel(modelName, modelDir string, createLayer LayerCreator, createTensorLayer TensorLayerCreator, writeManifest ManifestWriter, fn func(status string)) error {
+func CreateModel(modelName, modelDir, quantize string, createLayer LayerCreator, createTensorLayer QuantizingTensorLayerCreator, writeManifest ManifestWriter, fn func(status string)) error {
 	var layers []LayerInfo
 	var configLayer LayerInfo
+	var totalParams int64 // Count parameters from original tensor shapes
 
 	// Components to process - extract individual tensors from each
-	components := []string{"text_encoder", "transformer", "vae"}
+	components := []string{"text_encoder", "transformer", "vae", "vision_language_encoder"}
 
 	for _, component := range components {
 		componentDir := filepath.Join(modelDir, component)
@@ -74,7 +76,11 @@ func CreateModel(modelName, modelDir string, createLayer LayerCreator, createTen
 			}
 
 			tensorNames := extractor.ListTensors()
-			fn(fmt.Sprintf("importing %s/%s (%d tensors)", component, entry.Name(), len(tensorNames)))
+			quantizeMsg := ""
+			if quantize == "fp8" && component != "vae" {
+				quantizeMsg = ", quantizing to fp8"
+			}
+			fn(fmt.Sprintf("importing %s/%s (%d tensors%s)", component, entry.Name(), len(tensorNames), quantizeMsg))
 
 			for _, tensorName := range tensorNames {
 				td, err := extractor.GetTensor(tensorName)
@@ -83,16 +89,30 @@ func CreateModel(modelName, modelDir string, createLayer LayerCreator, createTen
 					return fmt.Errorf("failed to get tensor %s: %w", tensorName, err)
 				}
 
+				// Count parameters from original tensor shape
+				if len(td.Shape) > 0 {
+					numElements := int64(1)
+					for _, dim := range td.Shape {
+						numElements *= int64(dim)
+					}
+					totalParams += numElements
+				}
+
 				// Store as minimal safetensors format (88 bytes header overhead)
 				// This enables native mmap loading via mlx_load_safetensors
 				// Use path-style name: "component/tensor_name"
 				fullName := component + "/" + tensorName
-				layer, err := createTensorLayer(td.SafetensorsReader(), fullName, td.Dtype, td.Shape)
+
+				// Determine if this tensor should be quantized
+				doQuantize := quantize == "fp8" && ShouldQuantize(tensorName, component)
+
+				// createTensorLayer returns multiple layers if quantizing (weight + scales)
+				newLayers, err := createTensorLayer(td.SafetensorsReader(), fullName, td.Dtype, td.Shape, doQuantize)
 				if err != nil {
 					extractor.Close()
 					return fmt.Errorf("failed to create layer for %s: %w", fullName, err)
 				}
-				layers = append(layers, layer)
+				layers = append(layers, newLayers...)
 			}
 
 			extractor.Close()
@@ -106,10 +126,13 @@ func CreateModel(modelName, modelDir string, createLayer LayerCreator, createTen
 		"text_encoder/generation_config.json",
 		"transformer/config.json",
 		"vae/config.json",
+		"vision_language_encoder/config.json",
 		"scheduler/scheduler_config.json",
 		"tokenizer/tokenizer.json",
 		"tokenizer/tokenizer_config.json",
 		"tokenizer/vocab.json",
+		"processor/tokenizer.json",        // GLM-Image main tokenizer
+		"processor/tokenizer_config.json", // GLM-Image tokenizer config
 	}
 
 	for _, cfgPath := range configFiles {
@@ -122,7 +145,7 @@ func CreateModel(modelName, modelDir string, createLayer LayerCreator, createTen
 
 		var r io.Reader
 
-		// For model_index.json, normalize to Ollama format
+		// For model_index.json, normalize to Ollama format and add metadata
 		if cfgPath == "model_index.json" {
 			data, err := os.ReadFile(fullPath)
 			if err != nil {
@@ -141,6 +164,16 @@ func CreateModel(modelName, modelDir string, createLayer LayerCreator, createTen
 			}
 			delete(cfg, "_diffusers_version")
 
+			// Add parameter count (counted from tensor shapes during import)
+			cfg["parameter_count"] = totalParams
+
+			// Add quantization info
+			if quantize == "fp8" {
+				cfg["quantization"] = "FP8"
+			} else {
+				cfg["quantization"] = "BF16"
+			}
+
 			data, err = json.MarshalIndent(cfg, "", "    ")
 			if err != nil {
 				return fmt.Errorf("failed to marshal %s: %w", cfgPath, err)

x/imagegen/image.go

@@ -60,9 +60,12 @@ func ArrayToImage(arr *mlx.Array) (*image.RGBA, error) {
 	}
 
 	// Transform to [H, W, C] for image conversion
-	img := mlx.Squeeze(arr, 0)
-	img = mlx.Transpose(img, 1, 2, 0)
-	img = mlx.Contiguous(img)
+	// Free intermediate arrays to avoid memory leak
+	squeezed := mlx.Squeeze(arr, 0)
+	transposed := mlx.Transpose(squeezed, 1, 2, 0)
+	squeezed.Free()
+	img := mlx.Contiguous(transposed)
+	transposed.Free()
 	mlx.Eval(img)
 
 	imgShape := img.Shape()

x/imagegen/imagegen.md

@@ -0,0 +1,19 @@
+# Image generation models (experimental)
+
+Experimental image generation models are available for **macOS** in Ollama:
+
+## Available models
+
+- [Z-Image-Turbo](https://ollama.com/x/z-image-turbo)
+
+```
+ollama run x/z-image-turbo
+```
+
+> **Note**: [`x`](https://ollama.com/x) is a username on ollama.com where the maintainer team uploads experimental models
+
+More models coming soon:
+
+1. Qwen-Image-2512
+2. Qwen-Image-Edit-2511
+3. GLM-Image

x/imagegen/memory.go

@@ -27,6 +27,7 @@ var modelVRAMEstimates = map[string]uint64{
 	"ZImagePipeline":    21 * GB, // ~21GB for Z-Image (text encoder + transformer + VAE)
 	"FluxPipeline":      21 * GB, // ~21GB for Flux (same architecture)
 	"QwenImagePipeline": 80 * GB, // TODO: verify actual requirements, using conservative estimate for now
+	"GlmImagePipeline":  80 * GB, // ~34GB weights + ~46GB working memory for 9B+7B hybrid model
 }
 
 // CheckPlatformSupport validates that image generation is supported on the current platform.

x/imagegen/mlx/mlx.go

@@ -607,6 +607,11 @@ func (a *Array) Valid() bool {
 	return a != nil && a.c.ctx != nil
 }
 
+// Kept returns true if the array is marked to survive Eval() cleanup.
+func (a *Array) Kept() bool {
+	return a != nil && a.kept
+}
+
 func int32ToCInt(s []int32) *C.int {
 	if len(s) == 0 {
 		return nil
@@ -1480,6 +1485,44 @@ func (a *Array) ItemInt32() int32 {
 	return int32(val)
 }
 
+// Bytes copies the raw bytes out of the array without type conversion.
+// Works with common dtypes (float32, int32, uint32, uint8).
+// For non-contiguous arrays, call Contiguous() first.
+// Note: Triggers cleanup of non-kept arrays.
+func (a *Array) Bytes() []byte {
+	cleanup()
+	nbytes := a.Nbytes()
+	if nbytes == 0 {
+		return nil
+	}
+
+	// Get raw pointer based on dtype
+	var ptr unsafe.Pointer
+	switch a.Dtype() {
+	case DtypeFloat32:
+		ptr = unsafe.Pointer(C.mlx_array_data_float32(a.c))
+	case DtypeInt32:
+		ptr = unsafe.Pointer(C.mlx_array_data_int32(a.c))
+	case DtypeUint32:
+		ptr = unsafe.Pointer(C.mlx_array_data_uint32(a.c))
+	case DtypeUint8:
+		ptr = unsafe.Pointer(C.mlx_array_data_uint8(a.c))
+	default:
+		// For other types (bf16, f16, etc), convert to float32
+		arr := AsType(a, DtypeFloat32)
+		arr.Eval()
+		ptr = unsafe.Pointer(C.mlx_array_data_float32(arr.c))
+		nbytes = arr.Nbytes()
+	}
+
+	if ptr == nil {
+		return nil
+	}
+	data := make([]byte, nbytes)
+	copy(data, unsafe.Slice((*byte)(ptr), nbytes))
+	return data
+}
+
 // ============ Utility ============
 
 // String returns a string representation
@@ -1658,6 +1701,34 @@ func (s *SafetensorsFile) Free() {
 	C.mlx_map_string_to_string_free(s.metadata)
 }
 
+// SaveSafetensors saves arrays to a safetensors file using MLX's native implementation.
+// This correctly handles all dtypes including uint32 for quantized weights.
+func SaveSafetensors(path string, arrays map[string]*Array) error {
+	cPath := C.CString(path)
+	defer C.free(unsafe.Pointer(cPath))
+
+	// Create the map
+	cArrays := C.mlx_map_string_to_array_new()
+	defer C.mlx_map_string_to_array_free(cArrays)
+
+	// Add each array to the map
+	for name, arr := range arrays {
+		cName := C.CString(name)
+		C.mlx_map_string_to_array_insert(cArrays, cName, arr.c)
+		C.free(unsafe.Pointer(cName))
+	}
+
+	// Create empty metadata (optional)
+	cMeta := C.mlx_map_string_to_string_new()
+	defer C.mlx_map_string_to_string_free(cMeta)
+
+	// Save
+	if C.mlx_save_safetensors(cPath, cArrays, cMeta) != 0 {
+		return fmt.Errorf("failed to save safetensors: %s", path)
+	}
+	return nil
+}
+
 // ============ NPY Loading ============
 
 // LoadNpy loads a numpy array from an npy file
@@ -1986,7 +2057,8 @@ func GatherQMM(x, w, scales *Array, biases, lhsIndices, rhsIndices *Array, trans
 // Returns (quantized_weights, scales, biases).
 // groupSize: number of elements quantized together (default 64)
 // bits: bits per element, 2, 4, or 8 (default 4)
-// mode: "affine" (default) or "mxfp4"
+// mode: "affine" (default), "mxfp4", or "mxfp8"
+// Note: mxfp8 mode returns nil biases (only weights and scales)
 func Quantize(w *Array, groupSize, bits int, mode string) (weights, scales, biases *Array) {
 	cMode := C.CString(mode)
 	defer C.free(unsafe.Pointer(cMode))
@@ -1995,14 +2067,21 @@ func Quantize(w *Array, groupSize, bits int, mode string) (weights, scales, bias
 	res := C.mlx_vector_array_new()
 	C.mlx_quantize(&res, w.c, optGroupSize, optBits, cMode, C.default_stream())
 
-	// Result is a vector of 3 arrays: [weights, scales, biases]
+	// Result is a vector of arrays: [weights, scales, biases?]
+	// mxfp8 mode returns only 2 elements (no biases)
+	vecSize := int(C.mlx_vector_array_size(res))
 	var w0, w1, w2 C.mlx_array
 	C.mlx_vector_array_get(&w0, res, 0)
 	C.mlx_vector_array_get(&w1, res, 1)
-	C.mlx_vector_array_get(&w2, res, 2)
+	if vecSize >= 3 {
+		C.mlx_vector_array_get(&w2, res, 2)
+	}
 	C.mlx_vector_array_free(res)
 
-	return newArray(w0), newArray(w1), newArray(w2)
+	if vecSize >= 3 {
+		return newArray(w0), newArray(w1), newArray(w2)
+	}
+	return newArray(w0), newArray(w1), nil
 }
 
 // Dequantize reconstructs weights from quantized form.

x/imagegen/models/glm_image/glm_image.go

@@ -0,0 +1,693 @@
+//go:build mlx
+
+// Package glm_image implements the GLM-Image hybrid AR + diffusion model.
+package glm_image
+
+import (
+	"context"
+	"fmt"
+	"math"
+	"path/filepath"
+	"time"
+
+	"github.com/ollama/ollama/x/imagegen"
+	"github.com/ollama/ollama/x/imagegen/mlx"
+)
+
+// ByT5Tokenizer is a simple byte-level tokenizer for ByT5
+// ByT5 uses bytes as tokens: each byte (0-255) maps to token ID (3-258)
+// Special tokens: 0=pad, 1=eos, 2=unk
+type ByT5Tokenizer struct {
+	PadTokenID int32
+	EOSTokenID int32
+	UNKTokenID int32
+}
+
+// NewByT5Tokenizer creates a new ByT5 tokenizer
+func NewByT5Tokenizer() *ByT5Tokenizer {
+	return &ByT5Tokenizer{
+		PadTokenID: 0,
+		EOSTokenID: 1,
+		UNKTokenID: 2,
+	}
+}
+
+// Encode converts a string to token IDs
+func (t *ByT5Tokenizer) Encode(text string) []int32 {
+	bytes := []byte(text)
+	tokens := make([]int32, len(bytes))
+	for i, b := range bytes {
+		// Standard ByT5 tokenization: bytes 0-255 map to tokens 3-258
+		// (tokens 0, 1, 2 are PAD, EOS, UNK)
+		tokens[i] = int32(b) + 3
+	}
+	return tokens
+}
+
+// Decode converts token IDs back to a string
+func (t *ByT5Tokenizer) Decode(tokens []int32) string {
+	bytes := make([]byte, 0, len(tokens))
+	for _, tok := range tokens {
+		if tok >= 3 && tok < 259 {
+			bytes = append(bytes, byte(tok-3))
+		}
+	}
+	return string(bytes)
+}
+
+// GenerateConfig holds all options for image generation.
+type GenerateConfig struct {
+	Prompt         string
+	NegativePrompt string       // For CFG (optional, not typically used with GLM-Image)
+	GuidanceScale  float32      // Guidance scale (default: 1.5)
+	Width          int32        // Image width (default: 1024, must be divisible by 32)
+	Height         int32        // Image height (default: 1024, must be divisible by 32)
+	Steps          int          // Diffusion denoising steps (default: 50)
+	Seed           int64        // Random seed
+	Progress       ProgressFunc // Optional progress callback
+
+	// AR generation options
+	MaxVisualTokens int32   // Max visual tokens to generate (default: 256)
+	Temperature     float32 // AR sampling temperature (default: 0.9)
+	TopP            float32 // Nucleus sampling (default: 0.75)
+}
+
+// ProgressFunc is called during generation with stage and step progress.
+type ProgressFunc func(stage string, step, totalSteps int)
+
+// Model represents a GLM-Image hybrid model.
+type Model struct {
+	ModelName             string
+	Tokenizer             *ByT5Tokenizer   // For T5 text encoder (glyph embeddings)
+	GLMTokenizer          *GLMTokenizer    // For AR model (visual token generation)
+	TextEncoder           *T5TextEncoder
+	VisionLanguageEncoder *VisionLanguageEncoder
+	Transformer           *DiffusionTransformer
+	VAEDecoder            *VAEDecoder
+}
+
+// Load loads the GLM-Image model from ollama blob storage.
+func (m *Model) Load(modelName string) error {
+	fmt.Printf("Loading GLM-Image model from manifest: %s...\n", modelName)
+	start := time.Now()
+
+	if mlx.GPUIsAvailable() {
+		mlx.SetDefaultDeviceGPU()
+		mlx.EnableCompile()
+	}
+
+	m.ModelName = modelName
+
+	// Load manifest
+	manifest, err := imagegen.LoadManifest(modelName)
+	if err != nil {
+		return fmt.Errorf("load manifest: %w", err)
+	}
+
+	// Create ByT5 tokenizer (byte-level, no vocabulary file needed)
+	// Used for T5 text encoder (glyph embeddings)
+	fmt.Print("  Creating ByT5 tokenizer... ")
+	m.Tokenizer = NewByT5Tokenizer()
+	fmt.Println("✓")
+
+	// Load GLM tokenizer for AR model (visual token generation)
+	fmt.Print("  Loading GLM tokenizer... ")
+	glmTok, err := NewGLMTokenizer(manifest)
+	if err != nil {
+		return fmt.Errorf("glm tokenizer: %w", err)
+	}
+	m.GLMTokenizer = glmTok
+	fmt.Println("✓")
+
+	// Load T5 text encoder (~830MB)
+	m.TextEncoder = &T5TextEncoder{}
+	if err := m.TextEncoder.Load(manifest); err != nil {
+		return fmt.Errorf("text encoder: %w", err)
+	}
+	mlx.Eval(mlx.Collect(m.TextEncoder)...)
+	fmt.Printf("  (%.1f GB, peak %.1f GB)\n",
+		float64(mlx.MetalGetActiveMemory())/(1024*1024*1024),
+		float64(mlx.MetalGetPeakMemory())/(1024*1024*1024))
+
+	// Load vision-language encoder (~19GB, 9B params)
+	m.VisionLanguageEncoder = &VisionLanguageEncoder{}
+	if err := m.VisionLanguageEncoder.Load(manifest); err != nil {
+		return fmt.Errorf("vision language encoder: %w", err)
+	}
+	mlx.Eval(mlx.Collect(m.VisionLanguageEncoder)...)
+	fmt.Printf("  (%.1f GB, peak %.1f GB)\n",
+		float64(mlx.MetalGetActiveMemory())/(1024*1024*1024),
+		float64(mlx.MetalGetPeakMemory())/(1024*1024*1024))
+
+	// Load diffusion transformer (~13GB, 7B params)
+	m.Transformer = &DiffusionTransformer{}
+	if err := m.Transformer.Load(manifest); err != nil {
+		return fmt.Errorf("transformer: %w", err)
+	}
+	mlx.Eval(mlx.Collect(m.Transformer)...)
+	fmt.Printf("  (%.1f GB, peak %.1f GB)\n",
+		float64(mlx.MetalGetActiveMemory())/(1024*1024*1024),
+		float64(mlx.MetalGetPeakMemory())/(1024*1024*1024))
+
+	// Load VAE decoder (~775MB)
+	m.VAEDecoder = &VAEDecoder{}
+	if err := m.VAEDecoder.Load(manifest); err != nil {
+		return fmt.Errorf("VAE decoder: %w", err)
+	}
+	mlx.Eval(mlx.Collect(m.VAEDecoder)...)
+	fmt.Printf("  (%.1f GB, peak %.1f GB)\n",
+		float64(mlx.MetalGetActiveMemory())/(1024*1024*1024),
+		float64(mlx.MetalGetPeakMemory())/(1024*1024*1024))
+
+	mem := mlx.MetalGetActiveMemory()
+	fmt.Printf("  Loaded in %.2fs (%.1f GB VRAM)\n", time.Since(start).Seconds(), float64(mem)/(1024*1024*1024))
+
+	return nil
+}
+
+// LoadFromPath loads the model from a directory path (not ollama manifest)
+func (m *Model) LoadFromPath(modelPath string) error {
+	fmt.Printf("Loading GLM-Image model from path: %s...\n", modelPath)
+	start := time.Now()
+
+	if mlx.GPUIsAvailable() {
+		mlx.SetDefaultDeviceGPU()
+		mlx.EnableCompile()
+	}
+
+	m.ModelName = modelPath
+
+	// Create ByT5 tokenizer (byte-level, no vocabulary file needed)
+	fmt.Print("  Creating ByT5 tokenizer... ")
+	m.Tokenizer = NewByT5Tokenizer()
+	fmt.Println("✓")
+
+	// Load GLM tokenizer for AR model (visual token generation)
+	fmt.Print("  Loading GLM tokenizer... ")
+	glmTok, err := NewGLMTokenizerFromPath(modelPath)
+	if err != nil {
+		return fmt.Errorf("glm tokenizer: %w", err)
+	}
+	m.GLMTokenizer = glmTok
+	fmt.Println("✓")
+
+	// Load T5 text encoder
+	m.TextEncoder = &T5TextEncoder{}
+	if err := m.TextEncoder.LoadFromPath(filepath.Join(modelPath, "text_encoder")); err != nil {
+		return fmt.Errorf("text encoder: %w", err)
+	}
+	mlx.Eval(mlx.Collect(m.TextEncoder)...)
+	fmt.Printf("  (%.1f GB, peak %.1f GB)\n",
+		float64(mlx.MetalGetActiveMemory())/(1024*1024*1024),
+		float64(mlx.MetalGetPeakMemory())/(1024*1024*1024))
+
+	// Load vision-language encoder
+	m.VisionLanguageEncoder = &VisionLanguageEncoder{}
+	if err := m.VisionLanguageEncoder.LoadFromPath(filepath.Join(modelPath, "vision_language_encoder")); err != nil {
+		return fmt.Errorf("vision language encoder: %w", err)
+	}
+	mlx.Eval(mlx.Collect(m.VisionLanguageEncoder)...)
+	fmt.Printf("  (%.1f GB, peak %.1f GB)\n",
+		float64(mlx.MetalGetActiveMemory())/(1024*1024*1024),
+		float64(mlx.MetalGetPeakMemory())/(1024*1024*1024))
+
+	// Load diffusion transformer
+	m.Transformer = &DiffusionTransformer{}
+	if err := m.Transformer.LoadFromPath(filepath.Join(modelPath, "transformer")); err != nil {
+		return fmt.Errorf("transformer: %w", err)
+	}
+	mlx.Eval(mlx.Collect(m.Transformer)...)
+	fmt.Printf("  (%.1f GB, peak %.1f GB)\n",
+		float64(mlx.MetalGetActiveMemory())/(1024*1024*1024),
+		float64(mlx.MetalGetPeakMemory())/(1024*1024*1024))
+
+	// Load VAE decoder
+	m.VAEDecoder = &VAEDecoder{}
+	if err := m.VAEDecoder.LoadFromPath(filepath.Join(modelPath, "vae")); err != nil {
+		return fmt.Errorf("VAE decoder: %w", err)
+	}
+	mlx.Eval(mlx.Collect(m.VAEDecoder)...)
+	fmt.Printf("  (%.1f GB, peak %.1f GB)\n",
+		float64(mlx.MetalGetActiveMemory())/(1024*1024*1024),
+		float64(mlx.MetalGetPeakMemory())/(1024*1024*1024))
+
+	mem := mlx.MetalGetActiveMemory()
+	fmt.Printf("  Loaded in %.2fs (%.1f GB VRAM)\n", time.Since(start).Seconds(), float64(mem)/(1024*1024*1024))
+
+	return nil
+}
+
+// Generate creates an image from a prompt.
+func (m *Model) Generate(prompt string, width, height int32, steps int, seed int64) (*mlx.Array, error) {
+	return m.GenerateFromConfig(context.Background(), &GenerateConfig{
+		Prompt: prompt,
+		Width:  width,
+		Height: height,
+		Steps:  steps,
+		Seed:   seed,
+	})
+}
+
+// GenerateWithProgress creates an image with progress callback.
+func (m *Model) GenerateWithProgress(prompt string, width, height int32, steps int, seed int64, progress ProgressFunc) (*mlx.Array, error) {
+	return m.GenerateFromConfig(context.Background(), &GenerateConfig{
+		Prompt:   prompt,
+		Width:    width,
+		Height:   height,
+		Steps:    steps,
+		Seed:     seed,
+		Progress: progress,
+	})
+}
+
+// GenerateFromConfig generates an image using the unified config struct.
+func (m *Model) GenerateFromConfig(ctx context.Context, cfg *GenerateConfig) (*mlx.Array, error) {
+	start := time.Now()
+	result, err := m.generate(ctx, cfg)
+	if err != nil {
+		return nil, err
+	}
+	fmt.Printf("Generated in %.2fs (%d diffusion steps)\n", time.Since(start).Seconds(), cfg.Steps)
+	return result, nil
+}
+
+// GenerateImage implements model.ImageModel interface.
+func (m *Model) GenerateImage(ctx context.Context, prompt string, width, height int32, steps int, seed int64) (*mlx.Array, error) {
+	return m.Generate(prompt, width, height, steps, seed)
+}
+
+// generate is the internal generation pipeline.
+func (m *Model) generate(ctx context.Context, cfg *GenerateConfig) (*mlx.Array, error) {
+	// Apply defaults
+	if cfg.Width <= 0 {
+		cfg.Width = 1024
+	}
+	if cfg.Height <= 0 {
+		cfg.Height = 1024
+	}
+	if cfg.Steps <= 0 {
+		cfg.Steps = 50
+	}
+	if cfg.GuidanceScale <= 0 {
+		cfg.GuidanceScale = 1.5
+	}
+	// Calculate MaxVisualTokens based on image dimensions
+	// GLM-Image generates TWO grids of visual tokens:
+	//   1. First: prev (small) grid - prevTokenH × prevTokenW tokens
+	//   2. Then: target (large) grid - tokenH × tokenW tokens
+	// After generation, we extract only the TARGET grid tokens for diffusion.
+	factor := int32(32)
+	tokenH := cfg.Height / factor
+	tokenW := cfg.Width / factor
+	targetGridTokens := tokenH * tokenW
+
+	// Compute prev grid dimensions using diffusers formula:
+	// ratio = token_h / token_w
+	// prev_token_h = int(sqrt(ratio) * 16)
+	// prev_token_w = int(sqrt(1/ratio) * 16)
+	ratio := float64(tokenH) / float64(tokenW)
+	prevTokenH := int32(math.Sqrt(ratio) * 16)
+	prevTokenW := int32(math.Sqrt(1/ratio) * 16)
+	prevGridTokens := prevTokenH * prevTokenW
+
+	// Total tokens to generate = prev grid + target grid
+	// (diffusers does max_new_tokens = total + 1 for EOS, but we stop on EOS anyway)
+	cfg.MaxVisualTokens = prevGridTokens + targetGridTokens
+	if cfg.Temperature <= 0 {
+		cfg.Temperature = 0.9
+	}
+	if cfg.TopP <= 0 {
+		cfg.TopP = 0.75
+	}
+
+	// Ensure dimensions are divisible by 32
+	cfg.Width = (cfg.Width / 32) * 32
+	cfg.Height = (cfg.Height / 32) * 32
+
+	tcfg := m.Transformer.Config
+	latentH := cfg.Height / 8
+	latentW := cfg.Width / 8
+
+	// Progress callback helper
+	progress := func(stage string, step, total int) {
+		if cfg.Progress != nil {
+			cfg.Progress(stage, step, total)
+		}
+	}
+
+	// === PHASE 1: T5 Text Encoding ===
+	fmt.Println("[T5] Encoding glyph text...")
+	progress("text_encoding", 0, 1)
+	textEmbed := m.TextEncoder.EncodePrompt(m.Tokenizer, cfg.Prompt)
+	mlx.Keep(textEmbed)
+	mlx.Eval(textEmbed)
+	fmt.Printf("[T5] Done, shape: %v\n", textEmbed.Shape())
+	progress("text_encoding", 1, 1)
+
+	// === PHASE 2: AR Visual Token Generation ===
+	fmt.Printf("[AR] Generating %d visual tokens...\n", cfg.MaxVisualTokens)
+	progress("ar_generation", 0, int(cfg.MaxVisualTokens))
+	visualTokens := m.VisionLanguageEncoder.Generate(
+		cfg.Prompt,
+		m.GLMTokenizer,
+		cfg.MaxVisualTokens,
+		cfg.Temperature,
+		cfg.TopP,
+		cfg.Seed,
+		cfg.Height,
+		cfg.Width,
+		func(step int) {
+			if step%100 == 0 || step < 10 {
+				fmt.Printf("[AR] Step %d/%d\n", step, cfg.MaxVisualTokens)
+			}
+			progress("ar_generation", step, int(cfg.MaxVisualTokens))
+		},
+	)
+	mlx.Keep(visualTokens)
+	mlx.Eval(visualTokens)
+	fmt.Printf("[AR] Done generating visual tokens\n")
+	progress("ar_generation", int(cfg.MaxVisualTokens), int(cfg.MaxVisualTokens))
+
+	vtShape := visualTokens.Shape()
+	totalGenerated := vtShape[1]
+	fmt.Printf("[AR] Generated %d tokens total\n", totalGenerated)
+
+	// Extract only the TARGET grid tokens (skip the prev grid tokens)
+	// diffusers: large_image_tokens = outputs[input_length + large_image_start_offset : ...]
+	// large_image_start_offset = prev_grid_size
+	var targetGridVisualTokens *mlx.Array
+	if totalGenerated >= prevGridTokens+targetGridTokens {
+		// Full generation completed - extract target grid
+		targetGridVisualTokens = mlx.Slice(visualTokens,
+			[]int32{0, prevGridTokens},
+			[]int32{1, prevGridTokens + targetGridTokens})
+		mlx.Keep(targetGridVisualTokens)
+		mlx.Eval(targetGridVisualTokens)
+	} else if totalGenerated > prevGridTokens {
+		// Partial target grid - take what we have
+		actualTargetTokens := totalGenerated - prevGridTokens
+		targetGridVisualTokens = mlx.Slice(visualTokens,
+			[]int32{0, prevGridTokens},
+			[]int32{1, totalGenerated})
+		mlx.Keep(targetGridVisualTokens)
+		mlx.Eval(targetGridVisualTokens)
+		fmt.Printf("WARNING: Partial target grid: got %d/%d target tokens\n",
+			actualTargetTokens, targetGridTokens)
+	} else {
+		// Not enough tokens - EOS came too early
+		return nil, fmt.Errorf("AR generation stopped too early: got %d tokens, need at least %d (prev grid) + 1",
+			totalGenerated, prevGridTokens)
+	}
+
+	// === PHASE 3: Diffusion Decoding ===
+	// Setup scheduler with dynamic shift based on image size
+	scheduler := NewFlowMatchScheduler(DefaultSchedulerConfig())
+	imgSeqLen := (latentH / tcfg.PatchSize) * (latentW / tcfg.PatchSize)
+	scheduler.SetTimestepsWithDynamicShift(cfg.Steps, imgSeqLen)
+
+	// Initialize noise latents [B, C, H, W]
+	latents := scheduler.InitNoise([]int32{1, tcfg.InChannels, latentH, latentW}, cfg.Seed)
+	mlx.Eval(latents)
+
+	// Upsample TARGET grid visual tokens 2x to match patch count (matching diffusers)
+	// target_grid tokens -> 2x upsample -> patch_count
+	// e.g., 32x32=1024 tokens -> 64x64=4096 patches for 1024x1024
+	visualTokensUpsampled := upsampleTokens(targetGridVisualTokens, tokenH, tokenW, 2)
+
+	// Prepare prior embeddings from upsampled visual tokens (VQ codebook lookup + projection)
+	priorEmbed := m.Transformer.EmbedPriorTokens(visualTokensUpsampled)
+	mlx.Keep(priorEmbed)
+	mlx.Eval(priorEmbed)
+
+	// Prepare text conditioning (project T5 embeddings)
+	textCond := m.Transformer.ProjectTextEmbeddings(textEmbed)
+	mlx.Keep(textCond)
+	mlx.Eval(textCond)
+
+	// === CFG Setup ===
+	// For classifier-free guidance, we need unconditional (negative) text embeddings
+	// GLM-Image uses empty string "" for negative prompt
+	doCFG := cfg.GuidanceScale > 1.0
+	var negativeTextCond *mlx.Array
+	if doCFG {
+		// Encode empty string for negative prompt
+		negativeTextEmbed := m.TextEncoder.EncodePrompt(m.Tokenizer, "")
+		mlx.Keep(negativeTextEmbed)
+		mlx.Eval(negativeTextEmbed)
+		negativeTextCond = m.Transformer.ProjectTextEmbeddings(negativeTextEmbed)
+		mlx.Keep(negativeTextCond)
+		mlx.Eval(negativeTextCond)
+		negativeTextEmbed.Free()
+	}
+
+	// Prepare conditioning inputs
+	targetSize := mlx.NewArray([]float32{float32(cfg.Height), float32(cfg.Width)}, []int32{1, 2})
+	cropCoords := mlx.NewArray([]float32{0, 0}, []int32{1, 2}) // Default: no crop offset
+	targetSize = mlx.ToBFloat16(targetSize)
+	cropCoords = mlx.ToBFloat16(cropCoords)
+	mlx.Keep(targetSize)
+	mlx.Keep(cropCoords)
+	mlx.Eval(targetSize, cropCoords)
+
+	pH := latentH / tcfg.PatchSize
+	pW := latentW / tcfg.PatchSize
+
+	// Denoising loop
+	fmt.Printf("[Diffusion] Starting %d denoising steps...\n", cfg.Steps)
+	progress("diffusion", 0, cfg.Steps)
+	for i := 0; i < cfg.Steps; i++ {
+		fmt.Printf("[Diffusion] Step %d/%d (timestep=%.1f)\n", i+1, cfg.Steps, scheduler.Timesteps[i]-1)
+		// Check for cancellation
+		if ctx != nil {
+			select {
+			case <-ctx.Done():
+				textEmbed.Free()
+				visualTokens.Free()
+				// visualTokensUpsampled points to visualTokens, don't double-free
+				priorEmbed.Free()
+				textCond.Free()
+				latents.Free()
+				return nil, ctx.Err()
+			default:
+			}
+		}
+
+		// Get timestep value for the transformer
+		// scheduler.Timesteps contains raw timestep values (1000 down to ~20)
+		// Pass timestep - 1 to match diffusers: timestep = t.expand(latents.shape[0]) - 1
+		timestepVal := scheduler.Timesteps[i] - 1
+		timestep := mlx.ToBFloat16(mlx.NewArray([]float32{timestepVal}, []int32{1}))
+
+		// Patchify latents [B, C, H, W] -> [B, L, C*p*p]
+		patches := PatchifyLatents(latents, tcfg.PatchSize)
+
+		// Transformer forward with MMDiT architecture
+		// Conditional pass (with text + prior embeddings)
+		outputCond := m.Transformer.ForwardWithPriorDrop(
+			patches,
+			priorEmbed,
+			textCond,
+			timestep,
+			targetSize,
+			cropCoords,
+			pH,
+			pW,
+			false, // priorTokenDrop = false for conditional
+		)
+
+		// Unpatchify [B, L, C*p*p] -> [B, C, H, W]
+		noisePredCond := UnpatchifyLatents(outputCond, latentH, latentW, tcfg.PatchSize, tcfg.OutChannels)
+
+		var noisePred *mlx.Array
+		if doCFG {
+			// Unconditional pass (empty text, dropped prior embeddings)
+			outputUncond := m.Transformer.ForwardWithPriorDrop(
+				patches,
+				priorEmbed, // Still passed but will be ignored due to priorTokenDrop=true
+				negativeTextCond,
+				timestep,
+				targetSize,
+				cropCoords,
+				pH,
+				pW,
+				true, // priorTokenDrop = true for unconditional
+			)
+			noisePredUncond := UnpatchifyLatents(outputUncond, latentH, latentW, tcfg.PatchSize, tcfg.OutChannels)
+
+			// CFG formula: noise_pred = uncond + guidance_scale * (cond - uncond)
+			diff := mlx.Sub(noisePredCond, noisePredUncond)
+			scaled := mlx.MulScalar(diff, cfg.GuidanceScale)
+			noisePred = mlx.Add(noisePredUncond, scaled)
+		} else {
+			noisePred = noisePredCond
+		}
+
+		// Scheduler step
+		oldLatents := latents
+		latents = scheduler.Step(noisePred, latents, i)
+		mlx.Eval(latents)
+		oldLatents.Free()
+
+		progress("diffusion", i+1, cfg.Steps)
+	}
+
+	// Cleanup intermediate arrays
+	textEmbed.Free()
+	visualTokens.Free()
+	// visualTokensUpsampled points to visualTokens, don't double-free
+	priorEmbed.Free()
+	textCond.Free()
+	if negativeTextCond != nil {
+		negativeTextCond.Free()
+	}
+	targetSize.Free()
+	cropCoords.Free()
+
+	// === PHASE 4: VAE Decode ===
+	progress("vae_decode", 0, 1)
+	decoded := m.VAEDecoder.Decode(latents)
+	mlx.Eval(decoded)
+	latents.Free()
+	progress("vae_decode", 1, 1)
+
+	return decoded, nil
+}
+
+// upsampleTokens performs nearest-neighbor upsampling of visual tokens
+// Converts from prev_grid (e.g., 16x16) to target_grid (e.g., 32x32 for 2x, 64x64 for 4x)
+// scale must be 2 or 4
+//
+// Handles early EOS gracefully: if tokens has fewer than prevH*prevW elements,
+// missing tokens are padded with 0 (visual token padding value).
+func upsampleTokens(tokens *mlx.Array, prevH, prevW int32, scale int32) *mlx.Array {
+	// tokens: [1, N] where N <= prevH*prevW (may be shorter if early EOS)
+	// Each token at (i, j) becomes scale*scale tokens in the output
+
+	mlx.Eval(tokens)
+	tokenData := tokens.DataInt32()
+	numTokens := int32(len(tokenData))
+	expectedTokens := prevH * prevW
+
+	// Warn if we got fewer tokens than expected (early EOS)
+	if numTokens < expectedTokens {
+		fmt.Printf("WARNING: upsampleTokens got %d tokens, expected %d (padding with 0)\n",
+			numTokens, expectedTokens)
+	}
+
+	targetH := prevH * scale
+	targetW := prevW * scale
+	upsampled := make([]int32, targetH*targetW)
+
+	for i := int32(0); i < prevH; i++ {
+		for j := int32(0); j < prevW; j++ {
+			srcIdx := i*prevW + j
+
+			// Handle early EOS: use 0 (padding) for missing tokens
+			var val int32
+			if srcIdx < numTokens {
+				val = tokenData[srcIdx]
+			} else {
+				val = 0 // Padding token
+			}
+
+			// Place in scale*scale positions
+			dstI := i * scale
+			dstJ := j * scale
+			for di := int32(0); di < scale; di++ {
+				for dj := int32(0); dj < scale; dj++ {
+					upsampled[(dstI+di)*targetW+(dstJ+dj)] = val
+				}
+			}
+		}
+	}
+
+	return mlx.NewArrayInt32(upsampled, []int32{1, targetH * targetW})
+}
+
+// PatchifyLatents converts [B, C, H, W] to [B, L, C*p*p]
+func PatchifyLatents(latents *mlx.Array, patchSize int32) *mlx.Array {
+	shape := latents.Shape()
+	B := shape[0]
+	C := shape[1]
+	H := shape[2]
+	W := shape[3]
+
+	pH := H / patchSize
+	pW := W / patchSize
+
+	// Reshape: [B, C, H, W] -> [B, C, pH, p, pW, p]
+	x := mlx.Reshape(latents, B, C, pH, patchSize, pW, patchSize)
+	// Transpose: -> [B, pH, pW, C, p, p]
+	x = mlx.Transpose(x, 0, 2, 4, 1, 3, 5)
+	// Flatten: -> [B, pH*pW, C*p*p]
+	return mlx.Reshape(x, B, pH*pW, C*patchSize*patchSize)
+}
+
+// UnpatchifyLatents converts [B, L, C*p*p] back to [B, C, H, W]
+func UnpatchifyLatents(patches *mlx.Array, H, W, patchSize, channels int32) *mlx.Array {
+	shape := patches.Shape()
+	B := shape[0]
+
+	pH := H / patchSize
+	pW := W / patchSize
+
+	// Reshape: [B, L, C*p*p] -> [B, pH, pW, C, p, p]
+	x := mlx.Reshape(patches, B, pH, pW, channels, patchSize, patchSize)
+	// Transpose: -> [B, C, pH, p, pW, p]
+	x = mlx.Transpose(x, 0, 3, 1, 4, 2, 5)
+	// Reshape: -> [B, C, H, W]
+	return mlx.Reshape(x, B, channels, pH*patchSize, pW*patchSize)
+}
+
+// CalculateShift computes the dynamic shift for flow matching based on image sequence length.
+func CalculateShift(imgSeqLen int32) float32 {
+	cfg := DefaultSchedulerConfig()
+	if !cfg.UseDynamicShifting {
+		return 0
+	}
+
+	// Sqrt-based shift calculation (matches diffusers)
+	m := float32(math.Sqrt(float64(imgSeqLen) / float64(cfg.BaseImageSeqLen)))
+	return m*cfg.MaxShift + cfg.BaseShift
+}
+
+// UpsampleTokens2x upsamples token IDs by 2x using nearest neighbor interpolation
+// tokens: [B, H*W] -> [B, (H*2)*(W*2)]
+// This matches diffusers' _upsample_token_ids function
+func UpsampleTokens2x(tokens *mlx.Array, gridH, gridW int32) *mlx.Array {
+	shape := tokens.Shape()
+	B := shape[0]
+
+	// Reshape to [B, 1, H, W] for interpolation
+	tokens = mlx.Reshape(tokens, B, 1, gridH, gridW)
+
+	// Convert to float for interpolation
+	tokensFloat := mlx.AsType(tokens, mlx.DtypeFloat32)
+
+	// 2x nearest neighbor upsample
+	// [B, 1, H, W] -> [B, 1, H*2, W*2]
+	upsampled := nearestUpsample2x(tokensFloat)
+
+	// Convert back to int and reshape to [B, H*2*W*2]
+	upsampled = mlx.AsType(upsampled, mlx.DtypeInt32)
+	return mlx.Reshape(upsampled, B, gridH*2*gridW*2)
+}
+
+// nearestUpsample2x performs 2x nearest neighbor upsampling on NCHW tensor
+func nearestUpsample2x(x *mlx.Array) *mlx.Array {
+	shape := x.Shape()
+	B := shape[0]
+	C := shape[1]
+	H := shape[2]
+	W := shape[3]
+
+	// Repeat each element 2x2
+	// [B, C, H, W] -> [B, C, H, 1, W, 1] -> [B, C, H, 2, W, 2] -> [B, C, H*2, W*2]
+	x = mlx.Reshape(x, B, C, H, 1, W, 1)
+
+	// Tile to repeat each pixel 2x2
+	x = mlx.Tile(x, []int32{1, 1, 1, 2, 1, 2})
+
+	// Reshape to final size
+	return mlx.Reshape(x, B, C, H*2, W*2)
+}

x/imagegen/models/glm_image/glm_tokenizer.go

@@ -0,0 +1,358 @@
+//go:build mlx
+
+package glm_image
+
+import (
+	"encoding/json"
+	"fmt"
+	"os"
+	"path/filepath"
+	"sort"
+	"strings"
+
+	"github.com/ollama/ollama/x/imagegen"
+)
+
+// GLMTokenizer implements the GLM tokenizer for the AR model
+// This is a BPE-style tokenizer with ignore_merges=true, meaning it does
+// greedy longest-match tokenization from the vocab without runtime merging.
+type GLMTokenizer struct {
+	Vocab        map[string]int32 // token string -> token ID
+	VocabReverse map[int32]string // token ID -> token string
+	SpecialTokens map[string]int32 // special token strings -> IDs
+
+	// Special token IDs
+	SopTokenID  int32 // <sop> = grid_bos_token (167845)
+	EopTokenID  int32 // <eop> = grid_eos_token (167846)
+	BosTokenID  int32 // <|dit_token_16384|> = visual BOS (16384)
+	EosTokenID  int32 // <|dit_token_16385|> = visual EOS (16385)
+	PadTokenID  int32
+
+	// Sorted vocab keys by length (longest first) for greedy matching
+	sortedTokens []string
+}
+
+// tokenizerJSON represents the structure of tokenizer.json
+type tokenizerJSON struct {
+	Model struct {
+		Vocab map[string]int32 `json:"vocab"`
+	} `json:"model"`
+	AddedTokens []struct {
+		ID      int32  `json:"id"`
+		Content string `json:"content"`
+		Special bool   `json:"special"`
+	} `json:"added_tokens"`
+}
+
+// NewGLMTokenizer creates a GLM tokenizer from the model manifest
+func NewGLMTokenizer(manifest *imagegen.ModelManifest) (*GLMTokenizer, error) {
+	// Read tokenizer.json from processor directory in manifest
+	data, err := manifest.ReadConfig("processor/tokenizer.json")
+	if err != nil {
+		return nil, fmt.Errorf("failed to read tokenizer.json from manifest: %w", err)
+	}
+
+	var tj tokenizerJSON
+	if err := json.Unmarshal(data, &tj); err != nil {
+		return nil, fmt.Errorf("failed to parse tokenizer.json: %w", err)
+	}
+
+	tok := &GLMTokenizer{
+		Vocab:         make(map[string]int32),
+		VocabReverse:  make(map[int32]string),
+		SpecialTokens: make(map[string]int32),
+	}
+
+	// Load vocab from model section
+	for token, id := range tj.Model.Vocab {
+		tok.Vocab[token] = id
+		tok.VocabReverse[id] = token
+	}
+
+	// Load added tokens (special tokens including dit_tokens)
+	for _, at := range tj.AddedTokens {
+		tok.Vocab[at.Content] = at.ID
+		tok.VocabReverse[at.ID] = at.Content
+		if at.Special {
+			tok.SpecialTokens[at.Content] = at.ID
+		}
+	}
+
+	// Set special token IDs
+	tok.SopTokenID = 167845   // <sop>
+	tok.EopTokenID = 167846   // <eop>
+	tok.BosTokenID = 16384    // <|dit_token_16384|>
+	tok.EosTokenID = 16385    // <|dit_token_16385|>
+	tok.PadTokenID = 16385    // Same as EOS
+
+	// Build sorted token list for greedy matching (longest first)
+	tok.sortedTokens = make([]string, 0, len(tok.Vocab))
+	for token := range tok.Vocab {
+		tok.sortedTokens = append(tok.sortedTokens, token)
+	}
+	sort.Slice(tok.sortedTokens, func(i, j int) bool {
+		return len(tok.sortedTokens[i]) > len(tok.sortedTokens[j])
+	})
+
+	fmt.Printf("Loaded GLM tokenizer with %d tokens\n", len(tok.Vocab))
+
+	return tok, nil
+}
+
+// NewGLMTokenizerFromPath creates a GLM tokenizer from a directory path
+func NewGLMTokenizerFromPath(modelPath string) (*GLMTokenizer, error) {
+	// Read tokenizer.json from processor directory
+	tokenizerPath := filepath.Join(modelPath, "processor", "tokenizer.json")
+	data, err := os.ReadFile(tokenizerPath)
+	if err != nil {
+		return nil, fmt.Errorf("failed to read tokenizer.json: %w", err)
+	}
+
+	var tj tokenizerJSON
+	if err := json.Unmarshal(data, &tj); err != nil {
+		return nil, fmt.Errorf("failed to parse tokenizer.json: %w", err)
+	}
+
+	tok := &GLMTokenizer{
+		Vocab:         make(map[string]int32),
+		VocabReverse:  make(map[int32]string),
+		SpecialTokens: make(map[string]int32),
+	}
+
+	// Load vocab from model section
+	for token, id := range tj.Model.Vocab {
+		tok.Vocab[token] = id
+		tok.VocabReverse[id] = token
+	}
+
+	// Load added tokens (special tokens including dit_tokens)
+	for _, at := range tj.AddedTokens {
+		tok.Vocab[at.Content] = at.ID
+		tok.VocabReverse[at.ID] = at.Content
+		if at.Special {
+			tok.SpecialTokens[at.Content] = at.ID
+		}
+	}
+
+	// Set special token IDs
+	tok.SopTokenID = 167845 // <sop>
+	tok.EopTokenID = 167846 // <eop>
+	tok.BosTokenID = 16384  // <|dit_token_16384|>
+	tok.EosTokenID = 16385  // <|dit_token_16385|>
+	tok.PadTokenID = 16385  // Same as EOS
+
+	// Build sorted token list for greedy matching (longest first)
+	tok.sortedTokens = make([]string, 0, len(tok.Vocab))
+	for token := range tok.Vocab {
+		tok.sortedTokens = append(tok.sortedTokens, token)
+	}
+	sort.Slice(tok.sortedTokens, func(i, j int) bool {
+		return len(tok.sortedTokens[i]) > len(tok.sortedTokens[j])
+	})
+
+	fmt.Printf("Loaded GLM tokenizer with %d tokens\n", len(tok.Vocab))
+
+	return tok, nil
+}
+
+// Encode tokenizes a string into token IDs
+// This uses greedy longest-match tokenization with GPT-2 style space handling
+func (t *GLMTokenizer) Encode(text string) []int32 {
+	if text == "" {
+		return []int32{}
+	}
+
+	var tokens []int32
+
+	// First, check for and handle special tokens
+	// Replace special tokens with placeholders, encode, then restore
+	specialReplacements := make(map[string]int32)
+	for special, id := range t.SpecialTokens {
+		if strings.Contains(text, special) {
+			specialReplacements[special] = id
+		}
+	}
+
+	// Process text character by character with special token handling
+	i := 0
+	isFirstToken := true
+
+	for i < len(text) {
+		// Check for special tokens first
+		foundSpecial := false
+		for special, id := range specialReplacements {
+			if strings.HasPrefix(text[i:], special) {
+				tokens = append(tokens, id)
+				i += len(special)
+				isFirstToken = false
+				foundSpecial = true
+				break
+			}
+		}
+		if foundSpecial {
+			continue
+		}
+
+		// Handle regular text with GPT-2 style space prefix
+		// "Ġ" (U+0120) represents a space before a token
+		remaining := text[i:]
+
+		// Try to find the longest matching token
+		matched := false
+		for _, token := range t.sortedTokens {
+			// Skip special tokens in regular matching
+			if _, isSpecial := t.SpecialTokens[token]; isSpecial {
+				continue
+			}
+
+			// Check if this token matches
+			tokenText := token
+
+			// Handle the Ġ prefix (represents space)
+			if strings.HasPrefix(token, "Ġ") {
+				// This token expects a leading space
+				if i > 0 || !isFirstToken {
+					// Check if remaining starts with space + token content
+					tokenContent := token[len("Ġ"):]
+					if strings.HasPrefix(remaining, " "+tokenContent) {
+						tokens = append(tokens, t.Vocab[token])
+						i += 1 + len(tokenContent) // space + content
+						isFirstToken = false
+						matched = true
+						break
+					}
+				}
+			} else {
+				// Regular token without space prefix
+				if strings.HasPrefix(remaining, tokenText) {
+					tokens = append(tokens, t.Vocab[token])
+					i += len(tokenText)
+					isFirstToken = false
+					matched = true
+					break
+				}
+			}
+		}
+
+		if !matched {
+			// No token found - skip this character (or use UNK)
+			// For now, just skip unknown characters
+			i++
+		}
+	}
+
+	return tokens
+}
+
+// EncodeForGeneration encodes a prompt with grid tokens for image generation
+// Format: {prompt}<sop>{token_h} {token_w}<eop><sop>{prev_h} {prev_w}<eop><|dit_token_16384|>
+//
+// Uses GPT-2 style tokenization where " 32" becomes "Ġ32" (a single token with
+// space prefix), matching the HuggingFace tokenizer behavior.
+func (t *GLMTokenizer) EncodeForGeneration(prompt string, targetHeight, targetWidth int32) []int32 {
+	// Calculate grid dimensions
+	factor := int32(32)
+	height := (targetHeight / factor) * factor
+	width := (targetWidth / factor) * factor
+	tokenH := height / factor
+	tokenW := width / factor
+
+	// Calculate previous grid dimensions
+	ratio := float64(tokenH) / float64(tokenW)
+	prevTokenH := int32(sqrt(ratio) * 16)
+	prevTokenW := int32(sqrt(1.0/ratio) * 16)
+
+	// Encode the prompt text
+	promptTokens := t.Encode(prompt)
+
+	// Build the full sequence:
+	// [prompt tokens] <sop> [tokenH] [Ġ+tokenW] <eop> <sop> [prevH] [Ġ+prevW] <eop> <bos>
+	// Note: HF tokenizer treats " 32" as "Ġ32" (single token), not "Ġ" + "32"
+	var tokens []int32
+	tokens = append(tokens, promptTokens...)
+
+	// First grid: <sop> H W <eop>
+	// First number has no space prefix, second number has space prefix (Ġ)
+	tokens = append(tokens, t.SopTokenID)
+	tokens = append(tokens, t.encodeNumber(tokenH)...)
+	tokens = append(tokens, t.encodeSpaceNumber(tokenW)...) // " W" as Ġ+W
+	tokens = append(tokens, t.EopTokenID)
+
+	// Second grid: <sop> prevH prevW <eop>
+	tokens = append(tokens, t.SopTokenID)
+	tokens = append(tokens, t.encodeNumber(prevTokenH)...)
+	tokens = append(tokens, t.encodeSpaceNumber(prevTokenW)...) // " prevW" as Ġ+prevW
+	tokens = append(tokens, t.EopTokenID)
+
+	// BOS token (start of image generation)
+	tokens = append(tokens, t.BosTokenID)
+
+	return tokens
+}
+
+// encodeNumber encodes a number - first tries as a whole token, falls back to digit-by-digit
+func (t *GLMTokenizer) encodeNumber(n int32) []int32 {
+	s := fmt.Sprintf("%d", n)
+	// First try: look up the whole number as a single token
+	if id, ok := t.Vocab[s]; ok {
+		return []int32{id}
+	}
+	// Fallback: encode digit by digit
+	var tokens []int32
+	for _, c := range s {
+		if id, ok := t.Vocab[string(c)]; ok {
+			tokens = append(tokens, id)
+		}
+	}
+	return tokens
+}
+
+// encodeSpaceNumber encodes " N" as "ĠN" (space-prefixed number) matching HF tokenizer
+// GPT-2 style: " 32" becomes single token "Ġ32", not "Ġ" + "32"
+func (t *GLMTokenizer) encodeSpaceNumber(n int32) []int32 {
+	s := fmt.Sprintf("%d", n)
+
+	// First try: look up "Ġ{number}" as a single token (e.g., "Ġ32")
+	spaceToken := "Ġ" + s
+	if id, ok := t.Vocab[spaceToken]; ok {
+		return []int32{id}
+	}
+
+	// Fallback: bare space Ġ + number tokens
+	var tokens []int32
+	if spaceID, ok := t.Vocab["Ġ"]; ok {
+		tokens = append(tokens, spaceID)
+	}
+	tokens = append(tokens, t.encodeNumber(n)...)
+	return tokens
+}
+
+// sqrt is a helper for float64 sqrt
+func sqrt(x float64) float64 {
+	if x <= 0 {
+		return 0
+	}
+	// Newton's method
+	z := x
+	for i := 0; i < 10; i++ {
+		z = z - (z*z-x)/(2*z)
+	}
+	return z
+}
+
+// Decode converts token IDs back to a string
+func (t *GLMTokenizer) Decode(tokens []int32) string {
+	var sb strings.Builder
+	for _, id := range tokens {
+		if token, ok := t.VocabReverse[id]; ok {
+			// Handle Ġ prefix (convert back to space)
+			if strings.HasPrefix(token, "Ġ") {
+				sb.WriteString(" ")
+				sb.WriteString(token[len("Ġ"):])
+			} else {
+				sb.WriteString(token)
+			}
+		}
+	}
+	return sb.String()
+}

x/imagegen/models/glm_image/scheduler.go

@@ -0,0 +1,159 @@
+//go:build mlx
+
+package glm_image
+
+import (
+	"math"
+
+	"github.com/ollama/ollama/x/imagegen/mlx"
+)
+
+// FlowMatchSchedulerConfig holds scheduler configuration
+type FlowMatchSchedulerConfig struct {
+	NumTrainTimesteps  int32   `json:"num_train_timesteps"`   // 1000
+	BaseShift          float32 `json:"base_shift"`            // 0.25
+	MaxShift           float32 `json:"max_shift"`             // 0.75
+	BaseImageSeqLen    int32   `json:"base_image_seq_len"`    // 256
+	MaxImageSeqLen     int32   `json:"max_image_seq_len"`     // 4096
+	UseDynamicShifting bool    `json:"use_dynamic_shifting"`  // true
+	TimeShiftType      string  `json:"time_shift_type"`       // "linear"
+}
+
+// DefaultSchedulerConfig returns the default config for GLM-Image
+func DefaultSchedulerConfig() *FlowMatchSchedulerConfig {
+	return &FlowMatchSchedulerConfig{
+		NumTrainTimesteps:  1000,
+		BaseShift:          0.25,
+		MaxShift:           0.75,
+		BaseImageSeqLen:    256,
+		MaxImageSeqLen:     4096,
+		UseDynamicShifting: true,
+		TimeShiftType:      "linear",
+	}
+}
+
+// FlowMatchScheduler implements FlowMatchEulerDiscreteScheduler
+type FlowMatchScheduler struct {
+	Config    *FlowMatchSchedulerConfig
+	Timesteps []float32 // Raw timesteps for transformer conditioning (unshifted)
+	Sigmas    []float32 // Shifted sigmas for Euler step calculation
+	NumSteps  int
+}
+
+// NewFlowMatchScheduler creates a new scheduler
+func NewFlowMatchScheduler(cfg *FlowMatchSchedulerConfig) *FlowMatchScheduler {
+	return &FlowMatchScheduler{Config: cfg}
+}
+
+// SetTimestepsWithDynamicShift sets timesteps with dynamic shifting based on image size
+// Following diffusers: raw timesteps are used for conditioning, shifted sigmas for step calculation
+func (s *FlowMatchScheduler) SetTimestepsWithDynamicShift(numSteps int, imgSeqLen int32) {
+	s.NumSteps = numSteps
+
+	// Calculate shift (mu) based on image sequence length
+	mu := s.calculateShift(imgSeqLen)
+
+	// Create timesteps: linspace from sigma_max_t to sigma_min_t
+	// sigma_max = 1.0, sigma_min ~= 0.001 (near 0 but not exactly 0)
+	// Then apply time shift and append terminal sigma=0
+	s.Timesteps = make([]float32, numSteps)
+	s.Sigmas = make([]float32, numSteps+1) // +1 for terminal sigma
+
+	numTrainTimesteps := float32(s.Config.NumTrainTimesteps)
+
+	// Create base sigmas: linspace from 1.0 to small value (matching diffusers)
+	for i := 0; i < numSteps; i++ {
+		// linspace from 1000 to ~20 (sigma_min * num_train_timesteps)
+		tRaw := numTrainTimesteps - float32(i)*(numTrainTimesteps-1.0)/float32(numSteps-1)
+		s.Timesteps[i] = tRaw
+
+		// Convert to sigma [0, 1]
+		sigma := tRaw / numTrainTimesteps
+
+		// Apply time shift if enabled
+		if s.Config.UseDynamicShifting && mu > 0 {
+			sigma = s.applyShift(mu, sigma)
+		}
+
+		s.Sigmas[i] = sigma
+	}
+
+	// Append terminal sigma = 0 (the final clean image)
+	s.Sigmas[numSteps] = 0
+}
+
+// calculateShift computes dynamic shift based on image sequence length
+// Uses the sqrt-based formula from diffusers:
+// m = (image_seq_len / base_seq_len) ** 0.5
+// mu = m * max_shift + base_shift
+func (s *FlowMatchScheduler) calculateShift(imgSeqLen int32) float32 {
+	cfg := s.Config
+
+	if !cfg.UseDynamicShifting {
+		return 0
+	}
+
+	// Sqrt-based shift calculation (matches diffusers pipeline_glm_image.py)
+	m := float32(math.Sqrt(float64(imgSeqLen) / float64(cfg.BaseImageSeqLen)))
+	mu := m*cfg.MaxShift + cfg.BaseShift
+	return mu
+}
+
+// applyShift applies time shift transformation
+// mu: the computed shift value
+// t: sigma value in [0, 1]
+func (s *FlowMatchScheduler) applyShift(mu float32, t float32) float32 {
+	if t <= 0 {
+		return 0
+	}
+	if t >= 1 {
+		return 1
+	}
+
+	// sigma=1.0 for both shift types
+	sigma := float32(1.0)
+
+	if s.Config.TimeShiftType == "linear" {
+		// Linear: mu / (mu + (1/t - 1)^sigma)
+		return mu / (mu + float32(math.Pow(float64(1.0/t-1.0), float64(sigma))))
+	}
+
+	// Exponential (default): exp(mu) / (exp(mu) + (1/t - 1)^sigma)
+	expMu := float32(math.Exp(float64(mu)))
+	return expMu / (expMu + float32(math.Pow(float64(1.0/t-1.0), float64(sigma))))
+}
+
+// Step performs one denoising step
+func (s *FlowMatchScheduler) Step(modelOutput, sample *mlx.Array, stepIdx int) *mlx.Array {
+	sigma := s.Sigmas[stepIdx]
+	sigmaNext := s.Sigmas[stepIdx+1]
+
+	// Euler step: x_{t-dt} = x_t + dt * v_t
+	dt := sigmaNext - sigma // Negative (going from noise to clean)
+
+	scaledOutput := mlx.MulScalar(modelOutput, dt)
+	return mlx.Add(sample, scaledOutput)
+}
+
+// InitNoise creates initial noise
+func (s *FlowMatchScheduler) InitNoise(shape []int32, seed int64) *mlx.Array {
+	return mlx.RandomNormalWithDtype(shape, uint64(seed), mlx.DtypeBFloat16)
+}
+
+// AddNoise adds noise to clean samples for a given timestep (for img2img)
+func (s *FlowMatchScheduler) AddNoise(cleanSample, noise *mlx.Array, timestepIdx int) *mlx.Array {
+	// In flow matching: x_t = (1-sigma) * x_0 + sigma * noise
+	// Use sigmas (shifted) for the interpolation
+	sigma := s.Sigmas[timestepIdx]
+	oneMinusSigma := 1.0 - sigma
+
+	scaledClean := mlx.MulScalar(cleanSample, oneMinusSigma)
+	scaledNoise := mlx.MulScalar(noise, sigma)
+
+	return mlx.Add(scaledClean, scaledNoise)
+}
+
+// GetTimesteps returns all timesteps
+func (s *FlowMatchScheduler) GetTimesteps() []float32 {
+	return s.Timesteps
+}

x/imagegen/models/glm_image/text_encoder.go

@@ -0,0 +1,497 @@
+//go:build mlx
+
+package glm_image
+
+import (
+	"encoding/json"
+	"fmt"
+	"math"
+	"os"
+	"path/filepath"
+	"regexp"
+
+	"github.com/ollama/ollama/x/imagegen"
+	"github.com/ollama/ollama/x/imagegen/mlx"
+	"github.com/ollama/ollama/x/imagegen/nn"
+	"github.com/ollama/ollama/x/imagegen/safetensors"
+)
+
+// T5Config holds T5 encoder configuration
+type T5Config struct {
+	DModel      int32   `json:"d_model"`               // 1472
+	DFF         int32   `json:"d_ff"`                  // 3584
+	DKV         int32   `json:"d_kv"`                  // 64
+	NumHeads    int32   `json:"num_heads"`             // 6
+	NumLayers   int32   `json:"num_layers"`            // 12
+	VocabSize   int32   `json:"vocab_size"`            // 384 (byte-level)
+	LayerNormEps float32 `json:"layer_norm_epsilon"`   // 1e-6
+	IsGatedAct  bool    `json:"is_gated_act"`          // true (gated-gelu)
+
+	// Relative position bias
+	RelativeAttentionNumBuckets  int32 `json:"relative_attention_num_buckets"`  // 32
+	RelativeAttentionMaxDistance int32 `json:"relative_attention_max_distance"` // 128
+}
+
+// T5TextEncoder is the T5 encoder for text conditioning
+type T5TextEncoder struct {
+	Config *T5Config
+
+	// Embedding (shared for ByT5)
+	SharedEmbed *nn.Embedding `weight:"shared"`
+
+	// Encoder layers
+	Layers []*T5Block `weight:"encoder.block"`
+
+	// Final layer norm
+	FinalNorm *T5LayerNorm `weight:"encoder.final_layer_norm"`
+
+	// Relative position bias (from first layer, shared across all)
+	RelativeAttentionBias *mlx.Array `weight:"encoder.block.0.layer.0.SelfAttention.relative_attention_bias.weight"`
+}
+
+// T5Block is a single T5 encoder block
+type T5Block struct {
+	// Self attention
+	Layer0 *T5LayerSelfAttention `weight:"layer.0"`
+	// FFN
+	Layer1 *T5LayerFF `weight:"layer.1"`
+}
+
+// T5LayerSelfAttention is T5's self-attention layer
+type T5LayerSelfAttention struct {
+	SelfAttention *T5Attention `weight:"SelfAttention"`
+	LayerNorm     *T5LayerNorm `weight:"layer_norm"`
+}
+
+// T5Attention implements T5's relative attention
+type T5Attention struct {
+	Q *mlx.Array `weight:"q.weight"` // No bias in T5
+	K *mlx.Array `weight:"k.weight"`
+	V *mlx.Array `weight:"v.weight"`
+	O *mlx.Array `weight:"o.weight"`
+
+	NHeads int32
+	DKV    int32
+	Scale  float32
+}
+
+// T5LayerFF is T5's feedforward layer with gated-gelu
+type T5LayerFF struct {
+	DenseReluDense *T5DenseGatedGelu `weight:"DenseReluDense"`
+	LayerNorm      *T5LayerNorm      `weight:"layer_norm"`
+}
+
+// T5DenseGatedGelu is T5's gated-gelu FFN
+type T5DenseGatedGelu struct {
+	Wi0 *mlx.Array `weight:"wi_0.weight"` // gate projection
+	Wi1 *mlx.Array `weight:"wi_1.weight"` // up projection
+	Wo  *mlx.Array `weight:"wo.weight"`   // down projection
+}
+
+// T5LayerNorm is T5's RMSNorm variant (no bias, no mean subtraction)
+type T5LayerNorm struct {
+	Weight *mlx.Array `weight:"weight"`
+	Eps    float32
+}
+
+// Load loads the T5 text encoder from manifest
+func (m *T5TextEncoder) Load(manifest *imagegen.ModelManifest) error {
+	fmt.Print("  Loading T5 text encoder... ")
+
+	// Load config
+	var cfg T5Config
+	if err := manifest.ReadConfigJSON("text_encoder/config.json", &cfg); err != nil {
+		return fmt.Errorf("config: %w", err)
+	}
+	m.Config = &cfg
+
+	// Pre-allocate layers
+	m.Layers = make([]*T5Block, cfg.NumLayers)
+
+	// Load weights
+	weights, err := imagegen.LoadWeightsFromManifest(manifest, "text_encoder")
+	if err != nil {
+		return fmt.Errorf("weights: %w", err)
+	}
+	if err := weights.Load(0); err != nil {
+		return fmt.Errorf("load weights: %w", err)
+	}
+	defer weights.ReleaseAll()
+
+	if err := safetensors.LoadModule(m, weights, ""); err != nil {
+		return fmt.Errorf("load module: %w", err)
+	}
+
+	m.initComputedFields()
+	fmt.Println("✓")
+	return nil
+}
+
+// LoadFromPath loads the T5 text encoder from a directory path
+func (m *T5TextEncoder) LoadFromPath(path string) error {
+	fmt.Print("  Loading T5 text encoder... ")
+
+	// Load config
+	var cfg T5Config
+	configPath := filepath.Join(path, "config.json")
+	data, err := os.ReadFile(configPath)
+	if err != nil {
+		return fmt.Errorf("read config: %w", err)
+	}
+	if err := json.Unmarshal(data, &cfg); err != nil {
+		return fmt.Errorf("parse config: %w", err)
+	}
+	m.Config = &cfg
+
+	// Pre-allocate layers
+	m.Layers = make([]*T5Block, cfg.NumLayers)
+
+	// Load weights from safetensors files
+	weights, err := safetensors.LoadModelWeights(path)
+	if err != nil {
+		return fmt.Errorf("weights: %w", err)
+	}
+	if err := weights.Load(0); err != nil {
+		return fmt.Errorf("load weights: %w", err)
+	}
+	defer weights.ReleaseAll()
+
+	if err := safetensors.LoadModule(m, weights, ""); err != nil {
+		return fmt.Errorf("load module: %w", err)
+	}
+
+	m.initComputedFields()
+	fmt.Println("✓")
+	return nil
+}
+
+func (m *T5TextEncoder) initComputedFields() {
+	cfg := m.Config
+	m.FinalNorm.Eps = cfg.LayerNormEps
+	for _, block := range m.Layers {
+		attn := block.Layer0.SelfAttention
+		attn.NHeads = cfg.NumHeads
+		attn.DKV = cfg.DKV
+		attn.Scale = float32(1.0 / math.Sqrt(float64(cfg.DKV)))
+
+		block.Layer0.LayerNorm.Eps = cfg.LayerNormEps
+		block.Layer1.LayerNorm.Eps = cfg.LayerNormEps
+	}
+}
+
+// Forward encodes text tokens
+func (m *T5TextEncoder) Forward(tokens *mlx.Array) *mlx.Array {
+	cfg := m.Config
+
+	// Get embeddings
+	h := m.SharedEmbed.Forward(tokens)
+
+	// Compute relative position bias once
+	seqLen := tokens.Shape()[1]
+	posBias := m.computeRelativePositionBias(seqLen)
+
+	// Forward through layers
+	for _, block := range m.Layers {
+		h = block.Forward(h, posBias, cfg.LayerNormEps)
+	}
+
+	// Final norm
+	h = m.FinalNorm.Forward(h)
+
+	return h
+}
+
+// extractGlyphTexts extracts quoted text (glyphs) from the prompt
+// This matches diffusers' get_glyph_texts from pipeline_glm_image.py
+// Glyph texts are used for text rendering guidance in the generated image
+func extractGlyphTexts(prompt string) []string {
+	var glyphTexts []string
+
+	// Extract text in single quotes: 'text'
+	re1 := regexp.MustCompile(`'([^']*)'`)
+	for _, match := range re1.FindAllStringSubmatch(prompt, -1) {
+		if len(match) > 1 {
+			glyphTexts = append(glyphTexts, match[1])
+		}
+	}
+
+	// Extract text in Unicode curly double quotes: "text"
+	re2 := regexp.MustCompile(`"([^""]*)"`)
+	for _, match := range re2.FindAllStringSubmatch(prompt, -1) {
+		if len(match) > 1 {
+			glyphTexts = append(glyphTexts, match[1])
+		}
+	}
+
+	// Extract text in ASCII double quotes: "text"
+	re3 := regexp.MustCompile(`"([^"]*)"`)
+	for _, match := range re3.FindAllStringSubmatch(prompt, -1) {
+		if len(match) > 1 {
+			glyphTexts = append(glyphTexts, match[1])
+		}
+	}
+
+	// Extract text in Japanese quotes: 「text」
+	re4 := regexp.MustCompile(`「([^「」]*)」`)
+	for _, match := range re4.FindAllStringSubmatch(prompt, -1) {
+		if len(match) > 1 {
+			glyphTexts = append(glyphTexts, match[1])
+		}
+	}
+
+	return glyphTexts
+}
+
+// EncodePrompt encodes the prompt text using the ByT5 tokenizer and encoder
+// This provides text conditioning for the diffusion transformer via the glyph projector
+//
+// IMPORTANT: This encodes only the GLYPH TEXTS (quoted strings in the prompt), not the
+// full prompt. Glyph texts are used for text rendering guidance in the generated image.
+// Multiple glyph texts are encoded and concatenated to form the conditioning signal.
+// This matches diffusers' _get_glyph_embeds() behavior.
+func (m *T5TextEncoder) EncodePrompt(tok *ByT5Tokenizer, prompt string) *mlx.Array {
+	// Extract glyph texts from prompt (text in quotes)
+	glyphTexts := extractGlyphTexts(prompt)
+
+	// If no glyph texts found, encode empty string (matches diffusers: [""] fallback)
+	if len(glyphTexts) == 0 {
+		glyphTexts = []string{""}
+	}
+
+	// Encode each glyph text and collect token sequences
+	// Matching diffusers' _get_glyph_embeds() which batches all glyph texts
+	var allTokenSeqs [][]int32
+
+	for _, glyphText := range glyphTexts {
+		// ByT5 uses byte-level encoding: each byte (0-255) -> token (3-258)
+		tokens := tok.Encode(glyphText)
+
+		// Add EOS token (1) at the end to match HuggingFace tokenizer behavior
+		tokens = append(tokens, tok.EOSTokenID)
+
+		allTokenSeqs = append(allTokenSeqs, tokens)
+	}
+
+	// Process each glyph text through the encoder
+	var allEmbeddings []*mlx.Array
+	for _, tokens := range allTokenSeqs {
+		tokenLen := len(tokens)
+		if tokenLen == 0 {
+			continue
+		}
+
+		// Create token array [1, L]
+		tokensArr := mlx.NewArrayInt32(tokens, []int32{1, int32(tokenLen)})
+
+		// Forward through encoder
+		output := m.Forward(tokensArr)
+		mlx.Eval(output)
+
+		allEmbeddings = append(allEmbeddings, output)
+	}
+
+	// Concatenate all glyph embeddings along sequence dimension
+	var output *mlx.Array
+	if len(allEmbeddings) == 0 {
+		// Fallback: return single zero embedding
+		output = mlx.Zeros([]int32{1, 1, m.Config.DModel}, mlx.DtypeBFloat16)
+	} else if len(allEmbeddings) == 1 {
+		output = allEmbeddings[0]
+	} else {
+		output = mlx.Concatenate(allEmbeddings, 1)
+	}
+	mlx.Eval(output)
+
+	return output
+}
+
+// computeRelativePositionBias computes T5's relative position encoding
+func (m *T5TextEncoder) computeRelativePositionBias(seqLen int32) *mlx.Array {
+	cfg := m.Config
+
+	// Create relative position matrix
+	// For each (query_pos, key_pos) pair, compute bucketed relative position
+	numBuckets := cfg.RelativeAttentionNumBuckets
+	maxDistance := cfg.RelativeAttentionMaxDistance
+
+	// Create position indices
+	contextPos := make([]int32, seqLen*seqLen)
+	memoryPos := make([]int32, seqLen*seqLen)
+	for i := int32(0); i < seqLen; i++ {
+		for j := int32(0); j < seqLen; j++ {
+			contextPos[i*seqLen+j] = i
+			memoryPos[i*seqLen+j] = j
+		}
+	}
+
+	// Compute relative positions and bucket them
+	buckets := make([]int32, seqLen*seqLen)
+	for i := int32(0); i < seqLen*seqLen; i++ {
+		relPos := memoryPos[i] - contextPos[i]
+		buckets[i] = relativePosistionBucket(relPos, numBuckets, maxDistance, false)
+	}
+
+	// Create bucket indices array
+	bucketsArr := mlx.NewArrayInt32(buckets, []int32{seqLen, seqLen})
+
+	// Look up bias: RelativeAttentionBias shape is [numBuckets, numHeads] = [32, 6]
+	// Take along axis 0 (buckets dimension) -> [seqLen, seqLen, numHeads]
+	bias := mlx.Take(m.RelativeAttentionBias, bucketsArr, 0) // [seqLen, seqLen, numHeads]
+
+	// Transpose to [numHeads, seqLen, seqLen]
+	bias = mlx.Transpose(bias, 2, 0, 1) // [numHeads, seqLen, seqLen]
+	bias = mlx.ExpandDims(bias, 0)      // [1, numHeads, seqLen, seqLen]
+
+	return bias
+}
+
+// relativePosistionBucket computes the bucket for a relative position
+func relativePosistionBucket(relativePosition int32, numBuckets int32, maxDistance int32, bidirectional bool) int32 {
+	var bucket int32 = 0
+	var n int32 = -relativePosition
+
+	if bidirectional {
+		numBuckets /= 2
+		if n < 0 {
+			bucket += numBuckets
+			n = -n
+		}
+	} else {
+		if n < 0 {
+			n = 0
+		}
+	}
+
+	// Half buckets are for exact positions, half are for log-spaced
+	maxExact := numBuckets / 2
+	if n < maxExact {
+		bucket += n
+	} else {
+		// Log-spaced buckets
+		logVal := math.Log(float64(n)/float64(maxExact)) / math.Log(float64(maxDistance)/float64(maxExact))
+		bucket += maxExact + int32(logVal*float64(numBuckets-maxExact))
+		if bucket > numBuckets-1 {
+			bucket = numBuckets - 1
+		}
+	}
+
+	return bucket
+}
+
+// Forward for T5Block
+func (b *T5Block) Forward(x *mlx.Array, posBias *mlx.Array, eps float32) *mlx.Array {
+	// Self attention with residual
+	h := b.Layer0.Forward(x, posBias, eps)
+
+	// FFN with residual
+	h = b.Layer1.Forward(h, eps)
+
+	return h
+}
+
+// Forward for T5LayerSelfAttention
+func (l *T5LayerSelfAttention) Forward(x *mlx.Array, posBias *mlx.Array, eps float32) *mlx.Array {
+	// Pre-norm
+	normed := l.LayerNorm.Forward(x)
+
+	// Attention
+	attnOut := l.SelfAttention.Forward(normed, posBias)
+
+	// Residual
+	return mlx.Add(x, attnOut)
+}
+
+// Forward for T5Attention
+func (attn *T5Attention) Forward(x *mlx.Array, posBias *mlx.Array) *mlx.Array {
+	shape := x.Shape()
+	B := shape[0]
+	L := shape[1]
+	D := shape[2]
+
+	// Q, K, V projections (no bias)
+	// Weights are [out_features, in_features], so we use matmul with transpose
+	q := mlx.Matmul(x, mlx.Transpose(attn.Q, 1, 0))
+	k := mlx.Matmul(x, mlx.Transpose(attn.K, 1, 0))
+	v := mlx.Matmul(x, mlx.Transpose(attn.V, 1, 0))
+
+	// Reshape to [B, L, nheads, d_kv]
+	q = mlx.Reshape(q, B, L, attn.NHeads, attn.DKV)
+	k = mlx.Reshape(k, B, L, attn.NHeads, attn.DKV)
+	v = mlx.Reshape(v, B, L, attn.NHeads, attn.DKV)
+
+	// Transpose to [B, nheads, L, d_kv]
+	q = mlx.Transpose(q, 0, 2, 1, 3)
+	k = mlx.Transpose(k, 0, 2, 1, 3)
+	v = mlx.Transpose(v, 0, 2, 1, 3)
+
+	// Attention scores with relative position bias
+	// T5 uses UNSCALED dot-product attention: scores = q @ k.T + pos_bias
+	// (no 1/sqrt(d_k) scale factor like in standard transformers)
+	scores := mlx.Matmul(q, mlx.Transpose(k, 0, 1, 3, 2))
+	scores = mlx.Add(scores, posBias)
+
+	// Softmax
+	attnWeights := mlx.Softmax(scores, -1)
+
+	// Attend to values
+	out := mlx.Matmul(attnWeights, v)
+
+	// Transpose back [B, nheads, L, d_kv] -> [B, L, nheads, d_kv]
+	out = mlx.Transpose(out, 0, 2, 1, 3)
+	// Reshape to [B, L, D]
+	out = mlx.Reshape(out, B, L, attn.NHeads*attn.DKV)
+
+	// Output projection
+	out = mlx.Matmul(out, mlx.Transpose(attn.O, 1, 0))
+
+	_ = D // Silence unused warning
+	return out
+}
+
+// Forward for T5LayerFF
+func (l *T5LayerFF) Forward(x *mlx.Array, eps float32) *mlx.Array {
+	// Pre-norm
+	normed := l.LayerNorm.Forward(x)
+
+	// FFN
+	ffOut := l.DenseReluDense.Forward(normed)
+
+	// Residual
+	return mlx.Add(x, ffOut)
+}
+
+// geluNew implements the GELU activation with tanh approximation (gelu_new)
+// This matches HuggingFace transformers' gelu_new/OpenAI GPT implementation
+// Formula: 0.5 * x * (1 + tanh(sqrt(2/π) * (x + 0.044715 * x³)))
+func geluNew(x *mlx.Array) *mlx.Array {
+	sqrt2OverPi := float32(0.7978845608) // sqrt(2/π)
+	coeff := float32(0.044715)
+
+	x3 := mlx.Mul(mlx.Mul(x, x), x)
+	inner := mlx.MulScalar(mlx.Add(x, mlx.MulScalar(x3, coeff)), sqrt2OverPi)
+	return mlx.Mul(mlx.MulScalar(x, 0.5), mlx.AddScalar(mlx.Tanh(inner), 1.0))
+}
+
+// Forward for T5DenseGatedGelu (gated-gelu activation)
+func (d *T5DenseGatedGelu) Forward(x *mlx.Array) *mlx.Array {
+	// Gate projection with GELU activation (T5 v1.1/ByT5 uses gelu_new)
+	gate := mlx.Matmul(x, mlx.Transpose(d.Wi0, 1, 0))
+	gate = geluNew(gate)
+
+	// Up projection
+	up := mlx.Matmul(x, mlx.Transpose(d.Wi1, 1, 0))
+
+	// Gated output
+	h := mlx.Mul(gate, up)
+
+	// Down projection
+	return mlx.Matmul(h, mlx.Transpose(d.Wo, 1, 0))
+}
+
+// Forward for T5LayerNorm (RMSNorm variant)
+func (ln *T5LayerNorm) Forward(x *mlx.Array) *mlx.Array {
+	// T5 uses RMSNorm: x * rsqrt(mean(x^2) + eps) * weight
+	variance := mlx.Mean(mlx.Square(x), -1, true)
+	x = mlx.Mul(x, mlx.RSqrt(mlx.AddScalar(variance, ln.Eps)))
+	return mlx.Mul(x, ln.Weight)
+}

x/imagegen/models/glm_image/vae.go

@@ -0,0 +1,477 @@
+//go:build mlx
+
+package glm_image
+
+import (
+	"encoding/json"
+	"fmt"
+	"os"
+	"path/filepath"
+
+	"github.com/ollama/ollama/x/imagegen"
+	"github.com/ollama/ollama/x/imagegen/mlx"
+	"github.com/ollama/ollama/x/imagegen/safetensors"
+)
+
+// VAEConfig holds VAE decoder configuration
+type VAEConfig struct {
+	InChannels       int32     `json:"in_channels"`        // 3
+	OutChannels      int32     `json:"out_channels"`       // 3
+	LatentChannels   int32     `json:"latent_channels"`    // 16
+	BlockOutChannels []int32   `json:"block_out_channels"` // [128, 512, 1024, 1024]
+	LayersPerBlock   int32     `json:"layers_per_block"`   // 3
+	NormNumGroups    int32     `json:"norm_num_groups"`    // 32
+	ScalingFactor    float32   `json:"scaling_factor"`     // 0.18215
+	ShiftFactor      *float32  `json:"shift_factor"`       // null
+	LatentsMean      []float32 `json:"latents_mean"`       // [16 values]
+	LatentsStd       []float32 `json:"latents_std"`        // [16 values]
+}
+
+// VAEDecoder is the VAE latent decoder
+type VAEDecoder struct {
+	Config *VAEConfig
+
+	// Decoder components
+	ConvIn  *VAEConv2d    `weight:"decoder.conv_in"`
+	MidBlock *VAEMidBlock `weight:"decoder.mid_block"`
+	UpBlocks []*VAEUpBlock `weight:"decoder.up_blocks"`
+	ConvNormOut *GroupNorm `weight:"decoder.conv_norm_out"`
+	ConvOut *VAEConv2d    `weight:"decoder.conv_out"`
+}
+
+// VAEConv2d is a 2D convolution layer
+type VAEConv2d struct {
+	Weight *mlx.Array `weight:"weight"`
+	Bias   *mlx.Array `weight:"bias"`
+	Stride int32
+	Padding int32
+}
+
+// GroupNorm is group normalization
+type GroupNorm struct {
+	Weight    *mlx.Array `weight:"weight"`
+	Bias      *mlx.Array `weight:"bias"`
+	NumGroups int32
+	Eps       float32
+}
+
+// VAEMidBlock is the middle block of the VAE
+type VAEMidBlock struct {
+	Resnets []*VAEResnetBlock `weight:"resnets"`
+}
+
+// VAEUpBlock is an upsampling block
+type VAEUpBlock struct {
+	Resnets []*VAEResnetBlock `weight:"resnets"`
+	Upsamplers []*VAEUpsampler `weight:"upsamplers"`
+}
+
+// VAEResnetBlock is a residual block
+type VAEResnetBlock struct {
+	Norm1   *GroupNorm `weight:"norm1"`
+	Conv1   *VAEConv2d `weight:"conv1"`
+	Norm2   *GroupNorm `weight:"norm2"`
+	Conv2   *VAEConv2d `weight:"conv2"`
+	ConvShortcut *VAEConv2d `weight:"conv_shortcut,optional"` // Optional, for channel mismatch
+}
+
+// VAEUpsampler is an upsampling layer
+type VAEUpsampler struct {
+	Conv *VAEConv2d `weight:"conv"`
+}
+
+// Load loads the VAE decoder from manifest
+func (m *VAEDecoder) Load(manifest *imagegen.ModelManifest) error {
+	fmt.Print("  Loading VAE decoder... ")
+
+	// Load config
+	var cfg VAEConfig
+	if err := manifest.ReadConfigJSON("vae/config.json", &cfg); err != nil {
+		return fmt.Errorf("config: %w", err)
+	}
+	m.Config = &cfg
+
+	// Initialize structure based on config
+	numBlocks := len(cfg.BlockOutChannels)
+	m.UpBlocks = make([]*VAEUpBlock, numBlocks)
+
+	// Pre-allocate MidBlock resnets (VAE mid_block typically has 2 resnets)
+	m.MidBlock = &VAEMidBlock{
+		Resnets: make([]*VAEResnetBlock, 2),
+	}
+
+	// Pre-allocate UpBlocks with their resnets and upsamplers
+	// VAE decoder has layers_per_block+1 resnets per up_block (to match encoder)
+	// And all but the last up_block has an upsampler
+	for i := 0; i < numBlocks; i++ {
+		numResnets := cfg.LayersPerBlock + 1 // typically 4 resnets
+		m.UpBlocks[i] = &VAEUpBlock{
+			Resnets: make([]*VAEResnetBlock, numResnets),
+		}
+		// All but the last block has upsamplers
+		if i < numBlocks-1 {
+			m.UpBlocks[i].Upsamplers = make([]*VAEUpsampler, 1)
+		}
+	}
+
+	// Load weights
+	weights, err := imagegen.LoadWeightsFromManifest(manifest, "vae")
+	if err != nil {
+		return fmt.Errorf("weights: %w", err)
+	}
+	if err := weights.Load(mlx.DtypeBFloat16); err != nil {
+		return fmt.Errorf("load weights: %w", err)
+	}
+	defer weights.ReleaseAll()
+
+	if err := safetensors.LoadModule(m, weights, ""); err != nil {
+		return fmt.Errorf("load module: %w", err)
+	}
+
+	// Initialize GroupNorm parameters
+	m.initGroupNorms()
+
+	fmt.Println("✓")
+	return nil
+}
+
+// LoadFromPath loads the VAE decoder from a directory path
+func (m *VAEDecoder) LoadFromPath(path string) error {
+	fmt.Print("  Loading VAE decoder... ")
+
+	// Load config
+	var cfg VAEConfig
+	configPath := filepath.Join(path, "config.json")
+	data, err := os.ReadFile(configPath)
+	if err != nil {
+		return fmt.Errorf("read config: %w", err)
+	}
+	if err := json.Unmarshal(data, &cfg); err != nil {
+		return fmt.Errorf("parse config: %w", err)
+	}
+	m.Config = &cfg
+
+	// Initialize structure based on config
+	numBlocks := len(cfg.BlockOutChannels)
+	m.UpBlocks = make([]*VAEUpBlock, numBlocks)
+
+	// Pre-allocate MidBlock resnets (VAE mid_block typically has 2 resnets)
+	m.MidBlock = &VAEMidBlock{
+		Resnets: make([]*VAEResnetBlock, 2),
+	}
+
+	// Pre-allocate UpBlocks with their resnets and upsamplers
+	for i := 0; i < numBlocks; i++ {
+		numResnets := cfg.LayersPerBlock + 1
+		m.UpBlocks[i] = &VAEUpBlock{
+			Resnets: make([]*VAEResnetBlock, numResnets),
+		}
+		if i < numBlocks-1 {
+			m.UpBlocks[i].Upsamplers = make([]*VAEUpsampler, 1)
+		}
+	}
+
+	// Load weights from safetensors files
+	weights, err := safetensors.LoadModelWeights(path)
+	if err != nil {
+		return fmt.Errorf("weights: %w", err)
+	}
+	if err := weights.Load(mlx.DtypeBFloat16); err != nil {
+		return fmt.Errorf("load weights: %w", err)
+	}
+	defer weights.ReleaseAll()
+
+	if err := safetensors.LoadModule(m, weights, ""); err != nil {
+		return fmt.Errorf("load module: %w", err)
+	}
+
+	// Initialize GroupNorm parameters
+	m.initGroupNorms()
+
+	fmt.Println("✓")
+	return nil
+}
+
+func (m *VAEDecoder) initGroupNorms() {
+	cfg := m.Config
+	numGroups := cfg.NormNumGroups
+	eps := float32(1e-6) // Must match diffusers VAE (1e-6, not 1e-5)
+
+	if m.ConvNormOut != nil {
+		m.ConvNormOut.NumGroups = numGroups
+		m.ConvNormOut.Eps = eps
+	}
+
+	if m.MidBlock != nil {
+		for _, resnet := range m.MidBlock.Resnets {
+			if resnet.Norm1 != nil {
+				resnet.Norm1.NumGroups = numGroups
+				resnet.Norm1.Eps = eps
+			}
+			if resnet.Norm2 != nil {
+				resnet.Norm2.NumGroups = numGroups
+				resnet.Norm2.Eps = eps
+			}
+		}
+	}
+
+	for _, upBlock := range m.UpBlocks {
+		if upBlock == nil {
+			continue
+		}
+		for _, resnet := range upBlock.Resnets {
+			if resnet == nil {
+				continue
+			}
+			if resnet.Norm1 != nil {
+				resnet.Norm1.NumGroups = numGroups
+				resnet.Norm1.Eps = eps
+			}
+			if resnet.Norm2 != nil {
+				resnet.Norm2.NumGroups = numGroups
+				resnet.Norm2.Eps = eps
+			}
+		}
+	}
+}
+
+// Decode decodes latents to an image
+func (m *VAEDecoder) Decode(latents *mlx.Array) *mlx.Array {
+	cfg := m.Config
+
+	// Apply latent denormalization if mean/std are provided
+	// This matches diffusers GLM-Image: latents = latents * std + mean
+	// Note: GLM-Image does NOT divide by scaling_factor (unlike standard SD VAEs)
+	if len(cfg.LatentsMean) > 0 && len(cfg.LatentsStd) > 0 {
+		latents = m.denormalizeLatents(latents)
+	}
+
+	// Convert from NCHW to NHWC for processing
+	// [B, C, H, W] -> [B, H, W, C]
+	x := mlx.Transpose(latents, 0, 2, 3, 1)
+
+	// Initial convolution
+	x = m.ConvIn.Forward(x)
+
+	// Mid block
+	x = m.MidBlock.Forward(x)
+
+	// Up blocks (forward order - index 0 is at lowest resolution/highest channels)
+	for i := 0; i < len(m.UpBlocks); i++ {
+		if m.UpBlocks[i] != nil {
+			x = m.UpBlocks[i].Forward(x)
+		}
+	}
+
+	// Final normalization and convolution
+	x = m.ConvNormOut.Forward(x)
+	x = mlx.SiLU(x)
+	x = m.ConvOut.Forward(x)
+
+	// Convert back to NCHW
+	// [B, H, W, C] -> [B, C, H, W]
+	x = mlx.Transpose(x, 0, 3, 1, 2)
+
+	// Clamp to valid range and convert to [0, 1]
+	x = mlx.ClipScalar(x, -1.0, 1.0, true, true)
+	x = mlx.AddScalar(x, 1.0)
+	x = mlx.DivScalar(x, 2.0)
+
+	return x
+}
+
+// denormalizeLatents applies the latent mean/std denormalization
+func (m *VAEDecoder) denormalizeLatents(latents *mlx.Array) *mlx.Array {
+	cfg := m.Config
+
+	// Create mean and std arrays [1, C, 1, 1] for broadcasting
+	mean := mlx.NewArray(cfg.LatentsMean, []int32{1, int32(len(cfg.LatentsMean)), 1, 1})
+	std := mlx.NewArray(cfg.LatentsStd, []int32{1, int32(len(cfg.LatentsStd)), 1, 1})
+
+	// Denormalize: latents * std + mean
+	latents = mlx.Mul(latents, std)
+	latents = mlx.Add(latents, mean)
+
+	return latents
+}
+
+// Forward for VAEConv2d
+func (c *VAEConv2d) Forward(x *mlx.Array) *mlx.Array {
+	// x: [B, H, W, C_in] (NHWC)
+	// PyTorch weight: [C_out, C_in, kH, kW] (OIHW)
+	// MLX conv2d expects weight: [C_out, kH, kW, C_in] (OHWI)
+	// So we need to transpose from OIHW to OHWI
+
+	stride := c.Stride
+	if stride == 0 {
+		stride = 1
+	}
+	padding := c.Padding
+	if padding == 0 {
+		// Default to same padding for 3x3 kernels
+		wShape := c.Weight.Shape()
+		if len(wShape) >= 3 && wShape[2] == 3 {
+			padding = 1
+		}
+	}
+
+	// Transpose weight from OIHW [out, in, h, w] to OHWI [out, h, w, in]
+	weight := mlx.Transpose(c.Weight, 0, 2, 3, 1)
+
+	out := mlx.Conv2d(x, weight, stride, padding)
+	if c.Bias != nil {
+		// Bias: [C_out] -> [1, 1, 1, C_out]
+		bias := mlx.Reshape(c.Bias, 1, 1, 1, -1)
+		out = mlx.Add(out, bias)
+	}
+	return out
+}
+
+// Forward for GroupNorm
+func (gn *GroupNorm) Forward(x *mlx.Array) *mlx.Array {
+	// x: [B, H, W, C] (NHWC)
+	shape := x.Shape()
+	B := shape[0]
+	H := shape[1]
+	W := shape[2]
+	C := shape[3]
+
+	numGroups := gn.NumGroups
+	if numGroups == 0 {
+		numGroups = 32
+	}
+	groupSize := C / numGroups
+
+	// Reshape to [B, H, W, groups, groupSize]
+	x = mlx.Reshape(x, B, H, W, numGroups, groupSize)
+
+	// Compute mean and variance per group
+	mean := mlx.Mean(x, 1, true)
+	mean = mlx.Mean(mean, 2, true)
+	mean = mlx.Mean(mean, 4, true)
+
+	xCentered := mlx.Sub(x, mean)
+	variance := mlx.Mean(mlx.Square(xCentered), 1, true)
+	variance = mlx.Mean(variance, 2, true)
+	variance = mlx.Mean(variance, 4, true)
+
+	// Normalize
+	xNorm := mlx.Div(xCentered, mlx.Sqrt(mlx.AddScalar(variance, gn.Eps)))
+
+	// Reshape back
+	xNorm = mlx.Reshape(xNorm, B, H, W, C)
+
+	// Scale and shift
+	if gn.Weight != nil {
+		weight := mlx.Reshape(gn.Weight, 1, 1, 1, C)
+		xNorm = mlx.Mul(xNorm, weight)
+	}
+	if gn.Bias != nil {
+		bias := mlx.Reshape(gn.Bias, 1, 1, 1, C)
+		xNorm = mlx.Add(xNorm, bias)
+	}
+
+	return xNorm
+}
+
+// Forward for VAEMidBlock
+func (mb *VAEMidBlock) Forward(x *mlx.Array) *mlx.Array {
+	for _, resnet := range mb.Resnets {
+		x = resnet.Forward(x)
+	}
+	return x
+}
+
+// Forward for VAEUpBlock
+func (ub *VAEUpBlock) Forward(x *mlx.Array) *mlx.Array {
+	// Apply resnets
+	for _, resnet := range ub.Resnets {
+		if resnet != nil {
+			x = resnet.Forward(x)
+		}
+	}
+
+	// Apply upsamplers
+	for _, upsampler := range ub.Upsamplers {
+		if upsampler != nil {
+			x = upsampler.Forward(x)
+		}
+	}
+
+	return x
+}
+
+// Forward for VAEResnetBlock
+func (rb *VAEResnetBlock) Forward(x *mlx.Array) *mlx.Array {
+	residual := x
+
+	// First norm + activation + conv
+	h := rb.Norm1.Forward(x)
+	h = mlx.SiLU(h)
+	h = rb.Conv1.Forward(h)
+
+	// Second norm + activation + conv
+	h = rb.Norm2.Forward(h)
+	h = mlx.SiLU(h)
+	h = rb.Conv2.Forward(h)
+
+	// Shortcut for channel mismatch
+	if rb.ConvShortcut != nil {
+		residual = rb.ConvShortcut.Forward(residual)
+	}
+
+	return mlx.Add(h, residual)
+}
+
+// Forward for VAEUpsampler (2x nearest neighbor upsample + conv)
+func (us *VAEUpsampler) Forward(x *mlx.Array) *mlx.Array {
+	// x: [B, H, W, C]
+	// 2x nearest neighbor upsample
+	x = upsample2x(x)
+
+	// Conv
+	if us.Conv != nil {
+		x = us.Conv.Forward(x)
+	}
+
+	return x
+}
+
+// upsample2x performs 2x nearest neighbor upsampling.
+// Input and output are in NHWC format: [B, H, W, C] -> [B, H*2, W*2, C]
+func upsample2x(x *mlx.Array) *mlx.Array {
+	shape := x.Shape()
+	B := shape[0]
+	H := shape[1]
+	W := shape[2]
+	C := shape[3]
+
+	// Create indices [0, 0, 1, 1, 2, 2, ...] for nearest neighbor
+	hIndices := make([]int32, H*2)
+	for i := int32(0); i < H; i++ {
+		hIndices[i*2] = i
+		hIndices[i*2+1] = i
+	}
+	wIndices := make([]int32, W*2)
+	for i := int32(0); i < W; i++ {
+		wIndices[i*2] = i
+		wIndices[i*2+1] = i
+	}
+
+	hIdx := mlx.NewArrayInt32(hIndices, []int32{H * 2})
+	wIdx := mlx.NewArrayInt32(wIndices, []int32{W * 2})
+
+	// Take along height axis
+	x = mlx.Reshape(x, B*H, W, C)
+	x = mlx.Take(x, wIdx, 1) // [B*H, W*2, C]
+	x = mlx.Reshape(x, B, H, W*2, C)
+
+	// Take along width axis - transpose to [B, W*2, H, C], take, transpose back
+	x = mlx.Transpose(x, 0, 2, 1, 3) // [B, W*2, H, C]
+	x = mlx.Reshape(x, B*(W*2), H, C)
+	x = mlx.Take(x, hIdx, 1) // [B*(W*2), H*2, C]
+	x = mlx.Reshape(x, B, W*2, H*2, C)
+	x = mlx.Transpose(x, 0, 2, 1, 3) // [B, H*2, W*2, C]
+
+	return x
+}

x/imagegen/models/glm_image/vision_language_encoder.go

@@ -0,0 +1,982 @@
+//go:build mlx
+
+package glm_image
+
+import (
+	"encoding/json"
+	"fmt"
+	"math"
+	"os"
+	"path/filepath"
+
+	"github.com/ollama/ollama/x/imagegen"
+	"github.com/ollama/ollama/x/imagegen/cache"
+	"github.com/ollama/ollama/x/imagegen/mlx"
+	"github.com/ollama/ollama/x/imagegen/nn"
+	"github.com/ollama/ollama/x/imagegen/safetensors"
+)
+
+// VisionLanguageConfig holds GLM-Image AR generator configuration
+type VisionLanguageConfig struct {
+	// Text model config
+	HiddenSize        int32   `json:"hidden_size"`         // 4096
+	NumHiddenLayers   int32   `json:"num_hidden_layers"`   // 40
+	IntermediateSize  int32   `json:"intermediate_size"`   // 13696
+	NumAttentionHeads int32   `json:"num_attention_heads"` // 32
+	NumKeyValueHeads  int32   `json:"num_key_value_heads"` // 2
+	VocabSize         int32   `json:"vocab_size"`          // 168064
+	RMSNormEps        float32 `json:"rms_norm_eps"`        // 1e-5
+
+	// RoPE config
+	RopeTheta           float32 `json:"rope_theta"`            // 10000
+	PartialRotaryFactor float32 `json:"partial_rotary_factor"` // 0.5
+	MRoPESection        []int32 `json:"mrope_section"`         // [8, 12, 12]
+
+	// Visual token config
+	VisionVocabSize   int32 `json:"vision_vocab_size"`    // 16512
+	ImageStartTokenID int32 `json:"image_start_token_id"` // 16384
+	ImageEndTokenID   int32 `json:"image_end_token_id"`   // 16385
+	ImageTokenID      int32 `json:"image_token_id"`       // 167855
+
+	// Computed
+	HeadDim int32
+}
+
+// VisionLanguageEncoder is the 9B AR generator
+type VisionLanguageEncoder struct {
+	Config *VisionLanguageConfig
+
+	// Embedding
+	EmbedTokens *nn.Embedding `weight:"model.language_model.embed_tokens"`
+
+	// Transformer layers
+	Layers []*GLMBlock `weight:"model.language_model.layers"`
+
+	// Final norm
+	FinalNorm *nn.RMSNorm `weight:"model.language_model.norm"`
+
+	// LM Head
+	LMHead *mlx.Array `weight:"lm_head.weight"`
+}
+
+// GLMBlock is a single transformer block in GLM-4 style
+type GLMBlock struct {
+	// Pre-attention norm (GLM uses post-LN variant)
+	InputLayerNorm    *nn.RMSNorm `weight:"input_layernorm"`
+	PostSelfAttnNorm  *nn.RMSNorm `weight:"post_self_attn_layernorm"`
+	PostAttnLayerNorm *nn.RMSNorm `weight:"post_attention_layernorm"`
+	PostMLPLayerNorm  *nn.RMSNorm `weight:"post_mlp_layernorm"`
+
+	// Attention
+	SelfAttn *GLMAttention `weight:"self_attn"`
+
+	// MLP (fused gate_up)
+	MLP *GLMMLP `weight:"mlp"`
+}
+
+// GLMAttention implements GQA with partial rotary and MRoPE
+type GLMAttention struct {
+	QProj *mlx.Array `weight:"q_proj.weight"`
+	KProj *mlx.Array `weight:"k_proj.weight"`
+	VProj *mlx.Array `weight:"v_proj.weight"`
+	OProj *mlx.Array `weight:"o_proj.weight"`
+
+	// QKV have biases in GLM
+	QBias *mlx.Array `weight:"q_proj.bias"`
+	KBias *mlx.Array `weight:"k_proj.bias"`
+	VBias *mlx.Array `weight:"v_proj.bias"`
+
+	// Computed
+	NHeads           int32
+	NKVHeads         int32
+	HeadDim          int32
+	Scale            float32
+	PartialRotary    float32   // Only rotate this fraction of head_dim
+	RopeTheta        float32
+	MRoPESection     []int32   // [8, 12, 12] - frequency pairs per dimension (temporal, height, width)
+}
+
+// ARCache holds KV caches for all layers using the shared cache implementation
+type ARCache struct {
+	Layers []cache.Cache
+}
+
+// NewARCache creates a new cache for the given number of layers
+func NewARCache(numLayers int32) *ARCache {
+	layers := make([]cache.Cache, numLayers)
+	for i := range layers {
+		layers[i] = cache.NewKVCache()
+	}
+	return &ARCache{Layers: layers}
+}
+
+// Free releases all cached tensors
+func (c *ARCache) Free() {
+	for _, layer := range c.Layers {
+		for _, arr := range layer.State() {
+			if arr != nil {
+				arr.Free()
+			}
+		}
+	}
+}
+
+// GLMMLP implements fused gate_up SwiGLU MLP
+type GLMMLP struct {
+	// GLM uses fused gate_up_proj: [hidden, 2*intermediate]
+	GateUpProj *mlx.Array `weight:"gate_up_proj.weight"`
+	DownProj   *mlx.Array `weight:"down_proj.weight"`
+}
+
+// Load loads the vision-language encoder from manifest
+func (m *VisionLanguageEncoder) Load(manifest *imagegen.ModelManifest) error {
+	fmt.Print("  Loading vision-language encoder... ")
+
+	// Load config
+	var rawCfg struct {
+		TextConfig struct {
+			HiddenSize        int32   `json:"hidden_size"`
+			NumHiddenLayers   int32   `json:"num_hidden_layers"`
+			IntermediateSize  int32   `json:"intermediate_size"`
+			NumAttentionHeads int32   `json:"num_attention_heads"`
+			NumKeyValueHeads  int32   `json:"num_key_value_heads"`
+			VocabSize         int32   `json:"vocab_size"`
+			RMSNormEps        float32 `json:"rms_norm_eps"`
+			VisionVocabSize   int32   `json:"vision_vocab_size"`
+			RopeParameters    struct {
+				RopeTheta           float32 `json:"rope_theta"`
+				PartialRotaryFactor float32 `json:"partial_rotary_factor"`
+				MRoPESection        []int32 `json:"mrope_section"`
+			} `json:"rope_parameters"`
+		} `json:"text_config"`
+		ImageStartTokenID int32 `json:"image_start_token_id"`
+		ImageEndTokenID   int32 `json:"image_end_token_id"`
+		ImageTokenID      int32 `json:"image_token_id"`
+	}
+
+	if err := manifest.ReadConfigJSON("vision_language_encoder/config.json", &rawCfg); err != nil {
+		return fmt.Errorf("config: %w", err)
+	}
+
+	cfg := &VisionLanguageConfig{
+		HiddenSize:          rawCfg.TextConfig.HiddenSize,
+		NumHiddenLayers:     rawCfg.TextConfig.NumHiddenLayers,
+		IntermediateSize:    rawCfg.TextConfig.IntermediateSize,
+		NumAttentionHeads:   rawCfg.TextConfig.NumAttentionHeads,
+		NumKeyValueHeads:    rawCfg.TextConfig.NumKeyValueHeads,
+		VocabSize:           rawCfg.TextConfig.VocabSize,
+		RMSNormEps:          rawCfg.TextConfig.RMSNormEps,
+		VisionVocabSize:     rawCfg.TextConfig.VisionVocabSize,
+		RopeTheta:           rawCfg.TextConfig.RopeParameters.RopeTheta,
+		PartialRotaryFactor: rawCfg.TextConfig.RopeParameters.PartialRotaryFactor,
+		MRoPESection:        rawCfg.TextConfig.RopeParameters.MRoPESection,
+		ImageStartTokenID:   rawCfg.ImageStartTokenID,
+		ImageEndTokenID:     rawCfg.ImageEndTokenID,
+		ImageTokenID:        rawCfg.ImageTokenID,
+	}
+
+	cfg.HeadDim = cfg.HiddenSize / cfg.NumAttentionHeads
+	m.Config = cfg
+
+	// Pre-allocate layers
+	m.Layers = make([]*GLMBlock, cfg.NumHiddenLayers)
+
+	// Load weights
+	weights, err := imagegen.LoadWeightsFromManifest(manifest, "vision_language_encoder")
+	if err != nil {
+		return fmt.Errorf("weights: %w", err)
+	}
+	if err := weights.Load(mlx.DtypeBFloat16); err != nil {
+		return fmt.Errorf("load weights: %w", err)
+	}
+	defer weights.ReleaseAll()
+
+	if err := safetensors.LoadModule(m, weights, ""); err != nil {
+		return fmt.Errorf("load module: %w", err)
+	}
+
+	m.initComputedFields()
+	fmt.Printf("✓ [%d layers]\n", cfg.NumHiddenLayers)
+	return nil
+}
+
+// LoadFromPath loads the vision-language encoder from a directory path
+func (m *VisionLanguageEncoder) LoadFromPath(path string) error {
+	fmt.Print("  Loading vision-language encoder... ")
+
+	// Load config
+	var rawCfg struct {
+		TextConfig struct {
+			HiddenSize        int32   `json:"hidden_size"`
+			NumHiddenLayers   int32   `json:"num_hidden_layers"`
+			IntermediateSize  int32   `json:"intermediate_size"`
+			NumAttentionHeads int32   `json:"num_attention_heads"`
+			NumKeyValueHeads  int32   `json:"num_key_value_heads"`
+			VocabSize         int32   `json:"vocab_size"`
+			RMSNormEps        float32 `json:"rms_norm_eps"`
+			VisionVocabSize   int32   `json:"vision_vocab_size"`
+			RopeParameters    struct {
+				RopeTheta           float32 `json:"rope_theta"`
+				PartialRotaryFactor float32 `json:"partial_rotary_factor"`
+				MRoPESection        []int32 `json:"mrope_section"`
+			} `json:"rope_parameters"`
+		} `json:"text_config"`
+		ImageStartTokenID int32 `json:"image_start_token_id"`
+		ImageEndTokenID   int32 `json:"image_end_token_id"`
+		ImageTokenID      int32 `json:"image_token_id"`
+	}
+
+	configPath := filepath.Join(path, "config.json")
+	data, err := os.ReadFile(configPath)
+	if err != nil {
+		return fmt.Errorf("read config: %w", err)
+	}
+	if err := json.Unmarshal(data, &rawCfg); err != nil {
+		return fmt.Errorf("parse config: %w", err)
+	}
+
+	cfg := &VisionLanguageConfig{
+		HiddenSize:          rawCfg.TextConfig.HiddenSize,
+		NumHiddenLayers:     rawCfg.TextConfig.NumHiddenLayers,
+		IntermediateSize:    rawCfg.TextConfig.IntermediateSize,
+		NumAttentionHeads:   rawCfg.TextConfig.NumAttentionHeads,
+		NumKeyValueHeads:    rawCfg.TextConfig.NumKeyValueHeads,
+		VocabSize:           rawCfg.TextConfig.VocabSize,
+		RMSNormEps:          rawCfg.TextConfig.RMSNormEps,
+		VisionVocabSize:     rawCfg.TextConfig.VisionVocabSize,
+		RopeTheta:           rawCfg.TextConfig.RopeParameters.RopeTheta,
+		PartialRotaryFactor: rawCfg.TextConfig.RopeParameters.PartialRotaryFactor,
+		MRoPESection:        rawCfg.TextConfig.RopeParameters.MRoPESection,
+		ImageStartTokenID:   rawCfg.ImageStartTokenID,
+		ImageEndTokenID:     rawCfg.ImageEndTokenID,
+		ImageTokenID:        rawCfg.ImageTokenID,
+	}
+
+	cfg.HeadDim = cfg.HiddenSize / cfg.NumAttentionHeads
+	m.Config = cfg
+
+	// Pre-allocate layers
+	m.Layers = make([]*GLMBlock, cfg.NumHiddenLayers)
+
+	// Load weights
+	weights, err := safetensors.LoadModelWeights(path)
+	if err != nil {
+		return fmt.Errorf("weights: %w", err)
+	}
+	if err := weights.Load(mlx.DtypeBFloat16); err != nil {
+		return fmt.Errorf("load weights: %w", err)
+	}
+	defer weights.ReleaseAll()
+
+	if err := safetensors.LoadModule(m, weights, ""); err != nil {
+		return fmt.Errorf("load module: %w", err)
+	}
+
+	m.initComputedFields()
+	fmt.Printf("✓ [%d layers]\n", cfg.NumHiddenLayers)
+	return nil
+}
+
+func (m *VisionLanguageEncoder) initComputedFields() {
+	cfg := m.Config
+	for _, block := range m.Layers {
+		block.SelfAttn.NHeads = cfg.NumAttentionHeads
+		block.SelfAttn.NKVHeads = cfg.NumKeyValueHeads
+		block.SelfAttn.HeadDim = cfg.HeadDim
+		block.SelfAttn.Scale = float32(1.0 / math.Sqrt(float64(cfg.HeadDim)))
+		block.SelfAttn.PartialRotary = cfg.PartialRotaryFactor
+		block.SelfAttn.RopeTheta = cfg.RopeTheta
+		block.SelfAttn.MRoPESection = cfg.MRoPESection
+
+		// Set norm eps
+		block.InputLayerNorm.Eps = cfg.RMSNormEps
+		block.PostSelfAttnNorm.Eps = cfg.RMSNormEps
+		block.PostAttnLayerNorm.Eps = cfg.RMSNormEps
+		block.PostMLPLayerNorm.Eps = cfg.RMSNormEps
+	}
+	m.FinalNorm.Eps = cfg.RMSNormEps
+}
+
+// Generate autoregressively generates visual tokens with KV caching
+func (m *VisionLanguageEncoder) Generate(
+	prompt string,
+	tok *GLMTokenizer,
+	maxTokens int32,
+	temperature float32,
+	topP float32,
+	seed int64,
+	targetHeight, targetWidth int32,
+	progressFn func(int),
+) *mlx.Array {
+	cfg := m.Config
+
+	// Encode prompt with grid tokens using GLM tokenizer
+	// Format: {prompt}<sop>{h} {w}<eop><sop>{prev_h} {prev_w}<eop><|dit_token_16384|>
+	tokens := tok.EncodeForGeneration(prompt, targetHeight, targetWidth)
+
+	// Calculate grid dimensions for MRoPE position IDs
+	factor := int32(32)
+	tokenH := targetHeight / factor
+	tokenW := targetWidth / factor
+	ratio := float64(tokenH) / float64(tokenW)
+	prevTokenH := int32(math.Sqrt(ratio) * 16)
+	prevTokenW := int32(math.Sqrt(1.0/ratio) * 16)
+	prevGridSize := prevTokenH * prevTokenW
+
+	// Create KV cache for all layers
+	cache := NewARCache(cfg.NumHiddenLayers)
+	defer cache.Free()
+
+	// ===== PREFILL PHASE =====
+	// Process entire prompt at once, populate cache
+	promptLen := int32(len(tokens))
+	tokenArr := mlx.NewArrayInt32(tokens, []int32{1, promptLen})
+	h := m.EmbedTokens.Forward(tokenArr)
+	tokenArr.Free()
+
+	mlx.Eval(h)
+
+	// Compute position IDs for prefill (text tokens use same position for all dims)
+	prefillPositions := make([][]int32, 3)
+	for dim := 0; dim < 3; dim++ {
+		prefillPositions[dim] = make([]int32, promptLen)
+		for i := int32(0); i < promptLen; i++ {
+			prefillPositions[dim][i] = i
+		}
+	}
+
+	// Forward through layers (prefill)
+	for i, layer := range m.Layers {
+		oldH := h
+		h = layer.ForwardWithCache(h, promptLen, 0, cfg.RMSNormEps, cache.Layers[i], prefillPositions)
+		if i > 0 {
+			oldH.Free()
+		}
+	}
+	// Eval h and cache arrays together so cache is materialized
+	evalArgs := []*mlx.Array{h}
+	for _, lc := range cache.Layers {
+		evalArgs = append(evalArgs, lc.State()...)
+	}
+	mlx.Eval(evalArgs...)
+
+	// Final norm and get logits for last position
+	preNormH := h
+	h = m.FinalNorm.Forward(h, cfg.RMSNormEps)
+	preNormH.Free()
+
+	lastH := mlx.Slice(h, []int32{0, promptLen - 1, 0}, []int32{1, promptLen, cfg.HiddenSize})
+	h.Free()
+	lastH = mlx.Reshape(lastH, 1, cfg.HiddenSize)
+	logits := mlx.Matmul(lastH, mlx.Transpose(m.LMHead, 1, 0))
+	lastH.Free()
+
+	// Sample first token
+	var sampleCounter int64 = 0
+	nextToken := sampleVisualToken(logits, temperature, topP, cfg, seed, &sampleCounter)
+	logits.Free()
+
+	// AR generation loop with caching
+	// Visual tokens are stored as VQ codebook indices [0, 16383]
+	// The LM head outputs indices [0, 16511] where:
+	// - [0, 16383] are VQ codes
+	// - 16384 is BOS
+	// - 16385 is EOS
+	visualTokens := make([]int32, 0, maxTokens)
+	posOffset := promptLen
+	visualTokenIdx := int32(0) // Index within visual token sequence for grid position calculation
+
+	// Preallocate slice for old cache state to reuse
+	oldCacheState := make([]*mlx.Array, 0, len(m.Layers)*2)
+
+	for i := int32(0); i < maxTokens; i++ {
+		if progressFn != nil {
+			progressFn(int(i))
+		}
+
+		// Check for end token (EOS = 16385)
+		if nextToken == cfg.ImageEndTokenID {
+			break
+		}
+
+		// Skip BOS token (16384), only store actual VQ codes [0, 16383]
+		if nextToken == cfg.ImageStartTokenID {
+			// BOS token - skip storing but continue generation
+		} else if nextToken < cfg.ImageStartTokenID {
+			// This is an actual VQ code [0, 16383] - store it
+			visualTokens = append(visualTokens, nextToken)
+		}
+		// Tokens >= 16386 are other special tokens, skip them
+
+		// ===== DECODE PHASE =====
+		// Save old cache state before forward (to free after eval)
+		oldCacheState = oldCacheState[:0]
+		for _, lc := range cache.Layers {
+			oldCacheState = append(oldCacheState, lc.State()...)
+		}
+
+		// Only process the new token, use cached K,V
+		tokenArr := mlx.NewArrayInt32([]int32{nextToken}, []int32{1, 1})
+		h := m.EmbedTokens.Forward(tokenArr)
+		tokenArr.Free()
+
+		// Compute MRoPE position IDs for this visual token
+		// Visual tokens are arranged in two grids: prev grid then target grid
+		// Position dimensions: [temporal, height, width]
+		decodePositions := computeVisualTokenPositions(
+			visualTokenIdx, posOffset, promptLen,
+			prevTokenH, prevTokenW, prevGridSize,
+			tokenH, tokenW,
+		)
+
+		// Forward through layers (decode with cache)
+		for j, layer := range m.Layers {
+			oldH := h
+			h = layer.ForwardWithCache(h, 1, posOffset, cfg.RMSNormEps, cache.Layers[j], decodePositions)
+			if j > 0 { // Don't free the embedding on first layer
+				oldH.Free()
+			}
+		}
+
+		// Eval h and new cache state
+		newCacheState := make([]*mlx.Array, 0, len(m.Layers)*2)
+		for _, lc := range cache.Layers {
+			newCacheState = append(newCacheState, lc.State()...)
+		}
+		mlx.Eval(append([]*mlx.Array{h}, newCacheState...)...)
+
+		// Free old cache state (now that new state is evaluated)
+		for _, arr := range oldCacheState {
+			if arr != nil {
+				arr.Free()
+			}
+		}
+
+		// Final norm
+		preNormH := h
+		h = m.FinalNorm.Forward(h, cfg.RMSNormEps)
+		preNormH.Free()
+
+		// Get logits (h is already [1, 1, hidden_size])
+		h = mlx.Reshape(h, 1, cfg.HiddenSize)
+		logits := mlx.Matmul(h, mlx.Transpose(m.LMHead, 1, 0))
+		h.Free()
+
+		// Sample next token
+		nextToken = sampleVisualToken(logits, temperature, topP, cfg, seed, &sampleCounter)
+		logits.Free()
+
+		posOffset++
+		visualTokenIdx++
+
+		// Periodically clear cache to release intermediate memory
+		if i%256 == 0 {
+			mlx.ClearCache()
+		}
+	}
+
+	if len(visualTokens) == 0 {
+		// Return at least one token to avoid empty tensor issues
+		visualTokens = append(visualTokens, 0)
+	}
+
+	return mlx.NewArrayInt32(visualTokens, []int32{1, int32(len(visualTokens))})
+}
+
+// computeVisualTokenPositions computes MRoPE position IDs for a visual token
+// Returns [3][1] position IDs for temporal, height, and width dimensions
+//
+// MRoPE position encoding for GLM-Image visual tokens:
+// - temporal: CONSTANT within each grid (= decode_pos at grid start)
+// - height: decode_pos + row index within grid
+// - width: decode_pos + column index within grid
+//
+// Between grids, decode_pos advances by max(grid_h, grid_w) to ensure
+// sufficient positional separation.
+func computeVisualTokenPositions(
+	visualIdx int32, absPos int32, promptLen int32,
+	prevH, prevW, prevSize int32,
+	targetH, targetW int32,
+) [][]int32 {
+	positions := make([][]int32, 3)
+	for dim := 0; dim < 3; dim++ {
+		positions[dim] = make([]int32, 1)
+	}
+
+	// First grid (prev grid) starts at decode_pos = promptLen
+	prevGridDecodePos := promptLen
+
+	// Second grid (target grid) starts after first grid
+	// next_pos = prev_decode_pos + max(prevH, prevW)
+	maxPrev := prevH
+	if prevW > maxPrev {
+		maxPrev = prevW
+	}
+	targetGridDecodePos := prevGridDecodePos + maxPrev
+
+	// Compute position IDs based on which grid the token is in
+	if visualIdx < prevSize {
+		// Token is in the prev grid (prev_token_h × prev_token_w)
+		row := visualIdx / prevW
+		col := visualIdx % prevW
+
+		// temporal is CONSTANT for all tokens in this grid
+		positions[0][0] = prevGridDecodePos
+		// height and width are relative to grid's decode_pos
+		positions[1][0] = prevGridDecodePos + row
+		positions[2][0] = prevGridDecodePos + col
+	} else {
+		// Token is in the target grid (token_h × token_w)
+		targetIdx := visualIdx - prevSize
+		row := targetIdx / targetW
+		col := targetIdx % targetW
+
+		// temporal is CONSTANT for all tokens in this grid
+		positions[0][0] = targetGridDecodePos
+		// height and width are relative to grid's decode_pos
+		positions[1][0] = targetGridDecodePos + row
+		positions[2][0] = targetGridDecodePos + col
+	}
+
+	_ = targetH // Used for documentation clarity
+	_ = absPos  // No longer used - kept for API compatibility
+	return positions
+}
+
+// sampleVisualToken samples from the visual vocabulary using top-p (nucleus) sampling
+// Note: For GLM-Image, greedy decoding is not allowed as it may cause repetitive outputs
+// Returns a visual token ID in range [0, 16511] which directly indexes into the embedding table
+// sampleCounter is incremented for each call to ensure different random values
+func sampleVisualToken(logits *mlx.Array, temperature float32, topP float32, cfg *VisionLanguageConfig, seed int64, sampleCounter *int64) int32 {
+	// The LMHead outputs logits for visual tokens only (shape [1, 16512])
+	// Output index directly corresponds to vocab ID [0, 16511]
+	// No offset needed - the visual tokens are at vocab IDs [0, 16511]
+	visualLogits := logits
+
+	// Apply temperature
+	if temperature != 1.0 && temperature > 0 {
+		visualLogits = mlx.DivScalar(visualLogits, temperature)
+	}
+
+	// Apply softmax to get probabilities
+	probs := mlx.Softmax(visualLogits, -1)
+	mlx.Eval(probs)
+
+	// Get the sampled index using top-p sampling
+	// This directly gives us the vocab ID in [0, 16511]
+	// Special tokens: 16384 = BOS, 16385 = EOS
+	// Use seed + counter for reproducible but different random values
+	effectiveSeed := seed + *sampleCounter
+	*sampleCounter++
+	return sampleTopP(probs, topP, effectiveSeed)
+}
+
+// sampleTopP implements nucleus (top-p) sampling
+// probs: [1, vocab_size] probability distribution
+// topP: cumulative probability threshold (e.g., 0.75)
+// seed: random seed for reproducible sampling
+func sampleTopP(probs *mlx.Array, topP float32, seed int64) int32 {
+	// Negate probs for descending sort (Argsort only does ascending)
+	negProbs := mlx.MulScalar(probs, -1)
+	sortedIndices := mlx.Argsort(negProbs, -1)
+	sortedProbs := mlx.TakeAlongAxis(probs, sortedIndices, -1)
+	cumProbs := mlx.Cumsum(sortedProbs, -1)
+	mlx.Eval(sortedIndices, sortedProbs, cumProbs)
+
+	// Find cutoff index where cumulative probability exceeds topP
+	probsData := sortedProbs.Data()
+	cumProbsData := cumProbs.Data()
+	indicesData := sortedIndices.DataInt32()
+
+	// Calculate cutoff and renormalize
+	var cutoffIdx int
+	var totalProb float32
+	for i, cp := range cumProbsData {
+		totalProb += probsData[i]
+		if cp >= topP {
+			cutoffIdx = i + 1 // Include this token
+			break
+		}
+	}
+	if cutoffIdx == 0 {
+		cutoffIdx = len(probsData) // Use all tokens if topP is very high
+	}
+
+	// Sample from the truncated distribution
+	// Renormalize the truncated probabilities
+	truncatedProbs := make([]float32, cutoffIdx)
+	for i := 0; i < cutoffIdx; i++ {
+		truncatedProbs[i] = probsData[i] / totalProb
+	}
+
+	// Sample using random number with provided seed for reproducibility
+	r := mlx.RandomUniform([]int32{1}, uint64(seed))
+	mlx.Eval(r)
+	randVal := r.Data()[0]
+
+	// Find the sampled token
+	var cumulative float32
+	for i := 0; i < cutoffIdx; i++ {
+		cumulative += truncatedProbs[i]
+		if randVal < cumulative {
+			return indicesData[i]
+		}
+	}
+
+	// Fallback to the last token in truncated set
+	return indicesData[cutoffIdx-1]
+}
+
+// Forward for GLMBlock
+func (b *GLMBlock) Forward(x *mlx.Array, seqLen int32, eps float32) *mlx.Array {
+	return b.ForwardWithCache(x, seqLen, 0, eps, nil, nil)
+}
+
+// ForwardWithCache performs block forward with optional KV caching and MRoPE
+// positionIDs: [3][L] - position indices for MRoPE (nil = use sequential positions)
+func (b *GLMBlock) ForwardWithCache(x *mlx.Array, seqLen int32, posOffset int32, eps float32, kvcache cache.Cache, positionIDs [][]int32) *mlx.Array {
+	// Pre-attention norm
+	normed := b.InputLayerNorm.Forward(x, eps)
+
+	// Self-attention with RoPE/MRoPE and cache
+	attnOut := b.SelfAttn.ForwardWithCache(normed, seqLen, posOffset, kvcache, positionIDs)
+
+	// Post-attention norm (GLM-4 style)
+	attnOut = b.PostSelfAttnNorm.Forward(attnOut, eps)
+
+	// Residual connection
+	x = mlx.Add(x, attnOut)
+
+	// Post-attention layer norm
+	normed = b.PostAttnLayerNorm.Forward(x, eps)
+
+	// MLP
+	mlpOut := b.MLP.Forward(normed)
+
+	// Post-MLP norm
+	mlpOut = b.PostMLPLayerNorm.Forward(mlpOut, eps)
+
+	// Residual connection
+	x = mlx.Add(x, mlpOut)
+
+	return x
+}
+
+// Forward for GLMAttention (without cache - used for prefill)
+func (attn *GLMAttention) Forward(x *mlx.Array, seqLen int32) *mlx.Array {
+	return attn.ForwardWithCache(x, seqLen, 0, nil, nil)
+}
+
+// ForwardWithCache performs attention with optional KV caching and MRoPE
+// posOffset is the position offset for RoPE (0 for prefill, cached_len for decode)
+// positionIDs: [3][L] - if nil, uses sequential positions for all dims (text mode)
+// kvcache is updated in-place if provided
+func (attn *GLMAttention) ForwardWithCache(x *mlx.Array, seqLen int32, posOffset int32, kvcache cache.Cache, positionIDs [][]int32) *mlx.Array {
+	shape := x.Shape()
+	B := shape[0]
+	L := shape[1]
+
+	// Q, K, V projections
+	q := mlx.Matmul(x, mlx.Transpose(attn.QProj, 1, 0))
+	k := mlx.Matmul(x, mlx.Transpose(attn.KProj, 1, 0))
+	v := mlx.Matmul(x, mlx.Transpose(attn.VProj, 1, 0))
+
+	// Add biases
+	if attn.QBias != nil {
+		q = mlx.Add(q, attn.QBias)
+	}
+	if attn.KBias != nil {
+		k = mlx.Add(k, attn.KBias)
+	}
+	if attn.VBias != nil {
+		v = mlx.Add(v, attn.VBias)
+	}
+
+	// Reshape to [B, L, nheads, head_dim]
+	q = mlx.Reshape(q, B, L, attn.NHeads, attn.HeadDim)
+	k = mlx.Reshape(k, B, L, attn.NKVHeads, attn.HeadDim)
+	v = mlx.Reshape(v, B, L, attn.NKVHeads, attn.HeadDim)
+
+	// Apply partial RoPE or MRoPE
+	rotaryDim := int32(float32(attn.HeadDim) * attn.PartialRotary)
+	if len(attn.MRoPESection) == 3 && positionIDs != nil {
+		// Use MRoPE with explicit position IDs
+		q = applyPartialMRoPE(q, positionIDs, rotaryDim, attn.RopeTheta, attn.MRoPESection)
+		k = applyPartialMRoPE(k, positionIDs, rotaryDim, attn.RopeTheta, attn.MRoPESection)
+	} else if len(attn.MRoPESection) == 3 {
+		// Use MRoPE with sequential positions (same for all dims - text mode)
+		seqPositions := make([][]int32, 3)
+		for dim := 0; dim < 3; dim++ {
+			seqPositions[dim] = make([]int32, L)
+			for i := int32(0); i < L; i++ {
+				seqPositions[dim][i] = i + posOffset
+			}
+		}
+		q = applyPartialMRoPE(q, seqPositions, rotaryDim, attn.RopeTheta, attn.MRoPESection)
+		k = applyPartialMRoPE(k, seqPositions, rotaryDim, attn.RopeTheta, attn.MRoPESection)
+	} else {
+		// Fallback to standard RoPE
+		q = applyPartialRoPEWithOffset(q, L, posOffset, rotaryDim, attn.RopeTheta)
+		k = applyPartialRoPEWithOffset(k, L, posOffset, rotaryDim, attn.RopeTheta)
+	}
+
+	// Transpose to [B, nheads, L, head_dim]
+	q = mlx.Transpose(q, 0, 2, 1, 3)
+	k = mlx.Transpose(k, 0, 2, 1, 3)
+	v = mlx.Transpose(v, 0, 2, 1, 3)
+
+	// Update cache and get full K, V for attention
+	if kvcache != nil {
+		k, v = kvcache.Update(k, v, int(L))
+	}
+
+	// Repeat KV for GQA
+	kExpanded := k
+	vExpanded := v
+	if attn.NKVHeads < attn.NHeads {
+		repeats := attn.NHeads / attn.NKVHeads
+		kExpanded = repeatKV(k, repeats)
+		vExpanded = repeatKV(v, repeats)
+	}
+
+	// Scaled dot-product attention with causal mask
+	out := mlx.ScaledDotProductAttention(q, kExpanded, vExpanded, attn.Scale, true)
+
+	// Transpose back [B, nheads, L, head_dim] -> [B, L, nheads, head_dim]
+	out = mlx.Transpose(out, 0, 2, 1, 3)
+	// Reshape to [B, L, hidden_size]
+	out = mlx.Reshape(out, B, L, attn.NHeads*attn.HeadDim)
+
+	// Output projection
+	out = mlx.Matmul(out, mlx.Transpose(attn.OProj, 1, 0))
+
+	return out
+}
+
+// applyPartialRoPE applies RoPE to only the first rotaryDim dimensions
+func applyPartialRoPE(x *mlx.Array, seqLen int32, rotaryDim int32, theta float32) *mlx.Array {
+	return applyPartialRoPEWithOffset(x, seqLen, 0, rotaryDim, theta)
+}
+
+// applyPartialRoPEWithOffset applies RoPE with a position offset
+func applyPartialRoPEWithOffset(x *mlx.Array, seqLen int32, posOffset int32, rotaryDim int32, theta float32) *mlx.Array {
+	shape := x.Shape()
+	B := shape[0]
+	L := shape[1]
+	H := shape[2]
+	D := shape[3]
+
+	if rotaryDim <= 0 || rotaryDim > D {
+		rotaryDim = D
+	}
+
+	// Split into rotary and pass-through parts
+	xRot := mlx.Slice(x, []int32{0, 0, 0, 0}, []int32{B, L, H, rotaryDim})
+	xPass := mlx.Slice(x, []int32{0, 0, 0, rotaryDim}, []int32{B, L, H, D})
+
+	// Apply RoPE to rotary part with position offset
+	xRot = applyRoPEWithOffset(xRot, L, posOffset, theta)
+
+	// Concatenate back
+	return mlx.Concatenate([]*mlx.Array{xRot, xPass}, 3)
+}
+
+// applyPartialMRoPE applies Multi-dimensional RoPE (MRoPE) to the first rotaryDim dimensions
+// positionIDs: [3, L] - position indices for each dimension (temporal, height, width)
+// mrope_section: [8, 12, 12] - frequency pairs per dimension
+// For text tokens: all 3 dimensions have the same sequential position
+// For image tokens: temporal=seq_idx, height=row, width=col
+func applyPartialMRoPE(x *mlx.Array, positionIDs [][]int32, rotaryDim int32, theta float32, mropeSection []int32) *mlx.Array {
+	shape := x.Shape()
+	B := shape[0]
+	L := shape[1]
+	H := shape[2]
+	D := shape[3]
+
+	if rotaryDim <= 0 || rotaryDim > D {
+		rotaryDim = D
+	}
+
+	// Split into rotary and pass-through parts
+	xRot := mlx.Slice(x, []int32{0, 0, 0, 0}, []int32{B, L, H, rotaryDim})
+	xPass := mlx.Slice(x, []int32{0, 0, 0, rotaryDim}, []int32{B, L, H, D})
+
+	// Apply MRoPE to rotary part
+	xRot = applyMRoPE(xRot, positionIDs, theta, mropeSection)
+
+	// Concatenate back
+	return mlx.Concatenate([]*mlx.Array{xRot, xPass}, 3)
+}
+
+// applyMRoPE applies multi-dimensional rotary position embedding
+// x: [B, L, H, D] where D is the rotary dimension
+// positionIDs: [3][L] - positions for temporal, height, width dimensions
+// mropeSection: [8, 12, 12] - frequency pairs per dimension
+func applyMRoPE(x *mlx.Array, positionIDs [][]int32, theta float32, mropeSection []int32) *mlx.Array {
+	shape := x.Shape()
+	B := shape[0]
+	L := shape[1]
+	H := shape[2]
+	D := shape[3]
+	half := D / 2
+
+	// Validate mrope_section sums to half (number of frequency pairs)
+	var totalPairs int32
+	for _, s := range mropeSection {
+		totalPairs += s
+	}
+	if totalPairs != half {
+		// Fallback to standard RoPE if section doesn't match
+		return applyRoPEWithOffset(x, L, 0, theta)
+	}
+
+	// Build angles for each position dimension (matching Python's MRoPE approach)
+	// Python: compute freqs for all dims, then apply_mrope selects freq ranges, then duplicate
+	// Order: [temporal_8, height_12, width_12] -> duplicate -> [t8, h12, w12, t8, h12, w12]
+	angleVals := make([]*mlx.Array, 3)
+
+	freqOffset := int32(0)
+	for dim := 0; dim < 3; dim++ {
+		numPairs := mropeSection[dim]
+		if numPairs == 0 {
+			continue
+		}
+
+		// Compute inverse frequencies for this section
+		// Each dimension uses DIFFERENT frequency ranges:
+		// - Temporal: frequencies 0 to section[0]-1
+		// - Height: frequencies section[0] to section[0]+section[1]-1
+		// - Width: frequencies section[0]+section[1] to sum(section)-1
+		freqsArr := make([]float32, numPairs)
+		for i := int32(0); i < numPairs; i++ {
+			globalIdx := freqOffset + i
+			freqsArr[i] = float32(1.0 / math.Pow(float64(theta), float64(2*globalIdx)/float64(D)))
+		}
+		freqs := mlx.NewArray(freqsArr, []int32{numPairs})
+
+		// Position indices for this dimension
+		posArr := make([]float32, L)
+		for i := int32(0); i < L; i++ {
+			posArr[i] = float32(positionIDs[dim][i])
+		}
+		pos := mlx.NewArray(posArr, []int32{L})
+
+		// Compute angles: [L, numPairs] = outer(pos, freqs)
+		posExpanded := mlx.Reshape(pos, L, 1)
+		freqsExpanded := mlx.Reshape(freqs, 1, numPairs)
+		angleVals[dim] = mlx.Mul(posExpanded, freqsExpanded)
+
+		freqOffset += numPairs
+	}
+
+	// Concatenate all sections: [L, half] = [L, 32]
+	allAngles := mlx.Concatenate(angleVals, 1)
+
+	// Duplicate AFTER concatenation: [L, D] = [L, 64]
+	// This gives: [temporal_8, height_12, width_12, temporal_8, height_12, width_12]
+	allAngles = mlx.Concatenate([]*mlx.Array{allAngles, allAngles}, 1)
+
+	// Compute cos/sin
+	allCos := mlx.Cos(allAngles)
+	allSin := mlx.Sin(allAngles)
+
+	// Reshape for broadcasting: [1, L, 1, D] to match x [B, L, H, D]
+	allCos = mlx.Reshape(allCos, 1, L, 1, D)
+	allSin = mlx.Reshape(allSin, 1, L, 1, D)
+
+	// x_rotated = cat([-x_imag, x_real], dim=-1)
+	x1 := mlx.Slice(x, []int32{0, 0, 0, 0}, []int32{B, L, H, half})  // x_real
+	x2 := mlx.Slice(x, []int32{0, 0, 0, half}, []int32{B, L, H, D})  // x_imag
+	x2Neg := mlx.MulScalar(x2, -1)                                   // -x_imag
+	xRotated := mlx.Concatenate([]*mlx.Array{x2Neg, x1}, 3)          // [-x_imag, x_real]
+
+	// out = x * cos + x_rotated * sin
+	return mlx.Add(mlx.Mul(x, allCos), mlx.Mul(xRotated, allSin))
+}
+
+// applyRoPE applies rotary position embedding
+func applyRoPE(x *mlx.Array, seqLen int32, theta float32) *mlx.Array {
+	return applyRoPEWithOffset(x, seqLen, 0, theta)
+}
+
+// applyRoPEWithOffset applies rotary position embedding with position offset
+// Uses the split-half approach (matches diffusers GLM-Image with use_real_unbind_dim=-2)
+func applyRoPEWithOffset(x *mlx.Array, seqLen int32, posOffset int32, theta float32) *mlx.Array {
+	shape := x.Shape()
+	B := shape[0]
+	L := shape[1]
+	H := shape[2]
+	D := shape[3]
+	half := D / 2
+
+	// Compute inverse frequencies: 1 / (theta^(2i/d))
+	freqsArr := make([]float32, half)
+	for i := int32(0); i < half; i++ {
+		freqsArr[i] = float32(1.0 / math.Pow(float64(theta), float64(2*i)/float64(D)))
+	}
+	freqs := mlx.NewArray(freqsArr, []int32{half})
+
+	// Position indices with offset
+	posArr := make([]float32, L)
+	for i := int32(0); i < L; i++ {
+		posArr[i] = float32(i + posOffset)
+	}
+	pos := mlx.NewArray(posArr, []int32{L})
+
+	// Compute angles: [L, half] = outer(pos, freqs)
+	posExpanded := mlx.Reshape(pos, L, 1)
+	freqsExpanded := mlx.Reshape(freqs, 1, half)
+	angles := mlx.Mul(posExpanded, freqsExpanded)
+
+	// Duplicate angles to match diffusers: cat([angles, angles], dim=-1) -> [L, D]
+	anglesDup := mlx.Concatenate([]*mlx.Array{angles, angles}, 1)
+
+	// Cos and sin: [L, 1, D] for broadcasting to [B, L, H, D]
+	cosVals := mlx.Cos(anglesDup)
+	sinVals := mlx.Sin(anglesDup)
+	cosVals = mlx.Reshape(cosVals, L, 1, D)
+	sinVals = mlx.Reshape(sinVals, L, 1, D)
+
+	// x_rotated = cat([-x_imag, x_real], dim=-1) where x_real=x[..., :half], x_imag=x[..., half:]
+	x1 := mlx.Slice(x, []int32{0, 0, 0, 0}, []int32{B, L, H, half})      // x_real
+	x2 := mlx.Slice(x, []int32{0, 0, 0, half}, []int32{B, L, H, D})      // x_imag
+	x2Neg := mlx.MulScalar(x2, -1)                                       // -x_imag
+	xRotated := mlx.Concatenate([]*mlx.Array{x2Neg, x1}, 3)              // [-x_imag, x_real]
+
+	// out = x * cos + x_rotated * sin
+	return mlx.Add(mlx.Mul(x, cosVals), mlx.Mul(xRotated, sinVals))
+}
+
+// repeatKV repeats key/value heads for GQA
+func repeatKV(x *mlx.Array, repeats int32) *mlx.Array {
+	if repeats == 1 {
+		return x
+	}
+	shape := x.Shape()
+	// x: [B, nkvheads, L, head_dim]
+	x = mlx.ExpandDims(x, 2)
+	// x: [B, nkvheads, 1, L, head_dim]
+	x = mlx.Tile(x, []int32{1, 1, repeats, 1, 1})
+	// x: [B, nkvheads, repeats, L, head_dim]
+	return mlx.Reshape(x, shape[0], shape[1]*repeats, shape[2], shape[3])
+}
+
+// Forward for GLMMLP (fused gate_up SwiGLU)
+func (m *GLMMLP) Forward(x *mlx.Array) *mlx.Array {
+	// gate_up_proj outputs [gate, up] concatenated
+	gateUp := mlx.Matmul(x, mlx.Transpose(m.GateUpProj, 1, 0))
+
+	shape := gateUp.Shape()
+	halfDim := shape[len(shape)-1] / 2
+
+	// Split into gate and up
+	gate := mlx.Slice(gateUp, []int32{0, 0, 0}, []int32{shape[0], shape[1], halfDim})
+	up := mlx.Slice(gateUp, []int32{0, 0, halfDim}, []int32{shape[0], shape[1], shape[2]})
+
+	// SwiGLU: silu(gate) * up
+	gate = mlx.SiLU(gate)
+	h := mlx.Mul(gate, up)
+
+	// Down projection
+	return mlx.Matmul(h, mlx.Transpose(m.DownProj, 1, 0))
+}

x/imagegen/models/qwen_image/qwen_image.go

@@ -222,6 +222,14 @@ func (m *Model) generate(cfg *GenerateConfig) (*mlx.Array, error) {
 		mlx.Keep(posEmb, negEmb)
 	}
 
+	// Pre-compute batched embeddings for CFG (single forward pass optimization)
+	var batchedEmb *mlx.Array
+	if useCFG {
+		batchedEmb = mlx.Concatenate([]*mlx.Array{posEmb, negEmb}, 0)
+		mlx.Keep(batchedEmb)
+		mlx.Eval(batchedEmb)
+	}
+
 	// Scheduler
 	scheduler := NewFlowMatchScheduler(DefaultSchedulerConfig())
 	scheduler.SetTimesteps(cfg.Steps, imgSeqLen)
@@ -264,10 +272,19 @@ func (m *Model) generate(cfg *GenerateConfig) (*mlx.Array, error) {
 
 		var output *mlx.Array
 		if useCFG {
-			// True CFG: run twice and combine with norm rescaling
+			// CFG Batching: single forward pass with batch=2
 			// Note: layer caching with CFG is not supported yet (would need 2 caches)
-			posOutput := m.Transformer.Forward(patches, posEmb, timestep, ropeCache.ImgFreqs, ropeCache.TxtFreqs)
-			negOutput := m.Transformer.Forward(patches, negEmb, timestep, ropeCache.ImgFreqs, ropeCache.TxtFreqs)
+			batchedPatches := mlx.Tile(patches, []int32{2, 1, 1})
+			batchedTimestep := mlx.Tile(timestep, []int32{2})
+
+			// Single batched forward pass
+			batchedOutput := m.Transformer.Forward(batchedPatches, batchedEmb, batchedTimestep, ropeCache.ImgFreqs, ropeCache.TxtFreqs)
+
+			// Split output: [2, L, D] -> pos [1, L, D], neg [1, L, D]
+			L := batchedOutput.Shape()[1]
+			D := batchedOutput.Shape()[2]
+			posOutput := mlx.Slice(batchedOutput, []int32{0, 0, 0}, []int32{1, L, D})
+			negOutput := mlx.Slice(batchedOutput, []int32{1, 0, 0}, []int32{2, L, D})
 
 			diff := mlx.Sub(posOutput, negOutput)
 			scaledDiff := mlx.MulScalar(diff, cfg.CFGScale)
@@ -305,6 +322,9 @@ func (m *Model) generate(cfg *GenerateConfig) (*mlx.Array, error) {
 	if negEmb != nil {
 		negEmb.Free()
 	}
+	if batchedEmb != nil {
+		batchedEmb.Free()
+	}
 	ropeCache.ImgFreqs.Free()
 	ropeCache.TxtFreqs.Free()
 	if stepCache != nil {

x/imagegen/models/qwen_image_edit/qwen_image_edit.go

@@ -241,6 +241,14 @@ func (m *Model) edit(inputImagePaths []string, cfg *GenerateConfig) (*mlx.Array,
 		mlx.Eval(posEmb, negEmb)
 	}
 
+	// Pre-compute batched embeddings for CFG (single forward pass optimization)
+	var batchedEmb *mlx.Array
+	if useCFG {
+		batchedEmb = mlx.Concatenate([]*mlx.Array{posEmb, negEmb}, 0)
+		mlx.Keep(batchedEmb)
+		mlx.Eval(batchedEmb)
+	}
+
 	// Encode all input images to latents and concatenate
 	fmt.Println("Encoding images to latents...")
 	allImageLatentsPacked := make([]*mlx.Array, len(vaeImages))
@@ -291,11 +299,18 @@ func (m *Model) edit(inputImagePaths []string, cfg *GenerateConfig) (*mlx.Array,
 
 		var output *mlx.Array
 		if useCFG {
-			posOutput := m.Transformer.Forward(latentInput, posEmb, timestep, ropeCache.ImgFreqs, ropeCache.TxtFreqs)
-			negOutput := m.Transformer.Forward(latentInput, negEmb, timestep, ropeCache.ImgFreqs, ropeCache.TxtFreqs)
+			// CFG Batching: single forward pass with batch=2
+			// Tile inputs: [1, L, D] -> [2, L, D]
+			batchedLatentInput := mlx.Tile(latentInput, []int32{2, 1, 1})
+			batchedTimestep := mlx.Tile(timestep, []int32{2})
 
-			posOutput = mlx.Slice(posOutput, []int32{0, 0, 0}, []int32{1, imgSeqLen, posOutput.Shape()[2]})
-			negOutput = mlx.Slice(negOutput, []int32{0, 0, 0}, []int32{1, imgSeqLen, negOutput.Shape()[2]})
+			// Single batched forward pass
+			batchedOutput := m.Transformer.Forward(batchedLatentInput, batchedEmb, batchedTimestep, ropeCache.ImgFreqs, ropeCache.TxtFreqs)
+
+			// Split output: [2, L, D] -> pos [1, L, D], neg [1, L, D]
+			D := batchedOutput.Shape()[2]
+			posOutput := mlx.Slice(batchedOutput, []int32{0, 0, 0}, []int32{1, imgSeqLen, D})
+			negOutput := mlx.Slice(batchedOutput, []int32{1, 0, 0}, []int32{2, imgSeqLen, D})
 
 			output = applyCFGWithNormRescale(posOutput, negOutput, cfg.CFGScale)
 		} else {
@@ -317,6 +332,9 @@ func (m *Model) edit(inputImagePaths []string, cfg *GenerateConfig) (*mlx.Array,
 	if negEmb != nil {
 		negEmb.Free()
 	}
+	if batchedEmb != nil {
+		batchedEmb.Free()
+	}
 	ropeCache.ImgFreqs.Free()
 	ropeCache.TxtFreqs.Free()
 	imageLatentsPacked.Free()

x/imagegen/models/zimage/text_encoder.go

@@ -28,12 +28,12 @@ type Qwen3Config struct {
 
 // Qwen3Attention implements Qwen3 attention with QK norms
 type Qwen3Attention struct {
-	QProj *nn.Linear  `weight:"q_proj"`
-	KProj *nn.Linear  `weight:"k_proj"`
-	VProj *nn.Linear  `weight:"v_proj"`
-	OProj *nn.Linear  `weight:"o_proj"`
-	QNorm *nn.RMSNorm `weight:"q_norm"`
-	KNorm *nn.RMSNorm `weight:"k_norm"`
+	QProj nn.LinearLayer `weight:"q_proj"`
+	KProj nn.LinearLayer `weight:"k_proj"`
+	VProj nn.LinearLayer `weight:"v_proj"`
+	OProj nn.LinearLayer `weight:"o_proj"`
+	QNorm *nn.RMSNorm    `weight:"q_norm"`
+	KNorm *nn.RMSNorm    `weight:"k_norm"`
 	// Computed fields
 	NHeads    int32
 	NKVHeads  int32
@@ -136,9 +136,9 @@ func repeatKV(x *mlx.Array, repeats int32) *mlx.Array {
 
 // Qwen3MLP implements Qwen3 SwiGLU MLP
 type Qwen3MLP struct {
-	GateProj *nn.Linear `weight:"gate_proj"`
-	UpProj   *nn.Linear `weight:"up_proj"`
-	DownProj *nn.Linear `weight:"down_proj"`
+	GateProj nn.LinearLayer `weight:"gate_proj"`
+	UpProj   nn.LinearLayer `weight:"up_proj"`
+	DownProj nn.LinearLayer `weight:"down_proj"`
 }
 
 // Forward applies the MLP

x/imagegen/models/zimage/transformer.go

@@ -36,8 +36,8 @@ type TransformerConfig struct {
 // TimestepEmbedder creates sinusoidal timestep embeddings
 // Output dimension is 256 (fixed), used for AdaLN modulation
 type TimestepEmbedder struct {
-	Linear1       *nn.Linear `weight:"mlp.0"`
-	Linear2       *nn.Linear `weight:"mlp.2"`
+	Linear1       nn.LinearLayer `weight:"mlp.0"`
+	Linear2       nn.LinearLayer `weight:"mlp.2"`
 	FreqEmbedSize int32      // 256 (computed)
 }
 
@@ -74,7 +74,7 @@ func (te *TimestepEmbedder) Forward(t *mlx.Array) *mlx.Array {
 
 // XEmbedder embeds image patches to model dimension
 type XEmbedder struct {
-	Linear *nn.Linear `weight:"2-1"`
+	Linear nn.LinearLayer `weight:"2-1"`
 }
 
 // Forward embeds patchified image latents
@@ -86,7 +86,7 @@ func (xe *XEmbedder) Forward(x *mlx.Array) *mlx.Array {
 // CapEmbedder projects caption features to model dimension
 type CapEmbedder struct {
 	Norm     *nn.RMSNorm `weight:"0"`
-	Linear   *nn.Linear  `weight:"1"`
+	Linear   nn.LinearLayer  `weight:"1"`
 	PadToken *mlx.Array  // loaded separately at root level
 }
 
@@ -100,12 +100,13 @@ func (ce *CapEmbedder) Forward(capFeats *mlx.Array) *mlx.Array {
 
 // FeedForward implements SwiGLU FFN
 type FeedForward struct {
-	W1     *nn.Linear `weight:"w1"` // gate projection
-	W2     *nn.Linear `weight:"w2"` // down projection
-	W3     *nn.Linear `weight:"w3"` // up projection
+	W1     nn.LinearLayer `weight:"w1"` // gate projection
+	W2     nn.LinearLayer `weight:"w2"` // down projection
+	W3     nn.LinearLayer `weight:"w3"` // up projection
 	OutDim int32      // computed from W2
 }
 
+
 // Forward applies SwiGLU: silu(W1(x)) * W3(x), then W2
 func (ff *FeedForward) Forward(x *mlx.Array) *mlx.Array {
 	shape := x.Shape()
@@ -115,6 +116,7 @@ func (ff *FeedForward) Forward(x *mlx.Array) *mlx.Array {
 
 	// Reshape for matmul
 	x = mlx.Reshape(x, B*L, D)
+
 	gate := ff.W1.Forward(x)
 	gate = mlx.SiLU(gate)
 	up := ff.W3.Forward(x)
@@ -126,17 +128,69 @@ func (ff *FeedForward) Forward(x *mlx.Array) *mlx.Array {
 
 // Attention implements multi-head attention with QK norm
 type Attention struct {
-	ToQ   *nn.Linear `weight:"to_q"`
-	ToK   *nn.Linear `weight:"to_k"`
-	ToV   *nn.Linear `weight:"to_v"`
-	ToOut *nn.Linear `weight:"to_out.0"`
+	ToQ   nn.LinearLayer `weight:"to_q"`
+	ToK   nn.LinearLayer `weight:"to_k"`
+	ToV   nn.LinearLayer `weight:"to_v"`
+	ToOut nn.LinearLayer `weight:"to_out.0"`
 	NormQ *mlx.Array `weight:"norm_q.weight"` // [head_dim] for per-head RMSNorm
 	NormK *mlx.Array `weight:"norm_k.weight"`
-	// Computed fields
-	NHeads  int32
-	HeadDim int32
-	Dim     int32
-	Scale   float32
+	// Fused QKV (computed at init time for efficiency, not loaded from weights)
+	ToQKV nn.LinearLayer `weight:"-"` // Fused Q+K+V projection (created by FuseQKV)
+	Fused bool       `weight:"-"` // Whether to use fused QKV path
+	// Computed fields (not loaded from weights)
+	NHeads  int32   `weight:"-"`
+	HeadDim int32   `weight:"-"`
+	Dim     int32   `weight:"-"`
+	Scale   float32 `weight:"-"`
+}
+
+// FuseQKV creates a fused QKV projection by concatenating weights.
+// This reduces 3 matmuls to 1 for a ~5-10% speedup.
+// Note: Fusion is skipped for quantized weights as it would require complex
+// dequant-concat-requant operations. The FP8 memory bandwidth savings outweigh
+// the ~5% fusion benefit.
+func (attn *Attention) FuseQKV() {
+	if attn.ToQ == nil || attn.ToK == nil || attn.ToV == nil {
+		return
+	}
+
+	// Skip fusion for quantized weights - type assert to check
+	toQ, qOk := attn.ToQ.(*nn.Linear)
+	toK, kOk := attn.ToK.(*nn.Linear)
+	toV, vOk := attn.ToV.(*nn.Linear)
+	if !qOk || !kOk || !vOk {
+		// One or more are QuantizedLinear, skip fusion
+		return
+	}
+
+	if toQ.Weight == nil || toK.Weight == nil || toV.Weight == nil {
+		return
+	}
+
+	// Concatenate weights: [dim, dim] x 3 -> [3*dim, dim]
+	// Weight shapes: ToQ.Weight [out_dim, in_dim], etc.
+	qWeight := toQ.Weight
+	kWeight := toK.Weight
+	vWeight := toV.Weight
+
+	// Concatenate along output dimension (axis 0)
+	fusedWeight := mlx.Concatenate([]*mlx.Array{qWeight, kWeight, vWeight}, 0)
+
+	// Evaluate fused weight to ensure it's materialized
+	mlx.Eval(fusedWeight)
+
+	// Create fused linear layer
+	fusedLinear := &nn.Linear{Weight: fusedWeight}
+
+	// Handle bias if present
+	if toQ.Bias != nil && toK.Bias != nil && toV.Bias != nil {
+		fusedBias := mlx.Concatenate([]*mlx.Array{toQ.Bias, toK.Bias, toV.Bias}, 0)
+		mlx.Eval(fusedBias)
+		fusedLinear.Bias = fusedBias
+	}
+
+	attn.ToQKV = fusedLinear
+	attn.Fused = true
 }
 
 // Forward computes attention
@@ -146,11 +200,24 @@ func (attn *Attention) Forward(x *mlx.Array, cos, sin *mlx.Array) *mlx.Array {
 	L := shape[1]
 	D := shape[2]
 
-	// Project Q, K, V
 	xFlat := mlx.Reshape(x, B*L, D)
-	q := attn.ToQ.Forward(xFlat)
-	k := attn.ToK.Forward(xFlat)
-	v := attn.ToV.Forward(xFlat)
+
+	var q, k, v *mlx.Array
+	if attn.Fused && attn.ToQKV != nil {
+		// Fused QKV path: single matmul then split
+		qkv := attn.ToQKV.Forward(xFlat) // [B*L, 3*dim]
+
+		// Split into Q, K, V along last dimension
+		// Each has shape [B*L, dim]
+		q = mlx.Slice(qkv, []int32{0, 0}, []int32{B * L, attn.Dim})
+		k = mlx.Slice(qkv, []int32{0, attn.Dim}, []int32{B * L, 2 * attn.Dim})
+		v = mlx.Slice(qkv, []int32{0, 2 * attn.Dim}, []int32{B * L, 3 * attn.Dim})
+	} else {
+		// Separate Q, K, V projections
+		q = attn.ToQ.Forward(xFlat)
+		k = attn.ToK.Forward(xFlat)
+		v = attn.ToV.Forward(xFlat)
+	}
 
 	// Reshape to [B, L, nheads, head_dim]
 	q = mlx.Reshape(q, B, L, attn.NHeads, attn.HeadDim)
@@ -227,7 +294,7 @@ type TransformerBlock struct {
 	AttentionNorm2 *nn.RMSNorm  `weight:"attention_norm2"`
 	FFNNorm1       *nn.RMSNorm  `weight:"ffn_norm1"`
 	FFNNorm2       *nn.RMSNorm  `weight:"ffn_norm2"`
-	AdaLN          *nn.Linear   `weight:"adaLN_modulation.0,optional"` // only if modulation
+	AdaLN          nn.LinearLayer   `weight:"adaLN_modulation.0,optional"` // only if modulation
 	// Computed fields
 	HasModulation bool
 	Dim           int32
@@ -281,8 +348,8 @@ func (tb *TransformerBlock) Forward(x *mlx.Array, adaln *mlx.Array, cos, sin *ml
 
 // FinalLayer outputs the denoised patches
 type FinalLayer struct {
-	AdaLN  *nn.Linear `weight:"adaLN_modulation.1"` // [256] -> [dim]
-	Output *nn.Linear `weight:"linear"`             // [dim] -> [out_channels]
+	AdaLN  nn.LinearLayer `weight:"adaLN_modulation.1"` // [256] -> [dim]
+	Output nn.LinearLayer `weight:"linear"`             // [dim] -> [out_channels]
 	OutDim int32      // computed from Output
 }
 
@@ -350,12 +417,11 @@ func (m *Transformer) Load(manifest *imagegen.ModelManifest) error {
 	m.ContextRefiners = make([]*TransformerBlock, cfg.NRefinerLayers)
 	m.Layers = make([]*TransformerBlock, cfg.NLayers)
 
-	// Load weights from tensor blobs with BF16 conversion
 	weights, err := imagegen.LoadWeightsFromManifest(manifest, "transformer")
 	if err != nil {
 		return fmt.Errorf("weights: %w", err)
 	}
-	if err := weights.Load(mlx.DtypeBFloat16); err != nil {
+	if err := weights.Load(0); err != nil {
 		return fmt.Errorf("load weights: %w", err)
 	}
 	defer weights.ReleaseAll()
@@ -377,7 +443,7 @@ func (m *Transformer) loadWeights(weights safetensors.WeightSource) error {
 func (m *Transformer) initComputedFields() {
 	cfg := m.TransformerConfig
 	m.TEmbed.FreqEmbedSize = 256
-	m.FinalLayer.OutDim = m.FinalLayer.Output.Weight.Shape()[0]
+	m.FinalLayer.OutDim = m.FinalLayer.Output.OutputDim()
 	m.CapEmbed.Norm.Eps = 1e-6
 
 	for _, block := range m.NoiseRefiners {
@@ -391,6 +457,20 @@ func (m *Transformer) initComputedFields() {
 	}
 }
 
+// FuseAllQKV fuses QKV projections in all attention layers for efficiency.
+// This reduces 3 matmuls to 1 per attention layer, providing ~5-10% speedup.
+func (m *Transformer) FuseAllQKV() {
+	for _, block := range m.NoiseRefiners {
+		block.Attention.FuseQKV()
+	}
+	for _, block := range m.ContextRefiners {
+		block.Attention.FuseQKV()
+	}
+	for _, block := range m.Layers {
+		block.Attention.FuseQKV()
+	}
+}
+
 // initTransformerBlock sets computed fields on a transformer block
 func initTransformerBlock(block *TransformerBlock, cfg *TransformerConfig) {
 	block.Dim = cfg.Dim
@@ -404,7 +484,7 @@ func initTransformerBlock(block *TransformerBlock, cfg *TransformerConfig) {
 	attn.Scale = float32(1.0 / math.Sqrt(float64(attn.HeadDim)))
 
 	// Init feedforward OutDim
-	block.FeedForward.OutDim = block.FeedForward.W2.Weight.Shape()[0]
+	block.FeedForward.OutDim = block.FeedForward.W2.OutputDim()
 
 	// Set eps on all RMSNorm layers
 	block.AttentionNorm1.Eps = cfg.NormEps
@@ -423,6 +503,8 @@ type RoPECache struct {
 	UnifiedSin *mlx.Array
 	ImgLen     int32
 	CapLen     int32
+	GridH      int32 // Image token grid height
+	GridW      int32 // Image token grid width
 }
 
 // PrepareRoPECache precomputes RoPE values for the given image and caption lengths.
@@ -456,6 +538,8 @@ func (m *Transformer) PrepareRoPECache(hTok, wTok, capLen int32) *RoPECache {
 		UnifiedSin: unifiedSin,
 		ImgLen:     imgLen,
 		CapLen:     capLen,
+		GridH:      hTok,
+		GridW:      wTok,
 	}
 }
 

x/imagegen/models/zimage/vae.go

@@ -104,6 +104,8 @@ func (gn *GroupNormLayer) forwardTiled(x *mlx.Array, B, H, W, C int32) *mlx.Arra
 	groupSize := C / gn.NumGroups
 
 	// Keep the input - we need it for slicing tiles later
+	// Track if we were the ones who kept it, so we can restore state after
+	wasKept := x.Kept()
 	mlx.Keep(x)
 
 	// Compute per-group mean and variance using flattened spatial dimensions
@@ -205,6 +207,10 @@ func (gn *GroupNormLayer) forwardTiled(x *mlx.Array, B, H, W, C int32) *mlx.Arra
 	}
 
 	// Clean up kept arrays
+	// Restore x's kept state - only free if we were the ones who kept it
+	if !wasKept {
+		x.Free()
+	}
 	mean.Free()
 	invStd.Free()
 	if weightGN != nil {
@@ -734,18 +740,26 @@ func (vae *VAEDecoder) Decode(latents *mlx.Array) *mlx.Array {
 	h := vae.ConvIn.Forward(z)
 	mlx.Eval(h)
 
+	prev := h
 	h = vae.MidBlock.Forward(h)
+	prev.Free()
 
 	for _, upBlock := range vae.UpBlocks {
+		prev = h
 		h = upBlock.Forward(h)
+		prev.Free()
 	}
 
-	prev := h
+	prev = h
 	h = vae.ConvNormOut.Forward(h)
 	mlx.Eval(h) // Eval after GroupNorm to avoid grid dimension issues
+	prev.Free()
+
+	prev = h
 	h = mlx.SiLU(h)
 	h = vae.ConvOut.Forward(h)
 	mlx.Eval(h)
+	prev.Free()
 
 	// VAE outputs [-1, 1], convert to [0, 1]
 	h = mlx.MulScalar(h, 0.5)
@@ -754,7 +768,6 @@ func (vae *VAEDecoder) Decode(latents *mlx.Array) *mlx.Array {
 
 	// Convert NHWC -> NCHW for output
 	h = mlx.Transpose(h, 0, 3, 1, 2)
-	prev.Free()
 	mlx.Eval(h)
 
 	return h

x/imagegen/models/zimage/zimage.go

@@ -26,10 +26,12 @@ type GenerateConfig struct {
 	Progress       ProgressFunc // Optional progress callback
 	CapturePath    string       // GPU capture path (debug)
 
-	// Layer caching options (speedup via shallow layer reuse)
-	LayerCache    bool // Enable layer caching (default: false)
-	CacheInterval int  // Refresh cache every N steps (default: 3)
-	CacheLayers   int  // Number of shallow layers to cache (default: 15)
+	// TeaCache options (timestep embedding aware caching)
+	TeaCache          bool    // TeaCache is always enabled for faster inference
+	TeaCacheThreshold float32 // Threshold for cache reuse (default: 0.1, lower = more aggressive)
+
+	// Fused QKV (fuse Q/K/V projections into single matmul)
+	FusedQKV bool // Enable fused QKV projection (default: false)
 }
 
 // ProgressFunc is called during generation with step progress.
@@ -42,6 +44,7 @@ type Model struct {
 	TextEncoder *Qwen3TextEncoder
 	Transformer *Transformer
 	VAEDecoder  *VAEDecoder
+	qkvFused    bool // Track if QKV has been fused (do only once)
 }
 
 // Load loads the Z-Image model from ollama blob storage.
@@ -196,13 +199,17 @@ func (m *Model) generate(ctx context.Context, cfg *GenerateConfig) (*mlx.Array,
 	if cfg.CFGScale <= 0 {
 		cfg.CFGScale = 4.0
 	}
-	if cfg.LayerCache {
-		if cfg.CacheInterval <= 0 {
-			cfg.CacheInterval = 3
-		}
-		if cfg.CacheLayers <= 0 {
-			cfg.CacheLayers = 15 // Half of 30 layers
-		}
+	// TeaCache enabled by default
+	cfg.TeaCache = true
+	if cfg.TeaCacheThreshold <= 0 {
+		cfg.TeaCacheThreshold = 0.15
+	}
+
+	// Enable fused QKV if requested (only fuse once)
+	if cfg.FusedQKV && !m.qkvFused {
+		m.Transformer.FuseAllQKV()
+		m.qkvFused = true
+		fmt.Println("  Fused QKV enabled")
 	}
 
 	useCFG := cfg.NegativePrompt != ""
@@ -260,12 +267,54 @@ func (m *Model) generate(ctx context.Context, cfg *GenerateConfig) (*mlx.Array,
 		mlx.Eval(ropeCache.UnifiedCos)
 	}
 
-	// Step cache for shallow layer reuse (DeepCache/Learning-to-Cache style)
-	var stepCache *cache.StepCache
-	if cfg.LayerCache {
-		stepCache = cache.NewStepCache(cfg.CacheLayers)
-		fmt.Printf("  Layer caching enabled: %d layers, refresh every %d steps\n",
-			cfg.CacheLayers, cfg.CacheInterval)
+	// Pre-compute batched embeddings for CFG (outside the loop for efficiency)
+	var batchedEmb *mlx.Array
+	if useCFG {
+		// Concatenate embeddings once: [1, L, D] + [1, L, D] -> [2, L, D]
+		batchedEmb = mlx.Concatenate([]*mlx.Array{posEmb, negEmb}, 0)
+		mlx.Keep(batchedEmb)
+		mlx.Eval(batchedEmb)
+	}
+
+	// TeaCache for timestep-aware caching
+	// For CFG mode, we cache pos/neg separately, skip early steps, and always compute CFG fresh
+	var teaCache *cache.TeaCache
+	if cfg.TeaCache {
+		skipEarly := 0
+		if useCFG {
+			skipEarly = 3 // Skip first 3 steps for CFG to preserve structure
+		}
+		teaCache = cache.NewTeaCache(&cache.TeaCacheConfig{
+			Threshold:      cfg.TeaCacheThreshold,
+			RescaleFactor:  1.0,
+			SkipEarlySteps: skipEarly,
+		})
+		if useCFG {
+			fmt.Printf("  TeaCache enabled (CFG mode): threshold=%.2f, skip first %d steps\n", cfg.TeaCacheThreshold, skipEarly)
+		} else {
+			fmt.Printf("  TeaCache enabled: threshold=%.2f\n", cfg.TeaCacheThreshold)
+		}
+	}
+
+	// cleanup frees all kept arrays when we need to abort early
+	cleanup := func() {
+		posEmb.Free()
+		if negEmb != nil {
+			negEmb.Free()
+		}
+		ropeCache.ImgCos.Free()
+		ropeCache.ImgSin.Free()
+		ropeCache.CapCos.Free()
+		ropeCache.CapSin.Free()
+		ropeCache.UnifiedCos.Free()
+		ropeCache.UnifiedSin.Free()
+		if batchedEmb != nil {
+			batchedEmb.Free()
+		}
+		if teaCache != nil {
+			teaCache.Free()
+		}
+		latents.Free()
 	}
 
 	// Denoising loop
@@ -277,6 +326,7 @@ func (m *Model) generate(ctx context.Context, cfg *GenerateConfig) (*mlx.Array,
 		if ctx != nil {
 			select {
 			case <-ctx.Done():
+				cleanup()
 				return nil, ctx.Err()
 			default:
 			}
@@ -289,50 +339,77 @@ func (m *Model) generate(ctx context.Context, cfg *GenerateConfig) (*mlx.Array,
 		}
 
 		tCurr := scheduler.Timesteps[i]
-		timestep := mlx.ToBFloat16(mlx.NewArray([]float32{1.0 - tCurr}, []int32{1}))
+		var noisePred *mlx.Array
 
-		patches := PatchifyLatents(latents, tcfg.PatchSize)
+		// TeaCache: check if we should compute or reuse cached output
+		shouldCompute := teaCache == nil || teaCache.ShouldCompute(i, tCurr)
 
-		var output *mlx.Array
-		if stepCache != nil {
-			// Use layer caching for faster inference
+		if shouldCompute {
+			timestep := mlx.ToBFloat16(mlx.NewArray([]float32{1.0 - tCurr}, []int32{1}))
+			patches := PatchifyLatents(latents, tcfg.PatchSize)
+
+			var output *mlx.Array
 			if useCFG {
-				posOutput := m.Transformer.ForwardWithCache(patches, timestep, posEmb, ropeCache,
-					stepCache, i, cfg.CacheInterval)
-				// Note: CFG with layer cache shares the cache between pos/neg
-				// This is approximate but fast - neg prompt uses same cached shallow layers
-				negOutput := m.Transformer.ForwardWithCache(patches, timestep, negEmb, ropeCache,
-					stepCache, i, cfg.CacheInterval)
-				diff := mlx.Sub(posOutput, negOutput)
+				// CFG Batching: single forward pass with batch=2
+				// Tile patches: [1, L, D] -> [2, L, D]
+				batchedPatches := mlx.Tile(patches, []int32{2, 1, 1})
+				// Tile timestep: [1] -> [2]
+				batchedTimestep := mlx.Tile(timestep, []int32{2})
+
+				// Single batched forward pass (RoPE broadcasts from [1,L,H,D] to [2,L,H,D])
+				batchedOutput := m.Transformer.Forward(batchedPatches, batchedTimestep, batchedEmb, ropeCache)
+
+				// Split output: [2, L, D] -> pos [1, L, D], neg [1, L, D]
+				outputShape := batchedOutput.Shape()
+				L := outputShape[1]
+				D := outputShape[2]
+				posOutput := mlx.Slice(batchedOutput, []int32{0, 0, 0}, []int32{1, L, D})
+				negOutput := mlx.Slice(batchedOutput, []int32{1, 0, 0}, []int32{2, L, D})
+
+				// Convert to noise predictions (unpatchify and negate)
+				posPred := UnpatchifyLatents(posOutput, tcfg.PatchSize, latentH, latentW, tcfg.InChannels)
+				posPred = mlx.Neg(posPred)
+				negPred := UnpatchifyLatents(negOutput, tcfg.PatchSize, latentH, latentW, tcfg.InChannels)
+				negPred = mlx.Neg(negPred)
+
+				// Cache pos/neg separately for TeaCache
+				if teaCache != nil {
+					teaCache.UpdateCFGCache(posPred, negPred, tCurr)
+					mlx.Keep(teaCache.Arrays()...)
+				}
+
+				// Apply CFG: noisePred = neg + scale * (pos - neg)
+				diff := mlx.Sub(posPred, negPred)
 				scaledDiff := mlx.MulScalar(diff, cfg.CFGScale)
-				output = mlx.Add(negOutput, scaledDiff)
-			} else {
-				output = m.Transformer.ForwardWithCache(patches, timestep, posEmb, ropeCache,
-					stepCache, i, cfg.CacheInterval)
-			}
-		} else {
-			// Standard forward without caching
-			if useCFG {
-				posOutput := m.Transformer.Forward(patches, timestep, posEmb, ropeCache)
-				negOutput := m.Transformer.Forward(patches, timestep, negEmb, ropeCache)
-				diff := mlx.Sub(posOutput, negOutput)
-				scaledDiff := mlx.MulScalar(diff, cfg.CFGScale)
-				output = mlx.Add(negOutput, scaledDiff)
+				noisePred = mlx.Add(negPred, scaledDiff)
 			} else {
+				// Non-CFG forward pass
 				output = m.Transformer.Forward(patches, timestep, posEmb, ropeCache)
-			}
-		}
+				noisePred = UnpatchifyLatents(output, tcfg.PatchSize, latentH, latentW, tcfg.InChannels)
+				noisePred = mlx.Neg(noisePred)
 
-		noisePred := UnpatchifyLatents(output, tcfg.PatchSize, latentH, latentW, tcfg.InChannels)
-		noisePred = mlx.Neg(noisePred)
+				// Update TeaCache
+				if teaCache != nil {
+					teaCache.UpdateCache(noisePred, tCurr)
+					mlx.Keep(teaCache.Arrays()...)
+				}
+			}
+		} else if useCFG && teaCache != nil && teaCache.HasCFGCache() {
+			// CFG mode: get cached pos/neg and compute CFG fresh
+			posPred, negPred := teaCache.GetCFGCached()
+			diff := mlx.Sub(posPred, negPred)
+			scaledDiff := mlx.MulScalar(diff, cfg.CFGScale)
+			noisePred = mlx.Add(negPred, scaledDiff)
+			fmt.Printf("    [TeaCache: reusing cached pos/neg outputs]\n")
+		} else {
+			// Non-CFG mode: reuse cached noise prediction
+			noisePred = teaCache.GetCached()
+			fmt.Printf("    [TeaCache: reusing cached output]\n")
+		}
 
 		oldLatents := latents
 		latents = scheduler.Step(noisePred, latents, i)
 
-		// Keep latents and any cached arrays
-		if stepCache != nil {
-			mlx.Keep(stepCache.Arrays()...)
-		}
 		mlx.Eval(latents)
 		oldLatents.Free()
 
@@ -361,8 +438,14 @@ func (m *Model) generate(ctx context.Context, cfg *GenerateConfig) (*mlx.Array,
 	ropeCache.CapSin.Free()
 	ropeCache.UnifiedCos.Free()
 	ropeCache.UnifiedSin.Free()
-	if stepCache != nil {
-		stepCache.Free()
+	if batchedEmb != nil {
+		batchedEmb.Free()
+	}
+	if teaCache != nil {
+		hits, misses := teaCache.Stats()
+		fmt.Printf("  TeaCache stats: %d hits, %d misses (%.1f%% cache rate)\n",
+			hits, misses, float64(hits)/float64(hits+misses)*100)
+		teaCache.Free()
 	}
 
 	// VAE decode

x/imagegen/nn/nn.go

@@ -10,6 +10,13 @@ type Layer interface {
 	Forward(x *mlx.Array) *mlx.Array
 }
 
+// LinearLayer is an interface for linear layers (both regular and quantized).
+// This allows swapping between Linear and QuantizedLinear at runtime.
+type LinearLayer interface {
+	Forward(x *mlx.Array) *mlx.Array
+	OutputDim() int32 // Returns the output dimension of the layer
+}
+
 // Linear applies an affine transformation: y = x @ W.T + b
 // Weight is stored as [out_features, in_features], matching PyTorch/MLX convention.
 type Linear struct {
@@ -49,6 +56,11 @@ func (l *Linear) Forward(x *mlx.Array) *mlx.Array {
 	return mlx.Linear(x, w)
 }
 
+// OutputDim returns the output dimension of the linear layer.
+func (l *Linear) OutputDim() int32 {
+	return l.Weight.Shape()[0]
+}
+
 // ToQuantized converts this Linear to a QuantizedLinear.
 func (l *Linear) ToQuantized(groupSize, bits int, mode string) *QuantizedLinear {
 	qw, scales, qbiases := mlx.Quantize(l.Weight, groupSize, bits, mode)
@@ -84,6 +96,13 @@ func (ql *QuantizedLinear) Forward(x *mlx.Array) *mlx.Array {
 	return out
 }
 
+// OutputDim returns the output dimension of the quantized linear layer.
+// For mxfp8/mxfp4, quantized weight shape is [out_features, in_features / group_size].
+// The output dimension is the first dimension of the weight.
+func (ql *QuantizedLinear) OutputDim() int32 {
+	return ql.Weight.Shape()[0]
+}
+
 // RMSNorm represents an RMS normalization layer.
 type RMSNorm struct {
 	Weight *mlx.Array `weight:"weight"`

x/imagegen/quantize.go

@@ -0,0 +1,22 @@
+package imagegen
+
+import (
+	"io"
+	"strings"
+)
+
+// QuantizingTensorLayerCreator creates tensor layers with optional quantization.
+// When quantize is true, returns multiple layers (weight + scales + biases).
+type QuantizingTensorLayerCreator func(r io.Reader, name, dtype string, shape []int32, quantize bool) ([]LayerInfo, error)
+
+// ShouldQuantize returns true if a tensor should be quantized.
+// Quantizes linear weights only, skipping VAE, embeddings, norms, and biases.
+func ShouldQuantize(name, component string) bool {
+	if component == "vae" {
+		return false
+	}
+	if strings.Contains(name, "embed") || strings.Contains(name, "norm") {
+		return false
+	}
+	return strings.HasSuffix(name, ".weight")
+}

x/imagegen/runner/runner.go

@@ -13,16 +13,21 @@ import (
 	"net/http"
 	"os"
 	"os/signal"
-	"path/filepath"
 	"sync"
 	"syscall"
 	"time"
 
 	"github.com/ollama/ollama/x/imagegen"
 	"github.com/ollama/ollama/x/imagegen/mlx"
+	"github.com/ollama/ollama/x/imagegen/models/glm_image"
 	"github.com/ollama/ollama/x/imagegen/models/zimage"
 )
 
+// ImageModel is the interface for image generation models
+type ImageModel interface {
+	GenerateImage(ctx context.Context, prompt string, width, height int32, steps int, seed int64) (*mlx.Array, error)
+}
+
 // Request is the image generation request format
 type Request struct {
 	Prompt string `json:"prompt"`
@@ -34,15 +39,17 @@ type Request struct {
 
 // Response is streamed back for each progress update
 type Response struct {
-	Content string `json:"content"`
+	Content string `json:"content,omitempty"`
+	Image   string `json:"image,omitempty"` // Base64-encoded PNG
 	Done    bool   `json:"done"`
 }
 
 // Server holds the model and handles requests
 type Server struct {
 	mu        sync.Mutex
-	model     *zimage.Model
+	model     ImageModel
 	modelName string
+	modelType string // "zimage" or "glm_image"
 }
 
 // Execute is the entry point for the image runner subprocess
@@ -72,15 +79,35 @@ func Execute(args []string) error {
 			requiredMemory/(1024*1024*1024), availableMemory/(1024*1024*1024))
 	}
 
-	// Load model
-	model := &zimage.Model{}
-	if err := model.Load(*modelName); err != nil {
-		return fmt.Errorf("failed to load model: %w", err)
+	// Detect model type and load appropriate model
+	modelType, err := detectModelType(*modelName)
+	if err != nil {
+		return fmt.Errorf("failed to detect model type: %w", err)
+	}
+
+	var model ImageModel
+	switch modelType {
+	case "GlmImagePipeline":
+		slog.Info("loading GLM-Image model")
+		m := &glm_image.Model{}
+		if err := m.Load(*modelName); err != nil {
+			return fmt.Errorf("failed to load GLM-Image model: %w", err)
+		}
+		model = m
+	default:
+		// Default to zimage for ZImagePipeline, FluxPipeline, and unknown types
+		slog.Info("loading Z-Image model")
+		m := &zimage.Model{}
+		if err := m.Load(*modelName); err != nil {
+			return fmt.Errorf("failed to load Z-Image model: %w", err)
+		}
+		model = m
 	}
 
 	server := &Server{
 		model:     model,
 		modelName: *modelName,
+		modelType: modelType,
 	}
 
 	// Set up HTTP handlers
@@ -144,7 +171,13 @@ func (s *Server) completionHandler(w http.ResponseWriter, r *http.Request) {
 		req.Height = 1024
 	}
 	if req.Steps <= 0 {
-		req.Steps = 9
+		// Default steps depend on model type
+		switch s.modelType {
+		case "GlmImagePipeline":
+			req.Steps = 50 // GLM-Image default
+		default:
+			req.Steps = 9 // Z-Image turbo default
+		}
 	}
 	if req.Seed <= 0 {
 		req.Seed = time.Now().UnixNano()
@@ -159,25 +192,9 @@ func (s *Server) completionHandler(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 
-	// Generate image
+	// Generate image using interface method
 	ctx := r.Context()
-	img, err := s.model.GenerateFromConfig(ctx, &zimage.GenerateConfig{
-		Prompt: req.Prompt,
-		Width:  req.Width,
-		Height: req.Height,
-		Steps:  req.Steps,
-		Seed:   req.Seed,
-		Progress: func(step, total int) {
-			resp := Response{
-				Content: fmt.Sprintf("\rGenerating: step %d/%d", step, total),
-				Done:    false,
-			}
-			data, _ := json.Marshal(resp)
-			w.Write(data)
-			w.Write([]byte("\n"))
-			flusher.Flush()
-		},
-	})
+	img, err := s.model.GenerateImage(ctx, req.Prompt, req.Width, req.Height, req.Steps, req.Seed)
 
 	if err != nil {
 		// Don't send error for cancellation
@@ -191,10 +208,10 @@ func (s *Server) completionHandler(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 
-	// Save image
-	outPath := filepath.Join(os.TempDir(), fmt.Sprintf("ollama-image-%d.png", time.Now().UnixNano()))
-	if err := imagegen.SaveImage(img, outPath); err != nil {
-		resp := Response{Content: fmt.Sprintf("error saving: %v", err), Done: true}
+	// Encode image as base64 PNG
+	imageData, err := imagegen.EncodeImageBase64(img)
+	if err != nil {
+		resp := Response{Content: fmt.Sprintf("error encoding: %v", err), Done: true}
 		data, _ := json.Marshal(resp)
 		w.Write(data)
 		w.Write([]byte("\n"))
@@ -204,14 +221,47 @@ func (s *Server) completionHandler(w http.ResponseWriter, r *http.Request) {
 	// Free the generated image array and clean up MLX state
 	img.Free()
 	mlx.ClearCache()
+	mlx.MetalResetPeakMemory()
 
-	// Send final response
+	// Send final response with image data
 	resp := Response{
-		Content: fmt.Sprintf("\n\nImage saved to: %s\n", outPath),
-		Done:    true,
+		Image: imageData,
+		Done:  true,
 	}
 	data, _ := json.Marshal(resp)
 	w.Write(data)
 	w.Write([]byte("\n"))
 	flusher.Flush()
 }
+
+// detectModelType reads the model manifest and returns the pipeline class name
+func detectModelType(modelName string) (string, error) {
+	manifest, err := imagegen.LoadManifest(modelName)
+	if err != nil {
+		return "", err
+	}
+
+	data, err := manifest.ReadConfig("model_index.json")
+	if err != nil {
+		return "ZImagePipeline", nil // Default to Z-Image
+	}
+
+	// Try both _class_name (diffusers format) and architecture (ollama format)
+	var index struct {
+		ClassName    string `json:"_class_name"`
+		Architecture string `json:"architecture"`
+	}
+	if err := json.Unmarshal(data, &index); err != nil {
+		return "ZImagePipeline", nil
+	}
+
+	// Prefer _class_name, fall back to architecture
+	className := index.ClassName
+	if className == "" {
+		className = index.Architecture
+	}
+	if className == "" {
+		return "ZImagePipeline", nil
+	}
+	return className, nil
+}

x/imagegen/safetensors/loader.go

@@ -8,6 +8,7 @@ import (
 	"strings"
 
 	"github.com/ollama/ollama/x/imagegen/mlx"
+	"github.com/ollama/ollama/x/imagegen/nn"
 )
 
 // WeightSource is an interface for loading weights.
@@ -102,6 +103,22 @@ func loadStruct(v reflect.Value, weights WeightSource, prefix string, errs *[]st
 			}
 		}
 
+		// Handle nn.LinearLayer interface fields specially
+		if field.Type == reflect.TypeOf((*nn.LinearLayer)(nil)).Elem() {
+			if !hasTag {
+				continue // no tag = skip
+			}
+			layer, err := LoadLinearLayer(weights, fullPath)
+			if err != nil {
+				if !optional {
+					*errs = append(*errs, fullPath+": "+err.Error())
+				}
+				continue
+			}
+			fieldVal.Set(reflect.ValueOf(layer))
+			continue
+		}
+
 		// Handle by kind
 		switch fieldVal.Kind() {
 		case reflect.Ptr:
@@ -176,3 +193,64 @@ func joinPath(prefix, suffix string) string {
 	}
 	return prefix + "." + suffix
 }
+
+// LoadLinearLayer loads a linear layer from weights, automatically detecting if it's quantized.
+// If {path}.weight_scale exists, dequantizes the weights.
+func LoadLinearLayer(weights WeightSource, path string) (nn.LinearLayer, error) {
+	// Check if this is a quantized layer by looking for scale tensor
+	scalePath := path + ".weight_scale"
+	if weights.HasTensor(scalePath) {
+		weight, err := weights.GetTensor(path + ".weight")
+		if err != nil {
+			return nil, fmt.Errorf("failed to load quantized weight %s: %w", path, err)
+		}
+
+		scales, err := weights.GetTensor(scalePath)
+		if err != nil {
+			return nil, fmt.Errorf("failed to load scales %s: %w", scalePath, err)
+		}
+
+		// Bias is optional
+		var bias *mlx.Array
+		biasPath := path + ".bias"
+		if weights.HasTensor(biasPath) {
+			bias, _ = weights.GetTensor(biasPath)
+		}
+
+		var qbiases *mlx.Array
+		qbiasPath := path + ".weight_qbias"
+		if weights.HasTensor(qbiasPath) {
+			qbiases, _ = weights.GetTensor(qbiasPath)
+		}
+
+		if mlx.MetalIsAvailable() {
+			return &nn.QuantizedLinear{
+				Weight:    weight,
+				Scales:    scales,
+				QBiases:   qbiases,
+				Bias:      bias,
+				GroupSize: 32,
+				Bits:      8,
+				Mode:      "affine",
+			}, nil
+		}
+
+		dequantized := mlx.Dequantize(weight, scales, qbiases, 32, 8, "affine")
+		return nn.NewLinear(dequantized, bias), nil
+	}
+
+	// Load as regular Linear
+	weight, err := weights.GetTensor(path + ".weight")
+	if err != nil {
+		return nil, fmt.Errorf("failed to load weight %s: %w", path, err)
+	}
+
+	// Bias is optional
+	var bias *mlx.Array
+	biasPath := path + ".bias"
+	if weights.HasTensor(biasPath) {
+		bias, _ = weights.GetTensor(biasPath)
+	}
+
+	return nn.NewLinear(weight, bias), nil
+}

x/imagegen/server.go

@@ -14,7 +14,9 @@ import (
 	"os"
 	"os/exec"
 	"path/filepath"
+	"runtime"
 	"strconv"
+	"strings"
 	"sync"
 	"time"
 
@@ -46,7 +48,8 @@ type completionRequest struct {
 
 // completionResponse is received from the subprocess
 type completionResponse struct {
-	Content string `json:"content"`
+	Content string `json:"content,omitempty"`
+	Image   string `json:"image,omitempty"`
 	Done    bool   `json:"done"`
 }
 
@@ -69,7 +72,7 @@ func NewServer(modelName string) (*Server, error) {
 		port = rand.Intn(65535-49152) + 49152
 	}
 
-	// Get the ollama executable path
+	// Get the ollama-mlx executable path (in same directory as current executable)
 	exe, err := os.Executable()
 	if err != nil {
 		return nil, fmt.Errorf("unable to lookup executable path: %w", err)
@@ -77,11 +80,42 @@ func NewServer(modelName string) (*Server, error) {
 	if eval, err := filepath.EvalSymlinks(exe); err == nil {
 		exe = eval
 	}
+	mlxExe := filepath.Join(filepath.Dir(exe), "ollama-mlx")
 
-	// Spawn subprocess: ollama runner --image-engine --model <path> --port <port>
-	cmd := exec.Command(exe, "runner", "--image-engine", "--model", modelName, "--port", strconv.Itoa(port))
+	// Spawn subprocess: ollama-mlx runner --image-engine --model <path> --port <port>
+	cmd := exec.Command(mlxExe, "runner", "--image-engine", "--model", modelName, "--port", strconv.Itoa(port))
 	cmd.Env = os.Environ()
 
+	// On Linux, set LD_LIBRARY_PATH to include MLX library directories
+	if runtime.GOOS == "linux" {
+		// Build library paths: start with LibOllamaPath, then add any mlx_* subdirectories
+		libraryPaths := []string{ml.LibOllamaPath}
+		if mlxDirs, err := filepath.Glob(filepath.Join(ml.LibOllamaPath, "mlx_*")); err == nil {
+			libraryPaths = append(libraryPaths, mlxDirs...)
+		}
+
+		// Append existing LD_LIBRARY_PATH if set
+		if existingPath, ok := os.LookupEnv("LD_LIBRARY_PATH"); ok {
+			libraryPaths = append(libraryPaths, filepath.SplitList(existingPath)...)
+		}
+
+		pathEnvVal := strings.Join(libraryPaths, string(filepath.ListSeparator))
+
+		// Update or add LD_LIBRARY_PATH in cmd.Env
+		found := false
+		for i := range cmd.Env {
+			if strings.HasPrefix(cmd.Env[i], "LD_LIBRARY_PATH=") {
+				cmd.Env[i] = "LD_LIBRARY_PATH=" + pathEnvVal
+				found = true
+				break
+			}
+		}
+		if !found {
+			cmd.Env = append(cmd.Env, "LD_LIBRARY_PATH="+pathEnvVal)
+		}
+		slog.Debug("mlx subprocess library path", "LD_LIBRARY_PATH", pathEnvVal)
+	}
+
 	s := &Server{
 		cmd:       cmd,
 		port:      port,
@@ -112,7 +146,7 @@ func NewServer(modelName string) (*Server, error) {
 		}
 	}()
 
-	slog.Info("starting image runner subprocess", "model", modelName, "port", port)
+	slog.Info("starting ollama-mlx image runner subprocess", "exe", mlxExe, "model", modelName, "port", port)
 	if err := cmd.Start(); err != nil {
 		return nil, fmt.Errorf("failed to start image runner: %w", err)
 	}
@@ -250,15 +284,23 @@ func (s *Server) Completion(ctx context.Context, req llm.CompletionRequest, fn f
 		return fmt.Errorf("completion request failed: %d", resp.StatusCode)
 	}
 
-	// Stream responses
+	// Stream responses - use large buffer for base64 image data
 	scanner := bufio.NewScanner(resp.Body)
+	scanner.Buffer(make([]byte, 1024*1024), 16*1024*1024) // 16MB max
 	for scanner.Scan() {
 		var cresp completionResponse
 		if err := json.Unmarshal(scanner.Bytes(), &cresp); err != nil {
 			continue
 		}
+
+		content := cresp.Content
+		// If this is the final response with an image, encode it in the content
+		if cresp.Done && cresp.Image != "" {
+			content = "IMAGE_BASE64:" + cresp.Image
+		}
+
 		fn(llm.CompletionResponse{
-			Content: cresp.Content,
+			Content: content,
 			Done:    cresp.Done,
 		})
 		if cresp.Done {

x/imagegen/transfer/download.go

@@ -45,24 +45,33 @@ func download(ctx context.Context, opts DownloadOptions) error {
 		return nil
 	}
 
-	// Filter existing
-	var blobs []Blob
+	// Calculate total from all blobs (for accurate progress reporting on resume)
 	var total int64
+	for _, b := range opts.Blobs {
+		total += b.Size
+	}
+
+	// Filter out already-downloaded blobs and track completed bytes
+	var blobs []Blob
+	var alreadyCompleted int64
 	for _, b := range opts.Blobs {
 		if fi, _ := os.Stat(filepath.Join(opts.DestDir, digestToPath(b.Digest))); fi != nil && fi.Size() == b.Size {
 			if opts.Logger != nil {
 				opts.Logger.Debug("blob already exists", "digest", b.Digest, "size", b.Size)
 			}
+			alreadyCompleted += b.Size
 			continue
 		}
 		blobs = append(blobs, b)
-		total += b.Size
 	}
 	if len(blobs) == 0 {
 		return nil
 	}
 
 	token := opts.Token
+	progress := newProgressTracker(total, opts.Progress)
+	progress.add(alreadyCompleted) // Report already-downloaded bytes upfront
+
 	d := &downloader{
 		client:       cmp.Or(opts.Client, defaultClient),
 		baseURL:      opts.BaseURL,
@@ -72,7 +81,7 @@ func download(ctx context.Context, opts DownloadOptions) error {
 		getToken:     opts.GetToken,
 		userAgent:    cmp.Or(opts.UserAgent, defaultUserAgent),
 		stallTimeout: cmp.Or(opts.StallTimeout, defaultStallTimeout),
-		progress:     newProgressTracker(total, opts.Progress),
+		progress:     progress,
 		speeds:       &speedTracker{},
 		logger:       opts.Logger,
 	}

x/imagegen/transfer/transfer.go

@@ -110,8 +110,6 @@ var defaultClient = &http.Client{
 		MaxIdleConnsPerHost: 100,
 		IdleConnTimeout:     90 * time.Second,
 	},
-	Timeout: 5 * time.Minute,
-	// Don't follow redirects automatically - we handle them manually
 	CheckRedirect: func(req *http.Request, via []*http.Request) error {
 		return http.ErrUseLastResponse
 	},

x/imagegen/transfer/transfer_test.go

@@ -284,6 +284,83 @@ func TestDownloadSkipsExisting(t *testing.T) {
 	}
 }
 
+func TestDownloadResumeProgressTotal(t *testing.T) {
+	// Test that when resuming a download with some blobs already present:
+	// 1. Total reflects ALL blob sizes (not just remaining)
+	// 2. Completed starts at the size of already-downloaded blobs
+	serverDir := t.TempDir()
+	blob1, data1 := createTestBlob(t, serverDir, 1000)
+	blob2, data2 := createTestBlob(t, serverDir, 2000)
+	blob3, data3 := createTestBlob(t, serverDir, 3000)
+
+	// Pre-populate client with blob1 and blob2 (simulating partial download)
+	clientDir := t.TempDir()
+	for _, b := range []struct {
+		blob Blob
+		data []byte
+	}{{blob1, data1}, {blob2, data2}} {
+		path := filepath.Join(clientDir, digestToPath(b.blob.Digest))
+		if err := os.MkdirAll(filepath.Dir(path), 0o755); err != nil {
+			t.Fatal(err)
+		}
+		if err := os.WriteFile(path, b.data, 0o644); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		digest := filepath.Base(r.URL.Path)
+		path := filepath.Join(serverDir, digestToPath(digest))
+		data, err := os.ReadFile(path)
+		if err != nil {
+			http.NotFound(w, r)
+			return
+		}
+		w.Header().Set("Content-Length", fmt.Sprintf("%d", len(data)))
+		w.WriteHeader(http.StatusOK)
+		w.Write(data)
+	}))
+	defer server.Close()
+
+	var firstCompleted, firstTotal int64
+	var gotFirstProgress bool
+	var mu sync.Mutex
+
+	err := Download(context.Background(), DownloadOptions{
+		Blobs:       []Blob{blob1, blob2, blob3},
+		BaseURL:     server.URL,
+		DestDir:     clientDir,
+		Concurrency: 1,
+		Progress: func(completed, total int64) {
+			mu.Lock()
+			defer mu.Unlock()
+			if !gotFirstProgress {
+				firstCompleted = completed
+				firstTotal = total
+				gotFirstProgress = true
+			}
+		},
+	})
+	if err != nil {
+		t.Fatalf("Download failed: %v", err)
+	}
+
+	// Total should be sum of ALL blobs, not just blob3
+	expectedTotal := blob1.Size + blob2.Size + blob3.Size
+	if firstTotal != expectedTotal {
+		t.Errorf("Total = %d, want %d (should include all blobs)", firstTotal, expectedTotal)
+	}
+
+	// First progress call should show already-completed bytes from blob1+blob2
+	expectedCompleted := blob1.Size + blob2.Size
+	if firstCompleted < expectedCompleted {
+		t.Errorf("First completed = %d, want >= %d (should include already-downloaded blobs)", firstCompleted, expectedCompleted)
+	}
+
+	// Verify blob3 was downloaded
+	verifyBlob(t, clientDir, blob3, data3)
+}
+
 func TestDownloadDigestMismatch(t *testing.T) {
 	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 		// Return wrong data

x/tools/registry.go

@@ -54,6 +54,16 @@ func (r *Registry) RegisterBash() {
 	r.Register(&BashTool{})
 }
 
+// RegisterWebSearch adds the web search tool to the registry.
+func (r *Registry) RegisterWebSearch() {
+	r.Register(&WebSearchTool{})
+}
+
+// RegisterWebFetch adds the web fetch tool to the registry.
+func (r *Registry) RegisterWebFetch() {
+	r.Register(&WebFetchTool{})
+}
+
 // Get retrieves a tool by name.
 func (r *Registry) Get(name string) (Tool, bool) {
 	tool, ok := r.tools[name]

x/tools/webfetch.go

@@ -0,0 +1,162 @@
+package tools
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"net/url"
+	"strconv"
+	"strings"
+	"time"
+
+	"github.com/ollama/ollama/api"
+	"github.com/ollama/ollama/auth"
+)
+
+const (
+	webFetchAPI     = "https://ollama.com/api/web_fetch"
+	webFetchTimeout = 30 * time.Second
+)
+
+// ErrWebFetchAuthRequired is returned when web fetch requires authentication
+var ErrWebFetchAuthRequired = errors.New("web fetch requires authentication")
+
+// WebFetchTool implements web page fetching using Ollama's hosted API.
+type WebFetchTool struct{}
+
+// Name returns the tool name.
+func (w *WebFetchTool) Name() string {
+	return "web_fetch"
+}
+
+// Description returns a description of the tool.
+func (w *WebFetchTool) Description() string {
+	return "Fetch and extract text content from a web page. Use this to read the full content of a URL found in search results or provided by the user."
+}
+
+// Schema returns the tool's parameter schema.
+func (w *WebFetchTool) Schema() api.ToolFunction {
+	props := api.NewToolPropertiesMap()
+	props.Set("url", api.ToolProperty{
+		Type:        api.PropertyType{"string"},
+		Description: "The URL to fetch and extract content from",
+	})
+	return api.ToolFunction{
+		Name:        w.Name(),
+		Description: w.Description(),
+		Parameters: api.ToolFunctionParameters{
+			Type:       "object",
+			Properties: props,
+			Required:   []string{"url"},
+		},
+	}
+}
+
+// webFetchRequest is the request body for the web fetch API.
+type webFetchRequest struct {
+	URL string `json:"url"`
+}
+
+// webFetchResponse is the response from the web fetch API.
+type webFetchResponse struct {
+	Title   string   `json:"title"`
+	Content string   `json:"content"`
+	Links   []string `json:"links,omitempty"`
+}
+
+// Execute fetches content from a web page.
+// Uses Ollama key signing for authentication - this makes requests via ollama.com API.
+func (w *WebFetchTool) Execute(args map[string]any) (string, error) {
+	urlStr, ok := args["url"].(string)
+	if !ok || urlStr == "" {
+		return "", fmt.Errorf("url parameter is required")
+	}
+
+	// Validate URL
+	if _, err := url.Parse(urlStr); err != nil {
+		return "", fmt.Errorf("invalid URL: %w", err)
+	}
+
+	// Prepare request
+	reqBody := webFetchRequest{
+		URL: urlStr,
+	}
+
+	jsonBody, err := json.Marshal(reqBody)
+	if err != nil {
+		return "", fmt.Errorf("marshaling request: %w", err)
+	}
+
+	// Parse URL and add timestamp for signing
+	fetchURL, err := url.Parse(webFetchAPI)
+	if err != nil {
+		return "", fmt.Errorf("parsing fetch URL: %w", err)
+	}
+
+	q := fetchURL.Query()
+	q.Add("ts", strconv.FormatInt(time.Now().Unix(), 10))
+	fetchURL.RawQuery = q.Encode()
+
+	// Sign the request using Ollama key (~/.ollama/id_ed25519)
+	ctx := context.Background()
+	data := fmt.Appendf(nil, "%s,%s", http.MethodPost, fetchURL.RequestURI())
+	signature, err := auth.Sign(ctx, data)
+	if err != nil {
+		return "", fmt.Errorf("signing request: %w", err)
+	}
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, fetchURL.String(), bytes.NewBuffer(jsonBody))
+	if err != nil {
+		return "", fmt.Errorf("creating request: %w", err)
+	}
+
+	req.Header.Set("Content-Type", "application/json")
+	if signature != "" {
+		req.Header.Set("Authorization", fmt.Sprintf("Bearer %s", signature))
+	}
+
+	// Send request
+	client := &http.Client{Timeout: webFetchTimeout}
+	resp, err := client.Do(req)
+	if err != nil {
+		return "", fmt.Errorf("sending request: %w", err)
+	}
+	defer resp.Body.Close()
+
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return "", fmt.Errorf("reading response: %w", err)
+	}
+
+	if resp.StatusCode == http.StatusUnauthorized {
+		return "", ErrWebFetchAuthRequired
+	}
+	if resp.StatusCode != http.StatusOK {
+		return "", fmt.Errorf("web fetch API returned status %d: %s", resp.StatusCode, string(body))
+	}
+
+	// Parse response
+	var fetchResp webFetchResponse
+	if err := json.Unmarshal(body, &fetchResp); err != nil {
+		return "", fmt.Errorf("parsing response: %w", err)
+	}
+
+	// Format result
+	var sb strings.Builder
+	if fetchResp.Title != "" {
+		sb.WriteString(fmt.Sprintf("Title: %s\n\n", fetchResp.Title))
+	}
+
+	if fetchResp.Content != "" {
+		sb.WriteString("Content:\n")
+		sb.WriteString(fetchResp.Content)
+	} else {
+		sb.WriteString("No content could be extracted from the page.")
+	}
+
+	return sb.String(), nil
+}