better i thikn

.github/actions/typecheck/action.yml

@@ -0,0 +1,12 @@
+name: Type Check
+
+description: "Run type checker"
+
+runs:
+  using: "composite"
+  steps:
+    - name: Run type checker
+      run: |
+        nix --extra-experimental-features nix-command --extra-experimental-features flakes develop -c just sync
+        nix --extra-experimental-features nix-command --extra-experimental-features flakes develop -c just check
+      shell: bash

.github/workflows/pipeline.yml

@@ -26,14 +26,73 @@ jobs:
           name: exo
           authToken: "${{ secrets.CACHIX_AUTH_TOKEN }}"
 
-      - name: Load nix develop environment
-        run: nix run github:nicknovitski/nix-develop/v1
+      - name: Configure git user
+        run: |
+          git config --local user.email "github-actions@users.noreply.github.com"
+          git config --local user.name  "github-actions bot"
+        shell: bash
 
-      - name: Sync dependencies
-        run: uv sync --all-packages
+      - name: Pull LFS files
+        run: |
+          echo "Pulling Git LFS files..."
+          git lfs pull
+        shell: bash
 
-      - name: Run type checker
-        run: uv run basedpyright --project pyproject.toml
+      - name: Setup Nix Environment
+        run: |
+          echo "Checking for nix installation..."
+          
+          # Check if nix binary exists directly
+          if [ -f /nix/var/nix/profiles/default/bin/nix ]; then
+            echo "Found nix binary at /nix/var/nix/profiles/default/bin/nix"
+            export PATH="/nix/var/nix/profiles/default/bin:$PATH"
+            echo "PATH=$PATH" >> $GITHUB_ENV
+            nix --version
+          elif [ -f /nix/var/nix/profiles/default/etc/profile.d/nix-daemon.sh ]; then
+            echo "Found nix profile script, sourcing..."
+            source /nix/var/nix/profiles/default/etc/profile.d/nix-daemon.sh
+            nix --version
+          elif command -v nix >/dev/null 2>&1; then
+            echo "Nix already in PATH"
+            nix --version
+          else
+            echo "Nix not found. Debugging info:"
+            echo "Contents of /nix/var/nix/profiles/default/:"
+            ls -la /nix/var/nix/profiles/default/ 2>/dev/null || echo "Directory not found"
+            echo "Contents of /nix/var/nix/profiles/default/bin/:"
+            ls -la /nix/var/nix/profiles/default/bin/ 2>/dev/null || echo "Directory not found"
+            exit 1
+          fi
+        shell: bash
+
+      - name: Configure basedpyright include for local MLX
+        run: |
+          RUNNER_LABELS='${{ toJSON(runner.labels) }}'
+          if echo "$RUNNER_LABELS" | grep -q "local_mlx"; then
+            if [ -d "/Users/Shared/mlx" ]; then
+              echo "Updating [tool.basedpyright].include to use /Users/Shared/mlx"
+              awk '
+                BEGIN { in=0 }
+                /^\[tool\.basedpyright\]/ { in=1; print; next }
+                in && /^\[/ { in=0 }  # next section
+                in && /^[ \t]*include[ \t]*=/ {
+                  print "include = [\"/Users/Shared/mlx\"]"
+                  next
+                }
+                { print }
+              ' pyproject.toml > pyproject.toml.tmp && mv pyproject.toml.tmp pyproject.toml
+
+              echo "New [tool.basedpyright] section:"
+              sed -n '/^\[tool\.basedpyright\]/,/^\[/p' pyproject.toml | sed '$d' || true
+            else
+              echo "local_mlx tag present but /Users/Shared/mlx not found; leaving pyproject unchanged."
+            fi
+          else
+            echo "Runner does not have 'local_mlx' tag; leaving pyproject unchanged."
+          fi
+        shell: bash
+
+      - uses: ./.github/actions/typecheck
 
   nix:
     name: Build and check (${{ matrix.system }})
@@ -132,14 +191,3 @@ jobs:
 
       - name: Run nix flake check
         run: nix flake check
-
-      - name: Run pytest (macOS only)
-        if: runner.os == 'macOS'
-        run: |
-          # Build the test environment (requires relaxed sandbox for uv2nix on macOS)
-          TEST_ENV=$(nix build '.#exo-test-env' --option sandbox relaxed --print-out-paths)
-
-          # Run pytest outside sandbox (needs GPU access for MLX)
-          export HOME="$RUNNER_TEMP"
-          export EXO_TESTS=1
-          $TEST_ENV/bin/python -m pytest src -m "not slow" --import-mode=importlib

app/EXO/EXO.xcodeproj/project.pbxproj

@@ -342,8 +342,6 @@
 				SDKROOT = macosx;
 				SWIFT_ACTIVE_COMPILATION_CONDITIONS = "DEBUG $(inherited)";
 				SWIFT_OPTIMIZATION_LEVEL = "-Onone";
-				SWIFT_TREAT_WARNINGS_AS_ERRORS = YES;
-				GCC_TREAT_WARNINGS_AS_ERRORS = YES;
 			};
 			name = Debug;
 		};
@@ -399,8 +397,6 @@
 				MTL_FAST_MATH = YES;
 				SDKROOT = macosx;
 				SWIFT_COMPILATION_MODE = wholemodule;
-				SWIFT_TREAT_WARNINGS_AS_ERRORS = YES;
-				GCC_TREAT_WARNINGS_AS_ERRORS = YES;
 			};
 			name = Release;
 		};

dashboard/package-lock.json

@@ -865,6 +865,7 @@
 			"integrity": "sha512-oH8tXw7EZnie8FdOWYrF7Yn4IKrqTFHhXvl8YxXxbKwTMcD/5NNCryUSEXRk2ZR4ojnub0P8rNrsVGHXWqIDtA==",
 			"dev": true,
 			"license": "MIT",
+			"peer": true,
 			"dependencies": {
 				"@standard-schema/spec": "^1.0.0",
 				"@sveltejs/acorn-typescript": "^1.0.5",
@@ -904,6 +905,7 @@
 			"integrity": "sha512-Y1Cs7hhTc+a5E9Va/xwKlAJoariQyHY+5zBgCZg4PFWNYQ1nMN9sjK1zhw1gK69DuqVP++sht/1GZg1aRwmAXQ==",
 			"dev": true,
 			"license": "MIT",
+			"peer": true,
 			"dependencies": {
 				"@sveltejs/vite-plugin-svelte-inspector": "^4.0.1",
 				"debug": "^4.4.1",
@@ -1520,6 +1522,7 @@
 			"integrity": "sha512-LCCV0HdSZZZb34qifBsyWlUmok6W7ouER+oQIGBScS8EsZsQbrtFTUrDX4hOl+CS6p7cnNC4td+qrSVGSCTUfQ==",
 			"dev": true,
 			"license": "MIT",
+			"peer": true,
 			"dependencies": {
 				"undici-types": "~6.21.0"
 			}
@@ -1529,6 +1532,7 @@
 			"resolved": "https://registry.npmjs.org/acorn/-/acorn-8.15.0.tgz",
 			"integrity": "sha512-NZyJarBfL7nWwIq+FDL6Zp/yHEhePMNnnJ0y3qfieCrmNvYct8uvtiV41UvlSe6apAfk0fY1FbWx+NwfmpvtTg==",
 			"license": "MIT",
+			"peer": true,
 			"bin": {
 				"acorn": "bin/acorn"
 			},
@@ -1941,6 +1945,7 @@
 			"integrity": "sha512-fmTRWbNMmsmWq6xJV8D19U/gw/bwrHfNXxrIN+HfZgnzqTHp9jOmKMhsTUjXOJnZOdZY9Q28y4yebKzqDKlxlQ==",
 			"dev": true,
 			"license": "ISC",
+			"peer": true,
 			"engines": {
 				"node": ">=12"
 			}
@@ -2648,6 +2653,7 @@
 			"integrity": "sha512-5gTmgEY/sqK6gFXLIsQNH19lWb4ebPDLA4SdLP7dsWkIXHWlG66oPuVvXSGFPppYZz8ZDZq0dYYrbHfBCVUb1Q==",
 			"dev": true,
 			"license": "MIT",
+			"peer": true,
 			"engines": {
 				"node": ">=12"
 			},
@@ -2690,6 +2696,7 @@
 			"integrity": "sha512-UOnG6LftzbdaHZcKoPFtOcCKztrQ57WkHDeRD9t/PTQtmT0NHSeWWepj6pS0z/N7+08BHFDQVUrfmfMRcZwbMg==",
 			"dev": true,
 			"license": "MIT",
+			"peer": true,
 			"bin": {
 				"prettier": "bin/prettier.cjs"
 			},
@@ -2862,6 +2869,7 @@
 			"resolved": "https://registry.npmjs.org/svelte/-/svelte-5.45.3.tgz",
 			"integrity": "sha512-ngKXNhNvwPzF43QqEhDOue7TQTrG09em1sd4HBxVF0Wr2gopAmdEWan+rgbdgK4fhBtSOTJO8bYU4chUG7VXZQ==",
 			"license": "MIT",
+			"peer": true,
 			"dependencies": {
 				"@jridgewell/remapping": "^2.3.4",
 				"@jridgewell/sourcemap-codec": "^1.5.0",
@@ -3006,6 +3014,7 @@
 			"integrity": "sha512-jl1vZzPDinLr9eUt3J/t7V6FgNEw9QjvBPdysz9KfQDD41fQrC2Y4vKQdiaUpFT4bXlb1RHhLpp8wtm6M5TgSw==",
 			"dev": true,
 			"license": "Apache-2.0",
+			"peer": true,
 			"bin": {
 				"tsc": "bin/tsc",
 				"tsserver": "bin/tsserver"
@@ -3027,6 +3036,7 @@
 			"integrity": "sha512-+Oxm7q9hDoLMyJOYfUYBuHQo+dkAloi33apOPP56pzj+vsdJDzr+j1NISE5pyaAuKL4A3UD34qd0lx5+kfKp2g==",
 			"dev": true,
 			"license": "MIT",
+			"peer": true,
 			"dependencies": {
 				"esbuild": "^0.25.0",
 				"fdir": "^6.4.4",

dashboard/src/lib/stores/app.svelte.ts

@@ -173,11 +173,6 @@ export interface PlacementPreviewResponse {
   previews: PlacementPreview[];
 }
 
-interface ImageApiResponse {
-  created: number;
-  data: Array<{ b64_json?: string; url?: string }>;
-}
-
 interface RawStateResponse {
   topology?: RawTopology;
   instances?: Record<
@@ -2100,137 +2095,107 @@ class AppStore {
         throw new Error(`API error: ${response.status} - ${errorText}`);
       }
 
-      // Streaming requires both stream=true AND partialImages > 0
-      const isStreaming = params.stream && params.partialImages > 0;
-
-      if (!isStreaming) {
-        // Non-streaming: parse JSON response directly
-        const jsonResponse = (await response.json()) as ImageApiResponse;
-        const format = params.outputFormat || "png";
-        const mimeType = `image/${format}`;
-
-        const attachments: MessageAttachment[] = jsonResponse.data
-          .filter((img) => img.b64_json)
-          .map((img, index) => ({
-            type: "generated-image" as const,
-            name: `generated-image-${index + 1}.${format}`,
-            preview: `data:${mimeType};base64,${img.b64_json}`,
-            mimeType,
-          }));
-
-        this.updateConversationMessage(
-          targetConversationId,
-          assistantMessage.id,
-          (msg) => {
-            msg.content = "";
-            msg.attachments = attachments;
-          },
-        );
-        this.syncActiveMessagesIfNeeded(targetConversationId);
-      } else {
-        // Streaming mode: use SSE parser
-        const reader = response.body?.getReader();
-        if (!reader) {
-          throw new Error("No response body");
-        }
-
-        interface ImageGenerationChunk {
-          data?: { b64_json?: string };
-          format?: string;
-          type?: "partial" | "final";
-          image_index?: number;
-          partial_index?: number;
-          total_partials?: number;
-        }
-
-        const numImages = params.numImages;
-
-        await this.parseSSEStream<ImageGenerationChunk>(
-          reader,
-          targetConversationId,
-          (parsed) => {
-            const imageData = parsed.data?.b64_json;
-
-            if (imageData) {
-              const format = parsed.format || "png";
-              const mimeType = `image/${format}`;
-              const imageIndex = parsed.image_index ?? 0;
-
-              if (parsed.type === "partial") {
-                // Update with partial image and progress
-                const partialNum = (parsed.partial_index ?? 0) + 1;
-                const totalPartials = parsed.total_partials ?? 3;
-                const progressText =
-                  numImages > 1
-                    ? `Generating image ${imageIndex + 1}/${numImages}... ${partialNum}/${totalPartials}`
-                    : `Generating... ${partialNum}/${totalPartials}`;
-
-                const partialAttachment: MessageAttachment = {
-                  type: "generated-image",
-                  name: `generated-image.${format}`,
-                  preview: `data:${mimeType};base64,${imageData}`,
-                  mimeType,
-                };
-
-                this.updateConversationMessage(
-                  targetConversationId,
-                  assistantMessage.id,
-                  (msg) => {
-                    msg.content = progressText;
-                    if (imageIndex === 0) {
-                      // First image - safe to replace attachments with partial preview
-                      msg.attachments = [partialAttachment];
-                    } else {
-                      // Subsequent images - keep existing finals, show partial at current position
-                      const existingAttachments = msg.attachments || [];
-                      // Keep only the completed final images (up to current imageIndex)
-                      const finals = existingAttachments.slice(0, imageIndex);
-                      msg.attachments = [...finals, partialAttachment];
-                    }
-                  },
-                );
-              } else if (parsed.type === "final") {
-                // Final image - replace partial at this position
-                const newAttachment: MessageAttachment = {
-                  type: "generated-image",
-                  name: `generated-image-${imageIndex + 1}.${format}`,
-                  preview: `data:${mimeType};base64,${imageData}`,
-                  mimeType,
-                };
-
-                this.updateConversationMessage(
-                  targetConversationId,
-                  assistantMessage.id,
-                  (msg) => {
-                    if (imageIndex === 0) {
-                      // First final image - replace any partial preview
-                      msg.attachments = [newAttachment];
-                    } else {
-                      // Subsequent images - keep previous finals, replace partial at current position
-                      const existingAttachments = msg.attachments || [];
-                      // Slice keeps indices 0 to imageIndex-1 (the previous final images)
-                      const previousFinals = existingAttachments.slice(
-                        0,
-                        imageIndex,
-                      );
-                      msg.attachments = [...previousFinals, newAttachment];
-                    }
-
-                    // Update progress message for multiple images
-                    if (numImages > 1 && imageIndex < numImages - 1) {
-                      msg.content = `Generating image ${imageIndex + 2}/${numImages}...`;
-                    } else {
-                      msg.content = "";
-                    }
-                  },
-                );
-              }
-
-              this.syncActiveMessagesIfNeeded(targetConversationId);
-            }
-          },
-        );
+      const reader = response.body?.getReader();
+      if (!reader) {
+        throw new Error("No response body");
       }
+
+      interface ImageGenerationChunk {
+        data?: { b64_json?: string };
+        format?: string;
+        type?: "partial" | "final";
+        image_index?: number;
+        partial_index?: number;
+        total_partials?: number;
+      }
+
+      const numImages = params.numImages;
+
+      await this.parseSSEStream<ImageGenerationChunk>(
+        reader,
+        targetConversationId,
+        (parsed) => {
+          const imageData = parsed.data?.b64_json;
+
+          if (imageData) {
+            const format = parsed.format || "png";
+            const mimeType = `image/${format}`;
+            const imageIndex = parsed.image_index ?? 0;
+
+            if (parsed.type === "partial") {
+              // Update with partial image and progress
+              const partialNum = (parsed.partial_index ?? 0) + 1;
+              const totalPartials = parsed.total_partials ?? 3;
+              const progressText =
+                numImages > 1
+                  ? `Generating image ${imageIndex + 1}/${numImages}... ${partialNum}/${totalPartials}`
+                  : `Generating... ${partialNum}/${totalPartials}`;
+
+              const partialAttachment: MessageAttachment = {
+                type: "generated-image",
+                name: `generated-image.${format}`,
+                preview: `data:${mimeType};base64,${imageData}`,
+                mimeType,
+              };
+
+              this.updateConversationMessage(
+                targetConversationId,
+                assistantMessage.id,
+                (msg) => {
+                  msg.content = progressText;
+                  if (imageIndex === 0) {
+                    // First image - safe to replace attachments with partial preview
+                    msg.attachments = [partialAttachment];
+                  } else {
+                    // Subsequent images - keep existing finals, show partial at current position
+                    const existingAttachments = msg.attachments || [];
+                    // Keep only the completed final images (up to current imageIndex)
+                    const finals = existingAttachments.slice(0, imageIndex);
+                    msg.attachments = [...finals, partialAttachment];
+                  }
+                },
+              );
+            } else if (parsed.type === "final") {
+              // Final image - replace partial at this position
+              const newAttachment: MessageAttachment = {
+                type: "generated-image",
+                name: `generated-image-${imageIndex + 1}.${format}`,
+                preview: `data:${mimeType};base64,${imageData}`,
+                mimeType,
+              };
+
+              this.updateConversationMessage(
+                targetConversationId,
+                assistantMessage.id,
+                (msg) => {
+                  if (imageIndex === 0) {
+                    // First final image - replace any partial preview
+                    msg.attachments = [newAttachment];
+                  } else {
+                    // Subsequent images - keep previous finals, replace partial at current position
+                    const existingAttachments = msg.attachments || [];
+                    // Slice keeps indices 0 to imageIndex-1 (the previous final images)
+                    const previousFinals = existingAttachments.slice(
+                      0,
+                      imageIndex,
+                    );
+                    msg.attachments = [...previousFinals, newAttachment];
+                  }
+
+                  // Update progress message for multiple images
+                  if (numImages > 1 && imageIndex < numImages - 1) {
+                    msg.content = `Generating image ${imageIndex + 2}/${numImages}...`;
+                  } else {
+                    msg.content = "";
+                  }
+                },
+              );
+            }
+
+            this.syncActiveMessagesIfNeeded(targetConversationId);
+          }
+        },
+      );
     } catch (error) {
       console.error("Error generating image:", error);
       this.handleStreamingError(
@@ -2378,98 +2343,69 @@ class AppStore {
         throw new Error(`API error: ${apiResponse.status} - ${errorText}`);
       }
 
-      // Streaming requires both stream=true AND partialImages > 0
-      const isStreaming = params.stream && params.partialImages > 0;
-
-      if (!isStreaming) {
-        // Non-streaming: parse JSON response directly
-        const jsonResponse = (await apiResponse.json()) as ImageApiResponse;
-        const format = params.outputFormat || "png";
-        const mimeType = `image/${format}`;
-        const attachments: MessageAttachment[] = jsonResponse.data
-          .filter((img) => img.b64_json)
-          .map((img) => ({
-            type: "generated-image" as const,
-            name: `edited-image.${format}`,
-            preview: `data:${mimeType};base64,${img.b64_json}`,
-            mimeType,
-          }));
-
-        this.updateConversationMessage(
-          targetConversationId,
-          assistantMessage.id,
-          (msg) => {
-            msg.content = "";
-            msg.attachments = attachments;
-          },
-        );
-        this.syncActiveMessagesIfNeeded(targetConversationId);
-      } else {
-        // Streaming mode: use SSE parser
-        const reader = apiResponse.body?.getReader();
-        if (!reader) {
-          throw new Error("No response body");
-        }
-
-        interface ImageEditChunk {
-          data?: { b64_json?: string };
-          format?: string;
-          type?: "partial" | "final";
-          partial_index?: number;
-          total_partials?: number;
-        }
-
-        await this.parseSSEStream<ImageEditChunk>(
-          reader,
-          targetConversationId,
-          (parsed) => {
-            const imageData = parsed.data?.b64_json;
-
-            if (imageData) {
-              const format = parsed.format || "png";
-              const mimeType = `image/${format}`;
-              if (parsed.type === "partial") {
-                // Update with partial image and progress
-                const partialNum = (parsed.partial_index ?? 0) + 1;
-                const totalPartials = parsed.total_partials ?? 3;
-                this.updateConversationMessage(
-                  targetConversationId,
-                  assistantMessage.id,
-                  (msg) => {
-                    msg.content = `Editing... ${partialNum}/${totalPartials}`;
-                    msg.attachments = [
-                      {
-                        type: "generated-image",
-                        name: `edited-image.${format}`,
-                        preview: `data:${mimeType};base64,${imageData}`,
-                        mimeType,
-                      },
-                    ];
-                  },
-                );
-              } else if (parsed.type === "final") {
-                // Final image
-                this.updateConversationMessage(
-                  targetConversationId,
-                  assistantMessage.id,
-                  (msg) => {
-                    msg.content = "";
-                    msg.attachments = [
-                      {
-                        type: "generated-image",
-                        name: `edited-image.${format}`,
-                        preview: `data:${mimeType};base64,${imageData}`,
-                        mimeType,
-                      },
-                    ];
-                  },
-                );
-              }
-              this.syncActiveMessagesIfNeeded(targetConversationId);
-            }
-          },
-        );
+      const reader = apiResponse.body?.getReader();
+      if (!reader) {
+        throw new Error("No response body");
       }
+
+      interface ImageEditChunk {
+        data?: { b64_json?: string };
+        format?: string;
+        type?: "partial" | "final";
+        partial_index?: number;
+        total_partials?: number;
+      }
+
+      await this.parseSSEStream<ImageEditChunk>(
+        reader,
+        targetConversationId,
+        (parsed) => {
+          const imageData = parsed.data?.b64_json;
+
+          if (imageData) {
+            const format = parsed.format || "png";
+            const mimeType = `image/${format}`;
+            if (parsed.type === "partial") {
+              // Update with partial image and progress
+              const partialNum = (parsed.partial_index ?? 0) + 1;
+              const totalPartials = parsed.total_partials ?? 3;
+              this.updateConversationMessage(
+                targetConversationId,
+                assistantMessage.id,
+                (msg) => {
+                  msg.content = `Editing... ${partialNum}/${totalPartials}`;
+                  msg.attachments = [
+                    {
+                      type: "generated-image",
+                      name: `edited-image.${format}`,
+                      preview: `data:${mimeType};base64,${imageData}`,
+                      mimeType,
+                    },
+                  ];
+                },
+              );
+            } else if (parsed.type === "final") {
+              // Final image
+              this.updateConversationMessage(
+                targetConversationId,
+                assistantMessage.id,
+                (msg) => {
+                  msg.content = "";
+                  msg.attachments = [
+                    {
+                      type: "generated-image",
+                      name: `edited-image.${format}`,
+                      preview: `data:${mimeType};base64,${imageData}`,
+                      mimeType,
+                    },
+                  ];
+                },
+              );
+            }
+            this.syncActiveMessagesIfNeeded(targetConversationId);
+          }
+        },
+      );
     } catch (error) {
       console.error("Error editing image:", error);
       this.handleStreamingError(

flake.lock

@@ -21,9 +21,7 @@
           "nixpkgs"
         ],
         "purescript-overlay": "purescript-overlay",
-        "pyproject-nix": [
-          "pyproject-nix"
-        ]
+        "pyproject-nix": "pyproject-nix"
       },
       "locked": {
         "lastModified": 1765953015,
@@ -151,44 +149,19 @@
         "type": "github"
       }
     },
-    "pyproject-build-systems": {
-      "inputs": {
-        "nixpkgs": [
-          "nixpkgs"
-        ],
-        "pyproject-nix": [
-          "pyproject-nix"
-        ],
-        "uv2nix": [
-          "uv2nix"
-        ]
-      },
-      "locked": {
-        "lastModified": 1763662255,
-        "narHash": "sha256-4bocaOyLa3AfiS8KrWjZQYu+IAta05u3gYZzZ6zXbT0=",
-        "owner": "pyproject-nix",
-        "repo": "build-system-pkgs",
-        "rev": "042904167604c681a090c07eb6967b4dd4dae88c",
-        "type": "github"
-      },
-      "original": {
-        "owner": "pyproject-nix",
-        "repo": "build-system-pkgs",
-        "type": "github"
-      }
-    },
     "pyproject-nix": {
       "inputs": {
         "nixpkgs": [
+          "dream2nix",
           "nixpkgs"
         ]
       },
       "locked": {
-        "lastModified": 1764134915,
-        "narHash": "sha256-xaKvtPx6YAnA3HQVp5LwyYG1MaN4LLehpQI8xEdBvBY=",
+        "lastModified": 1763017646,
+        "narHash": "sha256-Z+R2lveIp6Skn1VPH3taQIuMhABg1IizJd8oVdmdHsQ=",
         "owner": "pyproject-nix",
         "repo": "pyproject.nix",
-        "rev": "2c8df1383b32e5443c921f61224b198a2282a657",
+        "rev": "47bd6f296502842643078d66128f7b5e5370790c",
         "type": "github"
       },
       "original": {
@@ -205,10 +178,7 @@
         "flake-parts": "flake-parts",
         "nixpkgs": "nixpkgs",
         "nixpkgs-swift": "nixpkgs-swift",
-        "pyproject-build-systems": "pyproject-build-systems",
-        "pyproject-nix": "pyproject-nix",
-        "treefmt-nix": "treefmt-nix",
-        "uv2nix": "uv2nix"
+        "treefmt-nix": "treefmt-nix"
       }
     },
     "rust-analyzer-src": {
@@ -269,29 +239,6 @@
         "repo": "treefmt-nix",
         "type": "github"
       }
-    },
-    "uv2nix": {
-      "inputs": {
-        "nixpkgs": [
-          "nixpkgs"
-        ],
-        "pyproject-nix": [
-          "pyproject-nix"
-        ]
-      },
-      "locked": {
-        "lastModified": 1767701098,
-        "narHash": "sha256-CJhKZnWb3gumR9oTRjFvCg/6lYTGbZRU7xtvcyWIRwU=",
-        "owner": "pyproject-nix",
-        "repo": "uv2nix",
-        "rev": "9d357f0d2ce6f5f35ec7959d7e704452352eb4da",
-        "type": "github"
-      },
-      "original": {
-        "owner": "pyproject-nix",
-        "repo": "uv2nix",
-        "type": "github"
-      }
     }
   },
   "root": "root",

flake.nix

@@ -24,26 +24,6 @@
     dream2nix = {
       url = "github:nix-community/dream2nix";
       inputs.nixpkgs.follows = "nixpkgs";
-      inputs.pyproject-nix.follows = "pyproject-nix";
-    };
-
-    # Python packaging with uv2nix
-    pyproject-nix = {
-      url = "github:pyproject-nix/pyproject.nix";
-      inputs.nixpkgs.follows = "nixpkgs";
-    };
-
-    uv2nix = {
-      url = "github:pyproject-nix/uv2nix";
-      inputs.pyproject-nix.follows = "pyproject-nix";
-      inputs.nixpkgs.follows = "nixpkgs";
-    };
-
-    pyproject-build-systems = {
-      url = "github:pyproject-nix/build-system-pkgs";
-      inputs.pyproject-nix.follows = "pyproject-nix";
-      inputs.uv2nix.follows = "uv2nix";
-      inputs.nixpkgs.follows = "nixpkgs";
     };
 
     # Pinned nixpkgs for swift-format (swift is broken on x86_64-linux in newer nixpkgs)
@@ -68,7 +48,6 @@
         inputs.treefmt-nix.flakeModule
         ./dashboard/parts.nix
         ./rust/parts.nix
-        ./python/parts.nix
       ];
 
       perSystem =
@@ -109,6 +88,12 @@
             };
           };
 
+          checks.lint = pkgs.runCommand "lint-check" { } ''
+            export RUFF_CACHE_DIR="$TMPDIR/ruff-cache"
+            ${pkgs.ruff}/bin/ruff check ${inputs.self}/
+            touch $out
+          '';
+
           packages = lib.optionalAttrs pkgs.stdenv.hostPlatform.isDarwin (
             let
               uvLock = builtins.fromTOML (builtins.readFile ./uv.lock);

packaging/pyinstaller/exo.spec

@@ -10,6 +10,7 @@ PROJECT_ROOT = Path.cwd()
 SOURCE_ROOT = PROJECT_ROOT / "src"
 ENTRYPOINT = SOURCE_ROOT / "exo" / "__main__.py"
 DASHBOARD_DIR = PROJECT_ROOT / "dashboard" / "build"
+RESOURCES_DIR = PROJECT_ROOT / "resources"
 EXO_SHARED_MODELS_DIR = SOURCE_ROOT / "exo" / "shared" / "models"
 
 if not ENTRYPOINT.is_file():
@@ -18,6 +19,9 @@ if not ENTRYPOINT.is_file():
 if not DASHBOARD_DIR.is_dir():
     raise SystemExit(f"Dashboard assets are missing: {DASHBOARD_DIR}")
 
+if not RESOURCES_DIR.is_dir():
+    raise SystemExit(f"Resource assets are missing: {RESOURCES_DIR}")
+
 if not EXO_SHARED_MODELS_DIR.is_dir():
     raise SystemExit(f"Shared model assets are missing: {EXO_SHARED_MODELS_DIR}")
 
@@ -58,6 +62,7 @@ HIDDEN_IMPORTS = sorted(
 
 DATAS: list[tuple[str, str]] = [
     (str(DASHBOARD_DIR), "dashboard"),
+    (str(RESOURCES_DIR), "resources"),
     (str(MLX_LIB_DIR), "mlx/lib"),
     (str(EXO_SHARED_MODELS_DIR), "exo/shared/models"),
 ]

pyproject.toml

@@ -19,7 +19,7 @@ dependencies = [
     "anyio==4.11.0",
     "mlx==0.30.4; sys_platform == 'darwin'",
     "mlx[cpu]==0.30.4; sys_platform == 'linux'",
-    "mlx-lm==0.30.5",
+    "mlx-lm",
     "tiktoken>=0.12.0", # required for kimi k2 tokenizer
     "hypercorn>=0.18.0",
     "openai-harmony>=0.0.8",
@@ -31,6 +31,8 @@ dependencies = [
 ]
 
 [project.scripts]
+exo-master = "exo.master.main:main"
+exo-worker = "exo.worker.main:main"
 exo = "exo.main:main"
 
 # dependencies only required for development
@@ -44,6 +46,12 @@ dev = [
     "ruff>=0.11.13",
 ]
 
+# mlx[cuda] requires a newer version of mlx. the ideal on linux is: default to mlx[cpu] unless[cuda] specified.
+[project.optional-dependencies]
+# cuda = [
+#     "mlx[cuda]==0.26.3",
+# ]
+
 ###
 # workspace configuration
 ###
@@ -55,8 +63,8 @@ members = [
 
 [tool.uv.sources]
 exo_pyo3_bindings = { workspace = true }
+mlx-lm = { git = "https://github.com/ml-explore/mlx-lm", branch = "main" }
 # Uncomment to use local mlx/mlx-lm development versions:
-# mlx-lm = { git = "https://github.com/ml-explore/mlx-lm", branch = "main" }
 # mlx = { path = "/Users/Shared/mlx", editable=true }
 # mlx-lm = { path = "/Users/Shared/mlx-lm", editable=true }
 
@@ -108,7 +116,7 @@ environments = [
 ###
 
 [tool.ruff]
-extend-exclude = ["*mlx_typings/**", "rust/exo_pyo3_bindings/**"]
+extend-exclude = ["shared/protobufs/**", "*mlx_typings/**", "rust/exo_pyo3_bindings/**"]
 
 [tool.ruff.lint]
 extend-select = ["I", "N", "B", "A", "PIE", "SIM"]

python/parts.nix

@@ -1,93 +0,0 @@
-{ inputs, ... }:
-{
-  perSystem =
-    { config, self', pkgs, lib, system, ... }:
-    let
-      # Load workspace from uv.lock
-      workspace = inputs.uv2nix.lib.workspace.loadWorkspace {
-        workspaceRoot = inputs.self;
-      };
-
-      # Create overlay from workspace
-      # Use wheels from PyPI for most packages; we override mlx with our pure Nix Metal build
-      overlay = workspace.mkPyprojectOverlay { sourcePreference = "wheel"; };
-
-      # Override overlay to inject Nix-built components
-      exoOverlay = final: prev: {
-        # Replace workspace exo_pyo3_bindings with Nix-built wheel
-        exo-pyo3-bindings = pkgs.stdenv.mkDerivation {
-          pname = "exo-pyo3-bindings";
-          version = "0.1.0";
-          src = self'.packages.exo_pyo3_bindings;
-          # Install from pre-built wheel
-          nativeBuildInputs = [ final.pyprojectWheelHook ];
-          dontStrip = true;
-        };
-      };
-
-      python = pkgs.python313;
-
-      # Overlay to provide build systems and custom packages
-      buildSystemsOverlay = final: prev: {
-        # Use our pure Nix-built MLX with Metal support
-        mlx = self'.packages.mlx;
-
-        # mlx-lm is a git dependency that needs setuptools
-        mlx-lm = prev.mlx-lm.overrideAttrs (old: {
-          nativeBuildInputs = (old.nativeBuildInputs or [ ]) ++ [
-            final.setuptools
-          ];
-        });
-      };
-
-      pythonSet = (pkgs.callPackage inputs.pyproject-nix.build.packages {
-        inherit python;
-      }).overrideScope (
-        lib.composeManyExtensions [
-          inputs.pyproject-build-systems.overlays.default
-          overlay
-          exoOverlay
-          buildSystemsOverlay
-        ]
-      );
-      exoVenv = pythonSet.mkVirtualEnv "exo-env" workspace.deps.default;
-
-      # Virtual environment with dev dependencies for testing
-      testVenv = pythonSet.mkVirtualEnv "exo-test-env" (
-        workspace.deps.default // {
-          exo = [ "dev" ]; # Include pytest, pytest-asyncio, pytest-env
-        }
-      );
-
-      exoPackage = pkgs.runCommand "exo"
-        {
-          nativeBuildInputs = [ pkgs.makeWrapper ];
-        }
-        ''
-          mkdir -p $out/bin
-
-          # Create wrapper scripts
-          for script in exo exo-master exo-worker; do
-            makeWrapper ${exoVenv}/bin/$script $out/bin/$script \
-              --set DASHBOARD_DIR ${self'.packages.dashboard}
-          done
-        '';
-    in
-    {
-      # Python package only available on macOS (requires MLX/Metal)
-      packages = lib.optionalAttrs pkgs.stdenv.hostPlatform.isDarwin {
-        exo = exoPackage;
-        # Test environment for running pytest outside of Nix sandbox (needs GPU access)
-        exo-test-env = testVenv;
-      };
-
-      checks = {
-        # Ruff linting (works on all platforms)
-        lint = pkgs.runCommand "ruff-lint" { } ''
-          export RUFF_CACHE_DIR="$TMPDIR/ruff-cache"
-          ${pkgs.ruff}/bin/ruff check ${inputs.self}/
-          touch $out
-        '';
-      };
-    };
-}

resources/image_model_cards/exolabs--FLUX.1-Krea-dev-4bit.toml

@@ -0,0 +1,45 @@
+model_id = "exolabs/FLUX.1-Krea-dev-4bit"
+n_layers = 57
+hidden_size = 1
+supports_tensor = false
+tasks = ["TextToImage"]
+
+[storage_size]
+in_bytes = 15475325472
+
+[[components]]
+component_name = "text_encoder"
+component_path = "text_encoder/"
+n_layers = 12
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0
+
+[[components]]
+component_name = "text_encoder_2"
+component_path = "text_encoder_2/"
+n_layers = 24
+can_shard = false
+safetensors_index_filename = "model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 9524621312
+
+[[components]]
+component_name = "transformer"
+component_path = "transformer/"
+n_layers = 57
+can_shard = true
+safetensors_index_filename = "diffusion_pytorch_model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 5950704160
+
+[[components]]
+component_name = "vae"
+component_path = "vae/"
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0

resources/image_model_cards/exolabs--FLUX.1-Krea-dev-8bit.toml

@@ -0,0 +1,45 @@
+model_id = "exolabs/FLUX.1-Krea-dev-8bit"
+n_layers = 57
+hidden_size = 1
+supports_tensor = false
+tasks = ["TextToImage"]
+
+[storage_size]
+in_bytes = 21426029632
+
+[[components]]
+component_name = "text_encoder"
+component_path = "text_encoder/"
+n_layers = 12
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0
+
+[[components]]
+component_name = "text_encoder_2"
+component_path = "text_encoder_2/"
+n_layers = 24
+can_shard = false
+safetensors_index_filename = "model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 9524621312
+
+[[components]]
+component_name = "transformer"
+component_path = "transformer/"
+n_layers = 57
+can_shard = true
+safetensors_index_filename = "diffusion_pytorch_model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 11901408320
+
+[[components]]
+component_name = "vae"
+component_path = "vae/"
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0

resources/image_model_cards/exolabs--FLUX.1-Krea-dev.toml

@@ -0,0 +1,45 @@
+model_id = "exolabs/FLUX.1-Krea-dev"
+n_layers = 57
+hidden_size = 1
+supports_tensor = false
+tasks = ["TextToImage"]
+
+[storage_size]
+in_bytes = 33327437952
+
+[[components]]
+component_name = "text_encoder"
+component_path = "text_encoder/"
+n_layers = 12
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0
+
+[[components]]
+component_name = "text_encoder_2"
+component_path = "text_encoder_2/"
+n_layers = 24
+can_shard = false
+safetensors_index_filename = "model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 9524621312
+
+[[components]]
+component_name = "transformer"
+component_path = "transformer/"
+n_layers = 57
+can_shard = true
+safetensors_index_filename = "diffusion_pytorch_model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 23802816640
+
+[[components]]
+component_name = "vae"
+component_path = "vae/"
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0

resources/image_model_cards/exolabs--FLUX.1-dev-4bit.toml

@@ -0,0 +1,45 @@
+model_id = "exolabs/FLUX.1-dev-4bit"
+n_layers = 57
+hidden_size = 1
+supports_tensor = false
+tasks = ["TextToImage"]
+
+[storage_size]
+in_bytes = 15475325472
+
+[[components]]
+component_name = "text_encoder"
+component_path = "text_encoder/"
+n_layers = 12
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0
+
+[[components]]
+component_name = "text_encoder_2"
+component_path = "text_encoder_2/"
+n_layers = 24
+can_shard = false
+safetensors_index_filename = "model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 9524621312
+
+[[components]]
+component_name = "transformer"
+component_path = "transformer/"
+n_layers = 57
+can_shard = true
+safetensors_index_filename = "diffusion_pytorch_model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 5950704160
+
+[[components]]
+component_name = "vae"
+component_path = "vae/"
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0

resources/image_model_cards/exolabs--FLUX.1-dev-8bit.toml

@@ -0,0 +1,45 @@
+model_id = "exolabs/FLUX.1-dev-8bit"
+n_layers = 57
+hidden_size = 1
+supports_tensor = false
+tasks = ["TextToImage"]
+
+[storage_size]
+in_bytes = 21426029632
+
+[[components]]
+component_name = "text_encoder"
+component_path = "text_encoder/"
+n_layers = 12
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0
+
+[[components]]
+component_name = "text_encoder_2"
+component_path = "text_encoder_2/"
+n_layers = 24
+can_shard = false
+safetensors_index_filename = "model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 9524621312
+
+[[components]]
+component_name = "transformer"
+component_path = "transformer/"
+n_layers = 57
+can_shard = true
+safetensors_index_filename = "diffusion_pytorch_model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 11901408320
+
+[[components]]
+component_name = "vae"
+component_path = "vae/"
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0

resources/image_model_cards/exolabs--FLUX.1-dev.toml

@@ -0,0 +1,45 @@
+model_id = "exolabs/FLUX.1-dev"
+n_layers = 57
+hidden_size = 1
+supports_tensor = false
+tasks = ["TextToImage"]
+
+[storage_size]
+in_bytes = 33327437952
+
+[[components]]
+component_name = "text_encoder"
+component_path = "text_encoder/"
+n_layers = 12
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0
+
+[[components]]
+component_name = "text_encoder_2"
+component_path = "text_encoder_2/"
+n_layers = 24
+can_shard = false
+safetensors_index_filename = "model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 9524621312
+
+[[components]]
+component_name = "transformer"
+component_path = "transformer/"
+n_layers = 57
+can_shard = true
+safetensors_index_filename = "diffusion_pytorch_model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 23802816640
+
+[[components]]
+component_name = "vae"
+component_path = "vae/"
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0

resources/image_model_cards/exolabs--FLUX.1-schnell-4bit.toml

@@ -0,0 +1,45 @@
+model_id = "exolabs/FLUX.1-schnell-4bit"
+n_layers = 57
+hidden_size = 1
+supports_tensor = false
+tasks = ["TextToImage"]
+
+[storage_size]
+in_bytes = 15470210592
+
+[[components]]
+component_name = "text_encoder"
+component_path = "text_encoder/"
+n_layers = 12
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0
+
+[[components]]
+component_name = "text_encoder_2"
+component_path = "text_encoder_2/"
+n_layers = 24
+can_shard = false
+safetensors_index_filename = "model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 9524621312
+
+[[components]]
+component_name = "transformer"
+component_path = "transformer/"
+n_layers = 57
+can_shard = true
+safetensors_index_filename = "diffusion_pytorch_model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 5945589280
+
+[[components]]
+component_name = "vae"
+component_path = "vae/"
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0

resources/image_model_cards/exolabs--FLUX.1-schnell-8bit.toml

@@ -0,0 +1,45 @@
+model_id = "exolabs/FLUX.1-schnell-8bit"
+n_layers = 57
+hidden_size = 1
+supports_tensor = false
+tasks = ["TextToImage"]
+
+[storage_size]
+in_bytes = 21415799872
+
+[[components]]
+component_name = "text_encoder"
+component_path = "text_encoder/"
+n_layers = 12
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0
+
+[[components]]
+component_name = "text_encoder_2"
+component_path = "text_encoder_2/"
+n_layers = 24
+can_shard = false
+safetensors_index_filename = "model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 9524621312
+
+[[components]]
+component_name = "transformer"
+component_path = "transformer/"
+n_layers = 57
+can_shard = true
+safetensors_index_filename = "diffusion_pytorch_model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 11891178560
+
+[[components]]
+component_name = "vae"
+component_path = "vae/"
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0

resources/image_model_cards/exolabs--FLUX.1-schnell.toml

@@ -0,0 +1,45 @@
+model_id = "exolabs/FLUX.1-schnell"
+n_layers = 57
+hidden_size = 1
+supports_tensor = false
+tasks = ["TextToImage"]
+
+[storage_size]
+in_bytes = 33306978432
+
+[[components]]
+component_name = "text_encoder"
+component_path = "text_encoder/"
+n_layers = 12
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0
+
+[[components]]
+component_name = "text_encoder_2"
+component_path = "text_encoder_2/"
+n_layers = 24
+can_shard = false
+safetensors_index_filename = "model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 9524621312
+
+[[components]]
+component_name = "transformer"
+component_path = "transformer/"
+n_layers = 57
+can_shard = true
+safetensors_index_filename = "diffusion_pytorch_model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 23782357120
+
+[[components]]
+component_name = "vae"
+component_path = "vae/"
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0

resources/image_model_cards/exolabs--Qwen-Image-4bit.toml

@@ -0,0 +1,35 @@
+model_id = "exolabs/Qwen-Image-4bit"
+n_layers = 60
+hidden_size = 1
+supports_tensor = false
+tasks = ["TextToImage"]
+
+[storage_size]
+in_bytes = 26799533856
+
+[[components]]
+component_name = "text_encoder"
+component_path = "text_encoder/"
+n_layers = 12
+can_shard = false
+
+[components.storage_size]
+in_bytes = 16584333312
+
+[[components]]
+component_name = "transformer"
+component_path = "transformer/"
+n_layers = 60
+can_shard = true
+safetensors_index_filename = "diffusion_pytorch_model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 10215200544
+
+[[components]]
+component_name = "vae"
+component_path = "vae/"
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0

resources/image_model_cards/exolabs--Qwen-Image-8bit.toml

@@ -0,0 +1,35 @@
+model_id = "exolabs/Qwen-Image-8bit"
+n_layers = 60
+hidden_size = 1
+supports_tensor = false
+tasks = ["TextToImage"]
+
+[storage_size]
+in_bytes = 37014734400
+
+[[components]]
+component_name = "text_encoder"
+component_path = "text_encoder/"
+n_layers = 12
+can_shard = false
+
+[components.storage_size]
+in_bytes = 16584333312
+
+[[components]]
+component_name = "transformer"
+component_path = "transformer/"
+n_layers = 60
+can_shard = true
+safetensors_index_filename = "diffusion_pytorch_model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 20430401088
+
+[[components]]
+component_name = "vae"
+component_path = "vae/"
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0

resources/image_model_cards/exolabs--Qwen-Image-Edit-2509-4bit.toml

@@ -0,0 +1,35 @@
+model_id = "exolabs/Qwen-Image-Edit-2509-4bit"
+n_layers = 60
+hidden_size = 1
+supports_tensor = false
+tasks = ["ImageToImage"]
+
+[storage_size]
+in_bytes = 26799533856
+
+[[components]]
+component_name = "text_encoder"
+component_path = "text_encoder/"
+n_layers = 12
+can_shard = false
+
+[components.storage_size]
+in_bytes = 16584333312
+
+[[components]]
+component_name = "transformer"
+component_path = "transformer/"
+n_layers = 60
+can_shard = true
+safetensors_index_filename = "diffusion_pytorch_model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 10215200544
+
+[[components]]
+component_name = "vae"
+component_path = "vae/"
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0

resources/image_model_cards/exolabs--Qwen-Image-Edit-2509-8bit.toml

@@ -0,0 +1,35 @@
+model_id = "exolabs/Qwen-Image-Edit-2509-8bit"
+n_layers = 60
+hidden_size = 1
+supports_tensor = false
+tasks = ["ImageToImage"]
+
+[storage_size]
+in_bytes = 37014734400
+
+[[components]]
+component_name = "text_encoder"
+component_path = "text_encoder/"
+n_layers = 12
+can_shard = false
+
+[components.storage_size]
+in_bytes = 16584333312
+
+[[components]]
+component_name = "transformer"
+component_path = "transformer/"
+n_layers = 60
+can_shard = true
+safetensors_index_filename = "diffusion_pytorch_model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 20430401088
+
+[[components]]
+component_name = "vae"
+component_path = "vae/"
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0

resources/image_model_cards/exolabs--Qwen-Image-Edit-2509.toml

@@ -0,0 +1,35 @@
+model_id = "exolabs/Qwen-Image-Edit-2509"
+n_layers = 60
+hidden_size = 1
+supports_tensor = false
+tasks = ["ImageToImage"]
+
+[storage_size]
+in_bytes = 57445135488
+
+[[components]]
+component_name = "text_encoder"
+component_path = "text_encoder/"
+n_layers = 12
+can_shard = false
+
+[components.storage_size]
+in_bytes = 16584333312
+
+[[components]]
+component_name = "transformer"
+component_path = "transformer/"
+n_layers = 60
+can_shard = true
+safetensors_index_filename = "diffusion_pytorch_model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 40860802176
+
+[[components]]
+component_name = "vae"
+component_path = "vae/"
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0

resources/image_model_cards/exolabs--Qwen-Image.toml

@@ -0,0 +1,35 @@
+model_id = "exolabs/Qwen-Image"
+n_layers = 60
+hidden_size = 1
+supports_tensor = false
+tasks = ["TextToImage"]
+
+[storage_size]
+in_bytes = 57445135488
+
+[[components]]
+component_name = "text_encoder"
+component_path = "text_encoder/"
+n_layers = 12
+can_shard = false
+
+[components.storage_size]
+in_bytes = 16584333312
+
+[[components]]
+component_name = "transformer"
+component_path = "transformer/"
+n_layers = 60
+can_shard = true
+safetensors_index_filename = "diffusion_pytorch_model.safetensors.index.json"
+
+[components.storage_size]
+in_bytes = 40860802176
+
+[[components]]
+component_name = "vae"
+component_path = "vae/"
+can_shard = false
+
+[components.storage_size]
+in_bytes = 0

resources/inference_model_cards/mlx-community--DeepSeek-V3.1-4bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/DeepSeek-V3.1-4bit"
+n_layers = 61
+hidden_size = 7168
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 405874409472

resources/inference_model_cards/mlx-community--DeepSeek-V3.1-8bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/DeepSeek-V3.1-8bit"
+n_layers = 61
+hidden_size = 7168
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 765577920512

resources/inference_model_cards/mlx-community--GLM-4.5-Air-8bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/GLM-4.5-Air-8bit"
+n_layers = 46
+hidden_size = 4096
+supports_tensor = false
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 122406567936

resources/inference_model_cards/mlx-community--GLM-4.5-Air-bf16.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/GLM-4.5-Air-bf16"
+n_layers = 46
+hidden_size = 4096
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 229780750336

resources/inference_model_cards/mlx-community--GLM-4.7-4bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/GLM-4.7-4bit"
+n_layers = 91
+hidden_size = 5120
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 198556925568

resources/inference_model_cards/mlx-community--GLM-4.7-6bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/GLM-4.7-6bit"
+n_layers = 91
+hidden_size = 5120
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 286737579648

resources/inference_model_cards/mlx-community--GLM-4.7-8bit-gs32.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/GLM-4.7-8bit-gs32"
+n_layers = 91
+hidden_size = 5120
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 396963397248

resources/inference_model_cards/mlx-community--GLM-4.7-Flash-4bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/GLM-4.7-Flash-4bit"
+n_layers = 47
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 19327352832

resources/inference_model_cards/mlx-community--GLM-4.7-Flash-5bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/GLM-4.7-Flash-5bit"
+n_layers = 47
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 22548578304

resources/inference_model_cards/mlx-community--GLM-4.7-Flash-6bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/GLM-4.7-Flash-6bit"
+n_layers = 47
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 26843545600

resources/inference_model_cards/mlx-community--GLM-4.7-Flash-8bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/GLM-4.7-Flash-8bit"
+n_layers = 47
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 34359738368

resources/inference_model_cards/mlx-community--Kimi-K2-Instruct-4bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Kimi-K2-Instruct-4bit"
+n_layers = 61
+hidden_size = 7168
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 620622774272

resources/inference_model_cards/mlx-community--Kimi-K2-Thinking.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Kimi-K2-Thinking"
+n_layers = 61
+hidden_size = 7168
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 706522120192

resources/inference_model_cards/mlx-community--Kimi-K2.5.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Kimi-K2.5"
+n_layers = 61
+hidden_size = 7168
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 662498705408

resources/inference_model_cards/mlx-community--Llama-3.2-1B-Instruct-4bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Llama-3.2-1B-Instruct-4bit"
+n_layers = 16
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 729808896

resources/inference_model_cards/mlx-community--Llama-3.2-3B-Instruct-4bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Llama-3.2-3B-Instruct-4bit"
+n_layers = 28
+hidden_size = 3072
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 1863319552

resources/inference_model_cards/mlx-community--Llama-3.2-3B-Instruct-8bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Llama-3.2-3B-Instruct-8bit"
+n_layers = 28
+hidden_size = 3072
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 3501195264

resources/inference_model_cards/mlx-community--Llama-3.3-70B-Instruct-4bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Llama-3.3-70B-Instruct-4bit"
+n_layers = 80
+hidden_size = 8192
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 40652242944

resources/inference_model_cards/mlx-community--Llama-3.3-70B-Instruct-8bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Llama-3.3-70B-Instruct-8bit"
+n_layers = 80
+hidden_size = 8192
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 76799803392

resources/inference_model_cards/mlx-community--Meta-Llama-3.1-70B-Instruct-4bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Meta-Llama-3.1-70B-Instruct-4bit"
+n_layers = 80
+hidden_size = 8192
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 40652242944

resources/inference_model_cards/mlx-community--Meta-Llama-3.1-8B-Instruct-4bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Meta-Llama-3.1-8B-Instruct-4bit"
+n_layers = 32
+hidden_size = 4096
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 4637851648

resources/inference_model_cards/mlx-community--Meta-Llama-3.1-8B-Instruct-8bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Meta-Llama-3.1-8B-Instruct-8bit"
+n_layers = 32
+hidden_size = 4096
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 8954839040

resources/inference_model_cards/mlx-community--Meta-Llama-3.1-8B-Instruct-bf16.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Meta-Llama-3.1-8B-Instruct-bf16"
+n_layers = 32
+hidden_size = 4096
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 16882073600

resources/inference_model_cards/mlx-community--MiniMax-M2.1-3bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/MiniMax-M2.1-3bit"
+n_layers = 61
+hidden_size = 3072
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 100086644736

resources/inference_model_cards/mlx-community--MiniMax-M2.1-8bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/MiniMax-M2.1-8bit"
+n_layers = 61
+hidden_size = 3072
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 242986745856

resources/inference_model_cards/mlx-community--Qwen3-0.6B-4bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Qwen3-0.6B-4bit"
+n_layers = 28
+hidden_size = 1024
+supports_tensor = false
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 342884352

resources/inference_model_cards/mlx-community--Qwen3-0.6B-8bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Qwen3-0.6B-8bit"
+n_layers = 28
+hidden_size = 1024
+supports_tensor = false
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 698351616

resources/inference_model_cards/mlx-community--Qwen3-235B-A22B-Instruct-2507-4bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Qwen3-235B-A22B-Instruct-2507-4bit"
+n_layers = 94
+hidden_size = 4096
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 141733920768

resources/inference_model_cards/mlx-community--Qwen3-235B-A22B-Instruct-2507-8bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Qwen3-235B-A22B-Instruct-2507-8bit"
+n_layers = 94
+hidden_size = 4096
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 268435456000

resources/inference_model_cards/mlx-community--Qwen3-30B-A3B-4bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Qwen3-30B-A3B-4bit"
+n_layers = 48
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 17612931072

resources/inference_model_cards/mlx-community--Qwen3-30B-A3B-8bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Qwen3-30B-A3B-8bit"
+n_layers = 48
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 33279705088

resources/inference_model_cards/mlx-community--Qwen3-Coder-480B-A35B-Instruct-4bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Qwen3-Coder-480B-A35B-Instruct-4bit"
+n_layers = 62
+hidden_size = 6144
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 289910292480

resources/inference_model_cards/mlx-community--Qwen3-Coder-480B-A35B-Instruct-8bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Qwen3-Coder-480B-A35B-Instruct-8bit"
+n_layers = 62
+hidden_size = 6144
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 579820584960

resources/inference_model_cards/mlx-community--Qwen3-Next-80B-A3B-Instruct-4bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Qwen3-Next-80B-A3B-Instruct-4bit"
+n_layers = 48
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 46976204800

resources/inference_model_cards/mlx-community--Qwen3-Next-80B-A3B-Instruct-8bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Qwen3-Next-80B-A3B-Instruct-8bit"
+n_layers = 48
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 88814387200

resources/inference_model_cards/mlx-community--Qwen3-Next-80B-A3B-Thinking-4bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Qwen3-Next-80B-A3B-Thinking-4bit"
+n_layers = 48
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 88814387200

resources/inference_model_cards/mlx-community--Qwen3-Next-80B-A3B-Thinking-8bit.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/Qwen3-Next-80B-A3B-Thinking-8bit"
+n_layers = 48
+hidden_size = 2048
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 88814387200

resources/inference_model_cards/mlx-community--gpt-oss-120b-MXFP4-Q8.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/gpt-oss-120b-MXFP4-Q8"
+n_layers = 36
+hidden_size = 2880
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 70652212224

resources/inference_model_cards/mlx-community--gpt-oss-20b-MXFP4-Q8.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/gpt-oss-20b-MXFP4-Q8"
+n_layers = 24
+hidden_size = 2880
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 12025908224

resources/inference_model_cards/mlx-community--llama-3.3-70b-instruct-fp16.toml

@@ -0,0 +1,8 @@
+model_id = "mlx-community/llama-3.3-70b-instruct-fp16"
+n_layers = 80
+hidden_size = 8192
+supports_tensor = true
+tasks = ["TextGeneration"]
+
+[storage_size]
+in_bytes = 144383672320

src/exo/download/impl_shard_downloader.py

@@ -7,7 +7,7 @@ from loguru import logger
 
 from exo.download.download_utils import RepoDownloadProgress, download_shard
 from exo.download.shard_downloader import ShardDownloader
-from exo.shared.models.model_cards import MODEL_CARDS, ModelCard, ModelId
+from exo.shared.models.model_cards import ModelCard, ModelId, get_model_cards
 from exo.shared.types.worker.shards import (
     PipelineShardMetadata,
     ShardMetadata,
@@ -21,7 +21,7 @@ def exo_shard_downloader(max_parallel_downloads: int = 8) -> ShardDownloader:
 
 
 async def build_base_shard(model_id: ModelId) -> ShardMetadata:
-    model_card = await ModelCard.from_hf(model_id)
+    model_card = await ModelCard.fetch_from_hf(model_id)
     return PipelineShardMetadata(
         model_card=model_card,
         device_rank=0,
@@ -160,14 +160,15 @@ class ResumableShardDownloader(ShardDownloader):
         # Kick off download status coroutines concurrently
         tasks = [
             asyncio.create_task(_status_for_model(model_card.model_id))
-            for model_card in MODEL_CARDS.values()
+            for model_card in await get_model_cards()
         ]
 
         for task in asyncio.as_completed(tasks):
             try:
                 yield await task
+            # TODO: except Exception
             except Exception as e:
-                logger.warning(f"Error downloading shard: {type(e).__name__}")
+                logger.error("Error downloading shard:", e)
 
     async def get_shard_download_status_for_shard(
         self, shard: ShardMetadata

src/exo/master/api.py

@@ -22,16 +22,13 @@ from loguru import logger
 from exo.master.image_store import ImageStore
 from exo.master.placement import place_instance as get_instance_placements
 from exo.shared.apply import apply
-from exo.shared.constants import (
-    EXO_IMAGE_CACHE_DIR,
-    EXO_MAX_CHUNK_SIZE,
-)
+from exo.shared.constants import DASHBOARD_DIR, EXO_IMAGE_CACHE_DIR, EXO_MAX_CHUNK_SIZE
 from exo.shared.election import ElectionMessage
 from exo.shared.logging import InterceptLogger
 from exo.shared.models.model_cards import (
-    MODEL_CARDS,
     ModelCard,
     ModelId,
+    get_model_cards,
 )
 from exo.shared.types.api import (
     AdvancedImageParams,
@@ -65,9 +62,7 @@ from exo.shared.types.api import (
     StartDownloadParams,
     StartDownloadResponse,
     StreamingChoiceResponse,
-    StreamOptions,
     ToolCall,
-    Usage,
 )
 from exo.shared.types.chunks import (
     ErrorChunk,
@@ -106,7 +101,6 @@ from exo.shared.types.worker.instances import Instance, InstanceId, InstanceMeta
 from exo.shared.types.worker.shards import Sharding
 from exo.utils.banner import print_startup_banner
 from exo.utils.channels import Receiver, Sender, channel
-from exo.utils.dashboard_path import find_dashboard
 from exo.utils.event_buffer import OrderedBuffer
 
 
@@ -115,9 +109,7 @@ def _format_to_content_type(image_format: Literal["png", "jpeg", "webp"] | None)
 
 
 def chunk_to_response(
-    chunk: TokenChunk | ToolCallChunk,
-    command_id: CommandId,
-    usage: Usage | None,
+    chunk: TokenChunk | ToolCallChunk, command_id: CommandId
 ) -> ChatCompletionResponse:
     return ChatCompletionResponse(
         id=command_id,
@@ -142,22 +134,9 @@ def chunk_to_response(
                 finish_reason=chunk.finish_reason,
             )
         ],
-        usage=usage,
     )
 
 
-async def resolve_model_card(model_id: ModelId) -> ModelCard:
-    if model_id in MODEL_CARDS:
-        model_card = MODEL_CARDS[model_id]
-        return model_card
-
-    for card in MODEL_CARDS.values():
-        if card.model_id == ModelId(model_id):
-            return card
-
-    return await ModelCard.from_hf(model_id)
-
-
 class API:
     def __init__(
         self,
@@ -195,7 +174,7 @@ class API:
         self.app.mount(
             "/",
             StaticFiles(
-                directory=find_dashboard(),
+                directory=DASHBOARD_DIR,
                 html=True,
             ),
             name="dashboard",
@@ -279,7 +258,7 @@ class API:
 
     async def place_instance(self, payload: PlaceInstanceParams):
         command = PlaceInstance(
-            model_card=await resolve_model_card(payload.model_id),
+            model_card=await ModelCard.load(payload.model_id),
             sharding=payload.sharding,
             instance_meta=payload.instance_meta,
             min_nodes=payload.min_nodes,
@@ -296,7 +275,7 @@ class API:
         self, payload: CreateInstanceParams
     ) -> CreateInstanceResponse:
         instance = payload.instance
-        model_card = await resolve_model_card(instance.shard_assignments.model_id)
+        model_card = await ModelCard.load(instance.shard_assignments.model_id)
         required_memory = model_card.storage_size
         available_memory = self._calculate_total_available_memory()
 
@@ -324,7 +303,7 @@ class API:
         instance_meta: InstanceMeta = InstanceMeta.MlxRing,
         min_nodes: int = 1,
     ) -> Instance:
-        model_card = await resolve_model_card(model_id)
+        model_card = await ModelCard.load(model_id)
 
         try:
             placements = get_instance_placements(
@@ -366,10 +345,7 @@ class API:
         if len(list(self.state.topology.list_nodes())) == 0:
             return PlacementPreviewResponse(previews=[])
 
-        cards = [card for card in MODEL_CARDS.values() if card.model_id == model_id]
-        if not cards:
-            raise HTTPException(status_code=404, detail=f"Model {model_id} not found")
-
+        model_card = await ModelCard.load(model_id)
         instance_combinations: list[tuple[Sharding, InstanceMeta, int]] = []
         for sharding in (Sharding.Pipeline, Sharding.Tensor):
             for instance_meta in (InstanceMeta.MlxRing, InstanceMeta.MlxJaccl):
@@ -384,96 +360,93 @@ class API:
         # TODO: PDD
         # instance_combinations.append((Sharding.PrefillDecodeDisaggregation, InstanceMeta.MlxRing, 1))
 
-        for model_card in cards:
-            for sharding, instance_meta, min_nodes in instance_combinations:
-                try:
-                    placements = get_instance_placements(
-                        PlaceInstance(
-                            model_card=model_card,
-                            sharding=sharding,
-                            instance_meta=instance_meta,
-                            min_nodes=min_nodes,
-                        ),
-                        node_memory=self.state.node_memory,
-                        node_network=self.state.node_network,
-                        topology=self.state.topology,
-                        current_instances=self.state.instances,
-                        required_nodes=required_nodes,
-                    )
-                except ValueError as exc:
-                    if (model_card.model_id, sharding, instance_meta, 0) not in seen:
-                        previews.append(
-                            PlacementPreview(
-                                model_id=model_card.model_id,
-                                sharding=sharding,
-                                instance_meta=instance_meta,
-                                instance=None,
-                                error=str(exc),
-                            )
-                        )
-                    seen.add((model_card.model_id, sharding, instance_meta, 0))
-                    continue
-
-                current_ids = set(self.state.instances.keys())
-                new_instances = [
-                    instance
-                    for instance_id, instance in placements.items()
-                    if instance_id not in current_ids
-                ]
-
-                if len(new_instances) != 1:
-                    if (model_card.model_id, sharding, instance_meta, 0) not in seen:
-                        previews.append(
-                            PlacementPreview(
-                                model_id=model_card.model_id,
-                                sharding=sharding,
-                                instance_meta=instance_meta,
-                                instance=None,
-                                error="Expected exactly one new instance from placement",
-                            )
-                        )
-                    seen.add((model_card.model_id, sharding, instance_meta, 0))
-                    continue
-
-                instance = new_instances[0]
-                shard_assignments = instance.shard_assignments
-                placement_node_ids = list(shard_assignments.node_to_runner.keys())
-
-                memory_delta_by_node: dict[str, int] = {}
-                if placement_node_ids:
-                    total_bytes = model_card.storage_size.in_bytes
-                    per_node = total_bytes // len(placement_node_ids)
-                    remainder = total_bytes % len(placement_node_ids)
-                    for index, node_id in enumerate(
-                        sorted(placement_node_ids, key=str)
-                    ):
-                        extra = 1 if index < remainder else 0
-                        memory_delta_by_node[str(node_id)] = per_node + extra
-
-                if (
-                    model_card.model_id,
-                    sharding,
-                    instance_meta,
-                    len(placement_node_ids),
-                ) not in seen:
+        for sharding, instance_meta, min_nodes in instance_combinations:
+            try:
+                placements = get_instance_placements(
+                    PlaceInstance(
+                        model_card=model_card,
+                        sharding=sharding,
+                        instance_meta=instance_meta,
+                        min_nodes=min_nodes,
+                    ),
+                    node_memory=self.state.node_memory,
+                    node_network=self.state.node_network,
+                    topology=self.state.topology,
+                    current_instances=self.state.instances,
+                    required_nodes=required_nodes,
+                )
+            except ValueError as exc:
+                if (model_card.model_id, sharding, instance_meta, 0) not in seen:
                     previews.append(
                         PlacementPreview(
                             model_id=model_card.model_id,
                             sharding=sharding,
                             instance_meta=instance_meta,
-                            instance=instance,
-                            memory_delta_by_node=memory_delta_by_node or None,
-                            error=None,
+                            instance=None,
+                            error=str(exc),
                         )
                     )
-                seen.add(
-                    (
-                        model_card.model_id,
-                        sharding,
-                        instance_meta,
-                        len(placement_node_ids),
+                seen.add((model_card.model_id, sharding, instance_meta, 0))
+                continue
+
+            current_ids = set(self.state.instances.keys())
+            new_instances = [
+                instance
+                for instance_id, instance in placements.items()
+                if instance_id not in current_ids
+            ]
+
+            if len(new_instances) != 1:
+                if (model_card.model_id, sharding, instance_meta, 0) not in seen:
+                    previews.append(
+                        PlacementPreview(
+                            model_id=model_card.model_id,
+                            sharding=sharding,
+                            instance_meta=instance_meta,
+                            instance=None,
+                            error="Expected exactly one new instance from placement",
+                        )
+                    )
+                seen.add((model_card.model_id, sharding, instance_meta, 0))
+                continue
+
+            instance = new_instances[0]
+            shard_assignments = instance.shard_assignments
+            placement_node_ids = list(shard_assignments.node_to_runner.keys())
+
+            memory_delta_by_node: dict[str, int] = {}
+            if placement_node_ids:
+                total_bytes = model_card.storage_size.in_bytes
+                per_node = total_bytes // len(placement_node_ids)
+                remainder = total_bytes % len(placement_node_ids)
+                for index, node_id in enumerate(sorted(placement_node_ids, key=str)):
+                    extra = 1 if index < remainder else 0
+                    memory_delta_by_node[str(node_id)] = per_node + extra
+
+            if (
+                model_card.model_id,
+                sharding,
+                instance_meta,
+                len(placement_node_ids),
+            ) not in seen:
+                previews.append(
+                    PlacementPreview(
+                        model_id=model_card.model_id,
+                        sharding=sharding,
+                        instance_meta=instance_meta,
+                        instance=instance,
+                        memory_delta_by_node=memory_delta_by_node or None,
+                        error=None,
                     )
                 )
+            seen.add(
+                (
+                    model_card.model_id,
+                    sharding,
+                    instance_meta,
+                    len(placement_node_ids),
+                )
+            )
 
         return PlacementPreviewResponse(previews=previews)
 
@@ -527,10 +500,9 @@ class API:
                 del self._chat_completion_queues[command_id]
 
     async def _generate_chat_stream(
-        self, command_id: CommandId, stream_options: StreamOptions | None = None
+        self, command_id: CommandId
     ) -> AsyncGenerator[str, None]:
         """Generate chat completion stream as JSON strings."""
-        include_usage = stream_options.include_usage if stream_options else False
 
         async for chunk in self._chat_chunk_stream(command_id):
             assert not isinstance(chunk, ImageChunk)
@@ -546,10 +518,8 @@ class API:
                 yield "data: [DONE]\n\n"
                 return
 
-            usage = chunk.usage if include_usage else None
-
             chunk_response: ChatCompletionResponse = chunk_to_response(
-                chunk, command_id, usage=usage
+                chunk, command_id
             )
             logger.debug(f"chunk_response: {chunk_response}")
 
@@ -567,7 +537,6 @@ class API:
         tool_calls: list[ToolCall] = []
         model: str | None = None
         finish_reason: FinishReason | None = None
-        usage: Usage | None = None
 
         async for chunk in self._chat_chunk_stream(command_id):
             if isinstance(chunk, ErrorChunk):
@@ -592,9 +561,6 @@ class API:
                     for i, tool in enumerate(chunk.tool_calls)
                 )
 
-            if chunk.usage is not None:
-                usage = chunk.usage
-
             if chunk.finish_reason is not None:
                 finish_reason = chunk.finish_reason
 
@@ -616,7 +582,6 @@ class API:
                     finish_reason=finish_reason,
                 )
             ],
-            usage=usage,
         )
 
     async def _collect_chat_completion_with_stats(
@@ -686,7 +651,7 @@ class API:
         self, payload: ChatCompletionTaskParams
     ) -> ChatCompletionResponse | StreamingResponse:
         """Handle chat completions, supporting both streaming and non-streaming responses."""
-        model_card = await resolve_model_card(ModelId(payload.model))
+        model_card = await ModelCard.load(ModelId(payload.model))
         payload.model = model_card.model_id
 
         if not any(
@@ -704,7 +669,7 @@ class API:
         await self._send(command)
         if payload.stream:
             return StreamingResponse(
-                self._generate_chat_stream(command.command_id, payload.stream_options),
+                self._generate_chat_stream(command.command_id),
                 media_type="text/event-stream",
             )
 
@@ -713,7 +678,7 @@ class API:
     async def bench_chat_completions(
         self, payload: BenchChatCompletionTaskParams
     ) -> BenchChatCompletionResponse:
-        model_card = await resolve_model_card(ModelId(payload.model))
+        model_card = await ModelCard.load(ModelId(payload.model))
         payload.model = model_card.model_id
 
         if not any(
@@ -738,7 +703,7 @@ class API:
 
         Raises HTTPException 404 if no instance is found for the model.
         """
-        model_card = await resolve_model_card(ModelId(model))
+        model_card = await ModelCard.load(ModelId(model))
         resolved_model = model_card.model_id
         if not any(
             instance.shard_assignments.model_id == resolved_model
@@ -1244,7 +1209,7 @@ class API:
                     supports_tensor=card.supports_tensor,
                     tasks=[task.value for task in card.tasks],
                 )
-                for card in MODEL_CARDS.values()
+                for card in await get_model_cards()
             ]
         )
 

src/exo/shared/constants.py

@@ -2,6 +2,8 @@ import os
 import sys
 from pathlib import Path
 
+from exo.utils.dashboard_path import find_dashboard, find_resources
+
 _EXO_HOME_ENV = os.environ.get("EXO_HOME", None)
 
 
@@ -31,6 +33,14 @@ EXO_MODELS_DIR = (
     if _EXO_MODELS_DIR_ENV is None
     else Path.home() / _EXO_MODELS_DIR_ENV
 )
+_RESOURCES_DIR_ENV = os.environ.get("EXO_RESOURCES_DIR", None)
+RESOURCES_DIR = (
+    find_resources() if _RESOURCES_DIR_ENV is None else Path.home() / _RESOURCES_DIR_ENV
+)
+_DASHBOARD_DIR_ENV = os.environ.get("EXO_DASHBOARD_DIR", None)
+DASHBOARD_DIR = (
+    find_dashboard() if _RESOURCES_DIR_ENV is None else Path.home() / _RESOURCES_DIR_ENV
+)
 
 # Log files (data/logs or cache)
 EXO_LOG = EXO_CACHE_HOME / "exo.log"

src/exo/shared/models/model_cards.py

@@ -12,16 +12,42 @@ from pydantic import (
     BaseModel,
     Field,
     PositiveInt,
+    ValidationError,
     field_validator,
     model_validator,
 )
+from tomlkit.exceptions import TOMLKitError
 
-from exo.shared.constants import EXO_ENABLE_IMAGE_MODELS
+from exo.shared.constants import EXO_ENABLE_IMAGE_MODELS, RESOURCES_DIR
 from exo.shared.types.common import ModelId
 from exo.shared.types.memory import Memory
 from exo.utils.pydantic_ext import CamelCaseModel
 
-_card_cache: dict[str, "ModelCard"] = {}
+# kinda ugly...
+# TODO: load search path from config.toml
+_csp = [Path(RESOURCES_DIR)]
+if EXO_ENABLE_IMAGE_MODELS:
+    _csp.append(Path(RESOURCES_DIR) / "image_models")
+
+CARD_SEARCH_PATH = _csp
+
+_card_cache: dict[ModelId, "ModelCard"] = {}
+
+
+async def _populate_card_cache():
+    for path in CARD_SEARCH_PATH:
+        async for toml_file in path.rglob("*.toml"):
+            try:
+                card = await ModelCard.load_from_path(toml_file)
+                _card_cache[card.model_id] = card
+            except (ValidationError, TOMLKitError):
+                pass
+
+
+async def get_model_cards() -> list["ModelCard"]:
+    if len(_card_cache) == 0:
+        await _populate_card_cache()
+    return list(_card_cache.values())
 
 
 class ModelTask(str, Enum):
@@ -55,28 +81,37 @@ class ModelCard(CamelCaseModel):
 
     async def save(self, path: Path) -> None:
         async with await open_file(path, "w") as f:
-            py = self.model_dump()
+            py = self.model_dump(exclude_none=True)
             data = tomlkit.dumps(py)  # pyright: ignore[reportUnknownMemberType]
             await f.write(data)
 
+    async def save_to_default_path(self):
+        await self.save(Path(RESOURCES_DIR) / (self.model_id.normalize() + ".toml"))
+
     @staticmethod
     async def load_from_path(path: Path) -> "ModelCard":
         async with await open_file(path, "r") as f:
             py = tomlkit.loads(await f.read())
             return ModelCard.model_validate(py)
 
+    # Is it okay that model card.load defaults to network access if the card doesn't exist? do we want to be more explicit here?
     @staticmethod
     async def load(model_id: ModelId) -> "ModelCard":
-        for card in MODEL_CARDS.values():
-            if card.model_id == model_id:
-                return card
-        return await ModelCard.from_hf(model_id)
-
-    @staticmethod
-    async def from_hf(model_id: ModelId) -> "ModelCard":
-        """Fetches storage size and number of layers for a Hugging Face model, returns Pydantic ModelMeta."""
+        if len(_card_cache) == 0:
+            await _populate_card_cache()
         if (mc := _card_cache.get(model_id)) is not None:
             return mc
+
+        return await ModelCard.fetch_from_hf(model_id)
+
+    @staticmethod
+    async def fetch_from_hf(model_id: ModelId) -> "ModelCard":
+        """Fetches storage size and number of layers for a Hugging Face model, returns Pydantic ModelMeta."""
+        if len(_card_cache) == 0:
+            await _populate_card_cache()
+        if (mc := _card_cache.get(model_id)) is not None:
+            return mc
+        # TODO: failure if files do not exist
         config_data = await get_config_data(model_id)
         num_layers = config_data.layer_count
         mem_size_bytes = await get_safetensors_size(model_id)
@@ -89,544 +124,13 @@ class ModelCard(CamelCaseModel):
             supports_tensor=config_data.supports_tensor,
             tasks=[ModelTask.TextGeneration],
         )
+        await mc.save_to_default_path()
         _card_cache[model_id] = mc
         return mc
 
 
-MODEL_CARDS: dict[str, ModelCard] = {
-    # deepseek v3
-    "deepseek-v3.1-4bit": ModelCard(
-        model_id=ModelId("mlx-community/DeepSeek-V3.1-4bit"),
-        storage_size=Memory.from_gb(378),
-        n_layers=61,
-        hidden_size=7168,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "deepseek-v3.1-8bit": ModelCard(
-        model_id=ModelId("mlx-community/DeepSeek-V3.1-8bit"),
-        storage_size=Memory.from_gb(713),
-        n_layers=61,
-        hidden_size=7168,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    # kimi k2
-    "kimi-k2-instruct-4bit": ModelCard(
-        model_id=ModelId("mlx-community/Kimi-K2-Instruct-4bit"),
-        storage_size=Memory.from_gb(578),
-        n_layers=61,
-        hidden_size=7168,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "kimi-k2-thinking": ModelCard(
-        model_id=ModelId("mlx-community/Kimi-K2-Thinking"),
-        storage_size=Memory.from_gb(658),
-        n_layers=61,
-        hidden_size=7168,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "kimi-k2.5": ModelCard(
-        model_id=ModelId("mlx-community/Kimi-K2.5"),
-        storage_size=Memory.from_gb(617),
-        n_layers=61,
-        hidden_size=7168,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    # llama-3.1
-    "llama-3.1-8b": ModelCard(
-        model_id=ModelId("mlx-community/Meta-Llama-3.1-8B-Instruct-4bit"),
-        storage_size=Memory.from_mb(4423),
-        n_layers=32,
-        hidden_size=4096,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "llama-3.1-8b-8bit": ModelCard(
-        model_id=ModelId("mlx-community/Meta-Llama-3.1-8B-Instruct-8bit"),
-        storage_size=Memory.from_mb(8540),
-        n_layers=32,
-        hidden_size=4096,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "llama-3.1-8b-bf16": ModelCard(
-        model_id=ModelId("mlx-community/Meta-Llama-3.1-8B-Instruct-bf16"),
-        storage_size=Memory.from_mb(16100),
-        n_layers=32,
-        hidden_size=4096,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "llama-3.1-70b": ModelCard(
-        model_id=ModelId("mlx-community/Meta-Llama-3.1-70B-Instruct-4bit"),
-        storage_size=Memory.from_mb(38769),
-        n_layers=80,
-        hidden_size=8192,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    # llama-3.2
-    "llama-3.2-1b": ModelCard(
-        model_id=ModelId("mlx-community/Llama-3.2-1B-Instruct-4bit"),
-        storage_size=Memory.from_mb(696),
-        n_layers=16,
-        hidden_size=2048,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "llama-3.2-3b": ModelCard(
-        model_id=ModelId("mlx-community/Llama-3.2-3B-Instruct-4bit"),
-        storage_size=Memory.from_mb(1777),
-        n_layers=28,
-        hidden_size=3072,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "llama-3.2-3b-8bit": ModelCard(
-        model_id=ModelId("mlx-community/Llama-3.2-3B-Instruct-8bit"),
-        storage_size=Memory.from_mb(3339),
-        n_layers=28,
-        hidden_size=3072,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    # llama-3.3
-    "llama-3.3-70b": ModelCard(
-        model_id=ModelId("mlx-community/Llama-3.3-70B-Instruct-4bit"),
-        storage_size=Memory.from_mb(38769),
-        n_layers=80,
-        hidden_size=8192,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "llama-3.3-70b-8bit": ModelCard(
-        model_id=ModelId("mlx-community/Llama-3.3-70B-Instruct-8bit"),
-        storage_size=Memory.from_mb(73242),
-        n_layers=80,
-        hidden_size=8192,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "llama-3.3-70b-fp16": ModelCard(
-        model_id=ModelId("mlx-community/llama-3.3-70b-instruct-fp16"),
-        storage_size=Memory.from_mb(137695),
-        n_layers=80,
-        hidden_size=8192,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    # qwen3
-    "qwen3-0.6b": ModelCard(
-        model_id=ModelId("mlx-community/Qwen3-0.6B-4bit"),
-        storage_size=Memory.from_mb(327),
-        n_layers=28,
-        hidden_size=1024,
-        supports_tensor=False,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "qwen3-0.6b-8bit": ModelCard(
-        model_id=ModelId("mlx-community/Qwen3-0.6B-8bit"),
-        storage_size=Memory.from_mb(666),
-        n_layers=28,
-        hidden_size=1024,
-        supports_tensor=False,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "qwen3-30b": ModelCard(
-        model_id=ModelId("mlx-community/Qwen3-30B-A3B-4bit"),
-        storage_size=Memory.from_mb(16797),
-        n_layers=48,
-        hidden_size=2048,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "qwen3-30b-8bit": ModelCard(
-        model_id=ModelId("mlx-community/Qwen3-30B-A3B-8bit"),
-        storage_size=Memory.from_mb(31738),
-        n_layers=48,
-        hidden_size=2048,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "qwen3-80b-a3B-4bit": ModelCard(
-        model_id=ModelId("mlx-community/Qwen3-Next-80B-A3B-Instruct-4bit"),
-        storage_size=Memory.from_mb(44800),
-        n_layers=48,
-        hidden_size=2048,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "qwen3-80b-a3B-8bit": ModelCard(
-        model_id=ModelId("mlx-community/Qwen3-Next-80B-A3B-Instruct-8bit"),
-        storage_size=Memory.from_mb(84700),
-        n_layers=48,
-        hidden_size=2048,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "qwen3-80b-a3B-thinking-4bit": ModelCard(
-        model_id=ModelId("mlx-community/Qwen3-Next-80B-A3B-Thinking-4bit"),
-        storage_size=Memory.from_mb(84700),
-        n_layers=48,
-        hidden_size=2048,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "qwen3-80b-a3B-thinking-8bit": ModelCard(
-        model_id=ModelId("mlx-community/Qwen3-Next-80B-A3B-Thinking-8bit"),
-        storage_size=Memory.from_mb(84700),
-        n_layers=48,
-        hidden_size=2048,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "qwen3-235b-a22b-4bit": ModelCard(
-        model_id=ModelId("mlx-community/Qwen3-235B-A22B-Instruct-2507-4bit"),
-        storage_size=Memory.from_gb(132),
-        n_layers=94,
-        hidden_size=4096,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "qwen3-235b-a22b-8bit": ModelCard(
-        model_id=ModelId("mlx-community/Qwen3-235B-A22B-Instruct-2507-8bit"),
-        storage_size=Memory.from_gb(250),
-        n_layers=94,
-        hidden_size=4096,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "qwen3-coder-480b-a35b-4bit": ModelCard(
-        model_id=ModelId("mlx-community/Qwen3-Coder-480B-A35B-Instruct-4bit"),
-        storage_size=Memory.from_gb(270),
-        n_layers=62,
-        hidden_size=6144,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "qwen3-coder-480b-a35b-8bit": ModelCard(
-        model_id=ModelId("mlx-community/Qwen3-Coder-480B-A35B-Instruct-8bit"),
-        storage_size=Memory.from_gb(540),
-        n_layers=62,
-        hidden_size=6144,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    # gpt-oss
-    "gpt-oss-120b-MXFP4-Q8": ModelCard(
-        model_id=ModelId("mlx-community/gpt-oss-120b-MXFP4-Q8"),
-        storage_size=Memory.from_kb(68_996_301),
-        n_layers=36,
-        hidden_size=2880,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "gpt-oss-20b-MXFP4-Q8": ModelCard(
-        model_id=ModelId("mlx-community/gpt-oss-20b-MXFP4-Q8"),
-        storage_size=Memory.from_kb(11_744_051),
-        n_layers=24,
-        hidden_size=2880,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    # glm 4.5
-    "glm-4.5-air-8bit": ModelCard(
-        # Needs to be quantized g32 or g16 to work with tensor parallel
-        model_id=ModelId("mlx-community/GLM-4.5-Air-8bit"),
-        storage_size=Memory.from_gb(114),
-        n_layers=46,
-        hidden_size=4096,
-        supports_tensor=False,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "glm-4.5-air-bf16": ModelCard(
-        model_id=ModelId("mlx-community/GLM-4.5-Air-bf16"),
-        storage_size=Memory.from_gb(214),
-        n_layers=46,
-        hidden_size=4096,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    # glm 4.7
-    "glm-4.7-4bit": ModelCard(
-        model_id=ModelId("mlx-community/GLM-4.7-4bit"),
-        storage_size=Memory.from_bytes(198556925568),
-        n_layers=91,
-        hidden_size=5120,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "glm-4.7-6bit": ModelCard(
-        model_id=ModelId("mlx-community/GLM-4.7-6bit"),
-        storage_size=Memory.from_bytes(286737579648),
-        n_layers=91,
-        hidden_size=5120,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "glm-4.7-8bit-gs32": ModelCard(
-        model_id=ModelId("mlx-community/GLM-4.7-8bit-gs32"),
-        storage_size=Memory.from_bytes(396963397248),
-        n_layers=91,
-        hidden_size=5120,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    # glm 4.7 flash
-    "glm-4.7-flash-4bit": ModelCard(
-        model_id=ModelId("mlx-community/GLM-4.7-Flash-4bit"),
-        storage_size=Memory.from_gb(18),
-        n_layers=47,
-        hidden_size=2048,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "glm-4.7-flash-5bit": ModelCard(
-        model_id=ModelId("mlx-community/GLM-4.7-Flash-5bit"),
-        storage_size=Memory.from_gb(21),
-        n_layers=47,
-        hidden_size=2048,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "glm-4.7-flash-6bit": ModelCard(
-        model_id=ModelId("mlx-community/GLM-4.7-Flash-6bit"),
-        storage_size=Memory.from_gb(25),
-        n_layers=47,
-        hidden_size=2048,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "glm-4.7-flash-8bit": ModelCard(
-        model_id=ModelId("mlx-community/GLM-4.7-Flash-8bit"),
-        storage_size=Memory.from_gb(32),
-        n_layers=47,
-        hidden_size=2048,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    # minimax-m2
-    "minimax-m2.1-8bit": ModelCard(
-        model_id=ModelId("mlx-community/MiniMax-M2.1-8bit"),
-        storage_size=Memory.from_bytes(242986745856),
-        n_layers=61,
-        hidden_size=3072,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-    "minimax-m2.1-3bit": ModelCard(
-        model_id=ModelId("mlx-community/MiniMax-M2.1-3bit"),
-        storage_size=Memory.from_bytes(100086644736),
-        n_layers=61,
-        hidden_size=3072,
-        supports_tensor=True,
-        tasks=[ModelTask.TextGeneration],
-    ),
-}
-
-_IMAGE_BASE_MODEL_CARDS: dict[str, ModelCard] = {
-    "flux1-schnell": ModelCard(
-        model_id=ModelId("exolabs/FLUX.1-schnell"),
-        storage_size=Memory.from_bytes(23782357120 + 9524621312),
-        n_layers=57,
-        hidden_size=1,
-        supports_tensor=False,
-        tasks=[ModelTask.TextToImage],
-        components=[
-            ComponentInfo(
-                component_name="text_encoder",
-                component_path="text_encoder/",
-                storage_size=Memory.from_kb(0),
-                n_layers=12,
-                can_shard=False,
-                safetensors_index_filename=None,
-            ),
-            ComponentInfo(
-                component_name="text_encoder_2",
-                component_path="text_encoder_2/",
-                storage_size=Memory.from_bytes(9524621312),
-                n_layers=24,
-                can_shard=False,
-                safetensors_index_filename="model.safetensors.index.json",
-            ),
-            ComponentInfo(
-                component_name="transformer",
-                component_path="transformer/",
-                storage_size=Memory.from_bytes(23782357120),
-                n_layers=57,
-                can_shard=True,
-                safetensors_index_filename="diffusion_pytorch_model.safetensors.index.json",
-            ),
-            ComponentInfo(
-                component_name="vae",
-                component_path="vae/",
-                storage_size=Memory.from_kb(0),
-                n_layers=None,
-                can_shard=False,
-                safetensors_index_filename=None,
-            ),
-        ],
-    ),
-    "flux1-dev": ModelCard(
-        model_id=ModelId("exolabs/FLUX.1-dev"),
-        storage_size=Memory.from_bytes(23782357120 + 9524621312),
-        n_layers=57,
-        hidden_size=1,
-        supports_tensor=False,
-        tasks=[ModelTask.TextToImage],
-        components=[
-            ComponentInfo(
-                component_name="text_encoder",
-                component_path="text_encoder/",
-                storage_size=Memory.from_kb(0),
-                n_layers=12,
-                can_shard=False,
-                safetensors_index_filename=None,
-            ),
-            ComponentInfo(
-                component_name="text_encoder_2",
-                component_path="text_encoder_2/",
-                storage_size=Memory.from_bytes(9524621312),
-                n_layers=24,
-                can_shard=False,
-                safetensors_index_filename="model.safetensors.index.json",
-            ),
-            ComponentInfo(
-                component_name="transformer",
-                component_path="transformer/",
-                storage_size=Memory.from_bytes(23802816640),
-                n_layers=57,
-                can_shard=True,
-                safetensors_index_filename="diffusion_pytorch_model.safetensors.index.json",
-            ),
-            ComponentInfo(
-                component_name="vae",
-                component_path="vae/",
-                storage_size=Memory.from_kb(0),
-                n_layers=None,
-                can_shard=False,
-                safetensors_index_filename=None,
-            ),
-        ],
-    ),
-    "flux1-krea-dev": ModelCard(
-        model_id=ModelId("exolabs/FLUX.1-Krea-dev"),
-        storage_size=Memory.from_bytes(23802816640 + 9524621312),  # Same as dev
-        n_layers=57,
-        hidden_size=1,
-        supports_tensor=False,
-        tasks=[ModelTask.TextToImage],
-        components=[
-            ComponentInfo(
-                component_name="text_encoder",
-                component_path="text_encoder/",
-                storage_size=Memory.from_kb(0),
-                n_layers=12,
-                can_shard=False,
-                safetensors_index_filename=None,
-            ),
-            ComponentInfo(
-                component_name="text_encoder_2",
-                component_path="text_encoder_2/",
-                storage_size=Memory.from_bytes(9524621312),
-                n_layers=24,
-                can_shard=False,
-                safetensors_index_filename="model.safetensors.index.json",
-            ),
-            ComponentInfo(
-                component_name="transformer",
-                component_path="transformer/",
-                storage_size=Memory.from_bytes(23802816640),
-                n_layers=57,
-                can_shard=True,
-                safetensors_index_filename="diffusion_pytorch_model.safetensors.index.json",
-            ),
-            ComponentInfo(
-                component_name="vae",
-                component_path="vae/",
-                storage_size=Memory.from_kb(0),
-                n_layers=None,
-                can_shard=False,
-                safetensors_index_filename=None,
-            ),
-        ],
-    ),
-    "qwen-image": ModelCard(
-        model_id=ModelId("exolabs/Qwen-Image"),
-        storage_size=Memory.from_bytes(16584333312 + 40860802176),
-        n_layers=60,
-        hidden_size=1,
-        supports_tensor=False,
-        tasks=[ModelTask.TextToImage],
-        components=[
-            ComponentInfo(
-                component_name="text_encoder",
-                component_path="text_encoder/",
-                storage_size=Memory.from_bytes(16584333312),
-                n_layers=12,
-                can_shard=False,
-                safetensors_index_filename=None,
-            ),
-            ComponentInfo(
-                component_name="transformer",
-                component_path="transformer/",
-                storage_size=Memory.from_bytes(40860802176),
-                n_layers=60,
-                can_shard=True,
-                safetensors_index_filename="diffusion_pytorch_model.safetensors.index.json",
-            ),
-            ComponentInfo(
-                component_name="vae",
-                component_path="vae/",
-                storage_size=Memory.from_kb(0),
-                n_layers=None,
-                can_shard=False,
-                safetensors_index_filename=None,
-            ),
-        ],
-    ),
-    "qwen-image-edit-2509": ModelCard(
-        model_id=ModelId("exolabs/Qwen-Image-Edit-2509"),
-        storage_size=Memory.from_bytes(16584333312 + 40860802176),
-        n_layers=60,
-        hidden_size=1,
-        supports_tensor=False,
-        tasks=[ModelTask.ImageToImage],
-        components=[
-            ComponentInfo(
-                component_name="text_encoder",
-                component_path="text_encoder/",
-                storage_size=Memory.from_bytes(16584333312),
-                n_layers=12,
-                can_shard=False,
-                safetensors_index_filename=None,
-            ),
-            ComponentInfo(
-                component_name="transformer",
-                component_path="transformer/",
-                storage_size=Memory.from_bytes(40860802176),
-                n_layers=60,
-                can_shard=True,
-                safetensors_index_filename="diffusion_pytorch_model.safetensors.index.json",
-            ),
-            ComponentInfo(
-                component_name="vae",
-                component_path="vae/",
-                storage_size=Memory.from_kb(0),
-                n_layers=None,
-                can_shard=False,
-                safetensors_index_filename=None,
-            ),
-        ],
-    ),
-}
-
-
-def _generate_image_model_quant_variants(
+# TODO: quantizing and dynamically creating model cards
+def _generate_image_model_quant_variants(  # pyright: ignore[reportUnusedFunction]
     base_name: str,
     base_card: ModelCard,
 ) -> dict[str, ModelCard]:
@@ -706,15 +210,6 @@ def _generate_image_model_quant_variants(
     return variants
 
 
-_image_model_cards: dict[str, ModelCard] = {}
-for _base_name, _base_card in _IMAGE_BASE_MODEL_CARDS.items():
-    _image_model_cards |= _generate_image_model_quant_variants(_base_name, _base_card)
-_IMAGE_MODEL_CARDS = _image_model_cards
-
-if EXO_ENABLE_IMAGE_MODELS:
-    MODEL_CARDS.update(_IMAGE_MODEL_CARDS)
-
-
 class ConfigData(BaseModel):
     model_config = {"extra": "ignore"}  # Allow unknown fields
 

src/exo/shared/types/api.py

@@ -11,7 +11,7 @@ from exo.shared.types.common import CommandId, NodeId
 from exo.shared.types.memory import Memory
 from exo.shared.types.worker.instances import Instance, InstanceId, InstanceMeta
 from exo.shared.types.worker.shards import Sharding, ShardMetadata
-from exo.utils.pydantic_ext import CamelCaseModel, ConfigDict, TaggedModel
+from exo.utils.pydantic_ext import CamelCaseModel
 
 FinishReason = Literal[
     "stop", "length", "tool_calls", "content_filter", "function_call", "error"
@@ -116,8 +116,8 @@ class Usage(BaseModel):
     prompt_tokens: int
     completion_tokens: int
     total_tokens: int
-    prompt_tokens_details: PromptTokensDetails
-    completion_tokens_details: CompletionTokensDetails
+    prompt_tokens_details: PromptTokensDetails | None = None
+    completion_tokens_details: CompletionTokensDetails | None = None
 
 
 class StreamingChoiceResponse(BaseModel):
@@ -170,13 +170,7 @@ class BenchChatCompletionResponse(ChatCompletionResponse):
     generation_stats: GenerationStats | None = None
 
 
-class StreamOptions(BaseModel):
-    include_usage: bool = False
-
-
-class ChatCompletionTaskParams(TaggedModel):
-    model_config = ConfigDict(extra="ignore")
-
+class ChatCompletionTaskParams(BaseModel):
     model: str
     frequency_penalty: float | None = None
     messages: list[ChatCompletionMessage]
@@ -190,7 +184,6 @@ class ChatCompletionTaskParams(TaggedModel):
     seed: int | None = None
     stop: str | list[str] | None = None
     stream: bool = False
-    stream_options: StreamOptions | None = None
     temperature: float | None = None
     top_p: float | None = None
     tools: list[dict[str, Any]] | None = None

src/exo/shared/types/chunks.py

@@ -2,7 +2,7 @@ from collections.abc import Generator
 from typing import Any, Literal
 
 from exo.shared.models.model_cards import ModelId
-from exo.shared.types.api import GenerationStats, ImageGenerationStats, Usage
+from exo.shared.types.api import GenerationStats, ImageGenerationStats
 from exo.utils.pydantic_ext import TaggedModel
 
 from .api import FinishReason
@@ -17,7 +17,6 @@ class BaseChunk(TaggedModel):
 class TokenChunk(BaseChunk):
     text: str
     token_id: int
-    usage: Usage | None
     finish_reason: Literal["stop", "length", "content_filter"] | None = None
     stats: GenerationStats | None = None
 
@@ -29,7 +28,6 @@ class ErrorChunk(BaseChunk):
 
 class ToolCallChunk(BaseChunk):
     tool_calls: list[ToolCallItem]
-    usage: Usage | None
     finish_reason: Literal["tool_calls"] = "tool_calls"
     stats: GenerationStats | None = None
 

src/exo/shared/types/commands.py

@@ -2,7 +2,6 @@ from pydantic import Field
 
 from exo.shared.models.model_cards import ModelCard, ModelId
 from exo.shared.types.api import (
-    BenchChatCompletionTaskParams,
     ChatCompletionTaskParams,
     ImageEditsInternalParams,
     ImageGenerationTaskParams,
@@ -23,7 +22,7 @@ class TestCommand(BaseCommand):
 
 
 class ChatCompletion(BaseCommand):
-    request_params: ChatCompletionTaskParams | BenchChatCompletionTaskParams
+    request_params: ChatCompletionTaskParams
 
 
 class ImageGeneration(BaseCommand):

src/exo/shared/types/tasks.py

@@ -3,7 +3,6 @@ from enum import Enum
 from pydantic import Field
 
 from exo.shared.types.api import (
-    BenchChatCompletionTaskParams,
     ChatCompletionTaskParams,
     ImageEditsInternalParams,
     ImageGenerationTaskParams,
@@ -55,7 +54,7 @@ class StartWarmup(BaseTask):  # emitted by Worker
 
 class ChatCompletion(BaseTask):  # emitted by Master
     command_id: CommandId
-    task_params: ChatCompletionTaskParams | BenchChatCompletionTaskParams
+    task_params: ChatCompletionTaskParams
 
     error_type: str | None = Field(default=None)
     error_message: str | None = Field(default=None)

src/exo/shared/types/worker/runner_response.py

@@ -6,7 +6,6 @@ from exo.shared.types.api import (
     GenerationStats,
     ImageGenerationStats,
     ToolCallItem,
-    Usage,
 )
 from exo.utils.pydantic_ext import TaggedModel
 
@@ -25,7 +24,6 @@ class GenerationResponse(BaseRunnerResponse):
     # logprobs: list[float] | None = None # too big. we can change to be top-k
     finish_reason: FinishReason | None = None
     stats: GenerationStats | None = None
-    usage: Usage | None
 
 
 class ImageGenerationResponse(BaseRunnerResponse):
@@ -59,7 +57,6 @@ class PartialImageResponse(BaseRunnerResponse):
 
 class ToolCallResponse(BaseRunnerResponse):
     tool_calls: list[ToolCallItem]
-    usage: Usage | None
 
 
 class FinishedResponse(BaseRunnerResponse):

src/exo/utils/dashboard_path.py

@@ -1,31 +1,45 @@
-import os
 import sys
 from pathlib import Path
 from typing import cast
 
 
+def find_resources() -> Path:
+    resources = _find_resources_in_repo() or _find_resources_in_bundle()
+    if resources is None:
+        raise FileNotFoundError(
+            "Unable to locate resources. Did you clone the repo properly?"
+        )
+    return resources
+
+
+def _find_resources_in_repo() -> Path | None:
+    current_module = Path(__file__).resolve()
+    for parent in current_module.parents:
+        build = parent / "resources"
+        if build.is_dir():
+            return build
+    return None
+
+
+def _find_resources_in_bundle() -> Path | None:
+    frozen_root = cast(str | None, getattr(sys, "_MEIPASS", None))
+    if frozen_root is None:
+        return None
+    candidate = Path(frozen_root) / "resources"
+    if candidate.is_dir():
+        return candidate
+    return None
+
+
 def find_dashboard() -> Path:
-    dashboard = (
-        _find_dashboard_in_env()
-        or _find_dashboard_in_repo()
-        or _find_dashboard_in_bundle()
-    )
+    dashboard = _find_dashboard_in_repo() or _find_dashboard_in_bundle()
     if not dashboard:
         raise FileNotFoundError(
-            "Unable to locate dashboard assets - make sure the dashboard has been built, or export DASHBOARD_DIR if you've built the dashboard elsewhere."
+            "Unable to locate dashboard assets - you probably forgot to run `cd dashboard && npm install && npm run build && cd ..`"
         )
     return dashboard
 
 
-def _find_dashboard_in_env() -> Path | None:
-    env = os.environ.get("DASHBOARD_DIR")
-    if not env:
-        return None
-    resolved_env = Path(env).expanduser().resolve()
-
-    return resolved_env
-
-
 def _find_dashboard_in_repo() -> Path | None:
     current_module = Path(__file__).resolve()
     for parent in current_module.parents:

src/exo/worker/engines/image/generate.py

@@ -98,8 +98,8 @@ def generate_image(
 
     partial_images = (
         task.partial_images
-        if task.partial_images is not None and task.stream is not None and task.stream
-        else 0
+        if task.partial_images is not None
+        else (3 if task.stream else 0)
     )
 
     image_path: Path | None = None

src/exo/worker/engines/image/pipeline/runner.py

@@ -348,7 +348,6 @@ class DiffusionRunner:
                 ctx.in_loop(  # pyright: ignore[reportAny]
                     t=t,
                     latents=latents,
-                    time_steps=time_steps,
                 )
 
                 mx.eval(latents)

src/exo/worker/engines/mlx/auto_parallel.py

@@ -201,9 +201,6 @@ def pipeline_auto_parallel(
     device_rank, world_size = model_shard_meta.device_rank, model_shard_meta.world_size
 
     layers = layers[start_layer:end_layer]
-    for layer in layers:
-        mx.eval(layer)  # type: ignore
-
     layers[0] = PipelineFirstLayer(layers[0], device_rank, group=group)
     layers[-1] = PipelineLastLayer(
         layers[-1],

src/exo/worker/engines/mlx/generator/generate.py

@@ -10,11 +10,8 @@ from mlx_lm.tokenizer_utils import TokenizerWrapper
 from exo.shared.types.api import (
     BenchChatCompletionTaskParams,
     ChatCompletionMessage,
-    CompletionTokensDetails,
     FinishReason,
     GenerationStats,
-    PromptTokensDetails,
-    Usage,
 )
 from exo.shared.types.memory import Memory
 from exo.shared.types.mlx import KVCacheType
@@ -42,7 +39,7 @@ def prefill(
     sampler: Callable[[mx.array], mx.array],
     prompt_tokens: mx.array,
     cache: KVCacheType,
-) -> tuple[float, int]:
+) -> float:
     """Prefill the KV cache with prompt tokens.
 
     This runs the model over the prompt tokens to populate the cache,
@@ -53,7 +50,7 @@ def prefill(
     """
     num_tokens = len(prompt_tokens)
     if num_tokens == 0:
-        return 0.0, 0
+        return 0.0
 
     logger.debug(f"Prefilling {num_tokens} tokens...")
     start_time = time.perf_counter()
@@ -88,7 +85,7 @@ def prefill(
         f"Prefill complete: {num_tokens} tokens in {elapsed:.2f}s "
         f"({tokens_per_sec:.1f} tok/s)"
     )
-    return tokens_per_sec, num_tokens
+    return tokens_per_sec
 
 
 def warmup_inference(
@@ -172,8 +169,6 @@ def mlx_generate(
     mx.reset_peak_memory()
     is_bench: bool = isinstance(task, BenchChatCompletionTaskParams)
 
-    logger.info(f"{is_bench=}")
-
     # Currently we support chat-completion tasks only.
     logger.debug(f"task_params: {task}")
 
@@ -209,9 +204,7 @@ def mlx_generate(
     )
 
     # Prefill cache with all tokens except the last one
-    prefill_tps, prefill_tokens = prefill(
-        model, tokenizer, sampler, prompt_tokens[:-1], caches
-    )
+    prefill_tps = prefill(model, tokenizer, sampler, prompt_tokens[:-1], caches)
 
     # stream_generate starts from the last token
     last_token = prompt_tokens[-1:]
@@ -219,43 +212,28 @@ def mlx_generate(
     max_tokens = task.max_tokens or MAX_TOKENS
     generated_text_parts: list[str] = []
     generation_start_time = time.perf_counter()
-    usage: Usage | None = None
-    in_thinking = False
-    reasoning_tokens = 0
-    think_start = tokenizer.think_start
-    think_end = tokenizer.think_end
-    for completion_tokens, out in enumerate(
-        stream_generate(
-            model=model,
-            tokenizer=tokenizer,
-            prompt=last_token,
-            max_tokens=max_tokens,
-            sampler=sampler,
-            logits_processors=logits_processors,
-            prompt_cache=caches,
-            # TODO: Dynamically change prefill step size to be the maximum possible without timing out.
-            prefill_step_size=2048,
-            kv_group_size=KV_GROUP_SIZE,
-            kv_bits=KV_BITS,
-        ),
-        start=1,
+    for out in stream_generate(
+        model=model,
+        tokenizer=tokenizer,
+        prompt=last_token,
+        max_tokens=max_tokens,
+        sampler=sampler,
+        logits_processors=logits_processors,
+        prompt_cache=caches,
+        # TODO: Dynamically change prefill step size to be the maximum possible without timing out.
+        prefill_step_size=2048,
+        kv_group_size=KV_GROUP_SIZE,
+        kv_bits=KV_BITS,
     ):
         generated_text_parts.append(out.text)
         logger.info(out.text)
 
-        if think_start is not None and out.text == think_start:
-            in_thinking = True
-        elif think_end is not None and out.text == think_end:
-            in_thinking = False
-        if in_thinking:
-            reasoning_tokens += 1
-
         stats: GenerationStats | None = None
         if out.finish_reason is not None:
             stats = GenerationStats(
                 prompt_tps=float(prefill_tps or out.prompt_tps),
                 generation_tps=float(out.generation_tps),
-                prompt_tokens=int(prefill_tokens + out.prompt_tokens),
+                prompt_tokens=int(out.prompt_tokens),
                 generation_tokens=int(out.generation_tokens),
                 peak_memory_usage=Memory.from_gb(out.peak_memory),
             )
@@ -267,24 +245,11 @@ def mlx_generate(
                     f"Model generated unexpected finish_reason: {out.finish_reason}"
                 )
 
-            usage = Usage(
-                prompt_tokens=int(out.prompt_tokens),
-                completion_tokens=completion_tokens,
-                total_tokens=int(out.prompt_tokens) + completion_tokens,
-                prompt_tokens_details=PromptTokensDetails(
-                    cached_tokens=prefix_hit_length
-                ),
-                completion_tokens_details=CompletionTokensDetails(
-                    reasoning_tokens=reasoning_tokens
-                ),
-            )
-
         yield GenerationResponse(
             text=out.text,
             token=out.token,
             finish_reason=cast(FinishReason | None, out.finish_reason),
             stats=stats,
-            usage=usage,
         )
 
         if out.finish_reason is not None:

src/exo/worker/runner/runner.py

@@ -277,11 +277,9 @@ def main(
                                 tokenizer.tool_parser,  # pyright: ignore[reportAny]
                             )
 
-                        completion_tokens = 0
                         for response in mlx_generator:
                             match response:
                                 case GenerationResponse():
-                                    completion_tokens += 1
                                     if (
                                         device_rank == 0
                                         and response.finish_reason == "error"
@@ -309,7 +307,6 @@ def main(
                                                     model=shard_metadata.model_card.model_id,
                                                     text=response.text,
                                                     token_id=response.token,
-                                                    usage=response.usage,
                                                     finish_reason=response.finish_reason,
                                                     stats=response.stats,
                                                 ),
@@ -323,7 +320,6 @@ def main(
                                                 chunk=ToolCallChunk(
                                                     tool_calls=response.tool_calls,
                                                     model=shard_metadata.model_card.model_id,
-                                                    usage=response.usage,
                                                 ),
                                             )
                                         )
@@ -539,10 +535,10 @@ def parse_gpt_oss(
                             name=current_tool_name,
                             arguments="".join(tool_arg_parts).strip(),
                         )
-                    ],
-                    usage=response.usage,
+                    ]
                 )
                 tool_arg_parts = []
+                break
             current_tool_name = recipient
 
         # If inside a tool call, accumulate arguments
@@ -688,7 +684,7 @@ def parse_tool_calls(
                     tools = [_validate_single_tool(tool) for tool in parsed]
                 else:
                     tools = [_validate_single_tool(parsed)]
-                yield ToolCallResponse(tool_calls=tools, usage=response.usage)
+                yield ToolCallResponse(tool_calls=tools)
 
             except (
                 json.JSONDecodeError,

src/exo/worker/tests/unittests/test_mlx/test_tokenizers.py

@@ -16,7 +16,7 @@ from exo.download.download_utils import (
     ensure_models_dir,
     fetch_file_list_with_cache,
 )
-from exo.shared.models.model_cards import MODEL_CARDS, ModelCard, ModelId
+from exo.shared.models.model_cards import ModelCard, ModelId, get_model_cards
 from exo.worker.engines.mlx.utils_mlx import (
     get_eos_token_ids_for_model,
     load_tokenizer_for_model_id,
@@ -76,7 +76,7 @@ def get_test_models() -> list[ModelCard]:
     """Get a representative sample of models to test."""
     # Pick one model from each family to test
     families: dict[str, ModelCard] = {}
-    for card in MODEL_CARDS.values():
+    for card in asyncio.run(get_model_cards()):
         # Extract family name (e.g., "llama-3.1" from "llama-3.1-8b")
         parts = card.model_id.short().split("-")
         family = "-".join(parts[:2]) if len(parts) >= 2 else parts[0]
@@ -296,7 +296,7 @@ async def test_tokenizer_special_tokens(model_card: ModelCard) -> None:
 async def test_kimi_tokenizer_specifically():
     """Test Kimi tokenizer with its specific patches and quirks."""
     kimi_models = [
-        card for card in MODEL_CARDS.values() if "kimi" in card.model_id.lower()
+        card for card in await get_model_cards() if "kimi" in card.model_id.lower()
     ]
 
     if not kimi_models:
@@ -343,7 +343,7 @@ async def test_kimi_tokenizer_specifically():
 async def test_glm_tokenizer_specifically():
     """Test GLM tokenizer with its specific EOS tokens."""
     glm_model_cards = [
-        card for card in MODEL_CARDS.values() if "glm" in card.model_id.lower()
+        card for card in await get_model_cards() if "glm" in card.model_id.lower()
     ]
 
     if not glm_model_cards:

src/exo/worker/tests/unittests/test_runner/test_event_ordering.py

@@ -120,7 +120,7 @@ def patch_out_mlx(monkeypatch: pytest.MonkeyPatch):
     monkeypatch.setattr(mlx_runner, "detect_thinking_prompt_suffix", make_nothin(False))
 
     def fake_generate(*_1: object, **_2: object):
-        yield GenerationResponse(token=0, text="hi", finish_reason="stop", usage=None)
+        yield GenerationResponse(token=0, text="hi", finish_reason="stop")
 
     monkeypatch.setattr(mlx_runner, "mlx_generate", fake_generate)
 
@@ -182,8 +182,6 @@ def test_events_processed_in_correct_order(patch_out_mlx: pytest.MonkeyPatch):
             text="hi",
             token_id=0,
             finish_reason="stop",
-            usage=None,
-            stats=None,
         ),
     )
 

tests/headless_runner.py

@@ -16,7 +16,7 @@ from exo.download.impl_shard_downloader import (
     exo_shard_downloader,
 )
 from exo.shared.logging import InterceptLogger, logger_setup
-from exo.shared.models.model_cards import MODEL_CARDS, ModelId
+from exo.shared.models.model_cards import ModelId
 from exo.shared.types.api import ChatCompletionMessage, ChatCompletionTaskParams
 from exo.shared.types.commands import CommandId
 from exo.shared.types.common import Host, NodeId
@@ -89,22 +89,26 @@ async def tb_detection():
 
 async def assert_downloads():
     sd = exo_shard_downloader()
-    # await sd.ensure_shard(await build_full_shard(MODEL_CARDS["qwen3-0.6b"].model_id))
     await sd.ensure_shard(
-        await build_full_shard(MODEL_CARDS["llama-3.1-8b-bf16"].model_id)
-    )
-    await sd.ensure_shard(await build_full_shard(MODEL_CARDS["qwen3-30b"].model_id))
-    await sd.ensure_shard(
-        await build_full_shard(MODEL_CARDS["gpt-oss-120b-MXFP4-Q8"].model_id)
+        await build_full_shard(ModelId("mlx-community/Qwen3-0.6B-4bit"))
     )
     await sd.ensure_shard(
-        await build_full_shard(MODEL_CARDS["gpt-oss-20b-4bit"].model_id)
+        await build_full_shard(ModelId("mlx-community/Meta-Llama-3.1-8B-Instruct-bf16"))
     )
     await sd.ensure_shard(
-        await build_full_shard(MODEL_CARDS["glm-4.7-8bit-gs32"].model_id)
+        await build_full_shard(ModelId("mlx-community/Meta-Llama-3.1-8B-Instruct-bf16"))
     )
     await sd.ensure_shard(
-        await build_full_shard(MODEL_CARDS["minimax-m2.1-8bit"].model_id)
+        await build_full_shard(ModelId("mlx-community/gpt-oss-120b-MXFP4-Q8"))
+    )
+    await sd.ensure_shard(
+        await build_full_shard(ModelId("mlx-community/gpt-oss-20b-MXFP4-Q8"))
+    )
+    await sd.ensure_shard(
+        await build_full_shard(ModelId("mlx-community/GLM-4.7-8bit-gs32"))
+    )
+    await sd.ensure_shard(
+        await build_full_shard(ModelId("mlx-community/MiniMax-M2.1-8bit"))
     )
 
 

tmp/config_examples/opencode.json

@@ -1,18 +0,0 @@
-{
-  "$schema": "https://opencode.ai/config.json",
-  "model": "exo/mlx-community/gpt-oss-120b-MXFP4-Q8",
-  "provider": {
-    "exo": {
-      "api": "http://localhost:52415/v1",
-      "models": {
-        "mlx-community/gpt-oss-120b-MXFP4-Q8": {
-          "name": "GPT OSS 120B",
-          "limit": {
-            "context": 32768,
-            "output": 8192
-          }
-        }
-      }
-    }
-  }
-}

uv.lock

@@ -192,14 +192,20 @@ sdist = { url = "https://files.pythonhosted.org/packages/eb/56/b1ba7935a17738ae8
 wheels = [
     { url = "https://files.pythonhosted.org/packages/b0/1e/d22cc63332bd59b06481ceaac49d6c507598642e2230f201649058a7e704/cffi-2.0.0-cp313-cp313-manylinux1_i686.manylinux2014_i686.manylinux_2_17_i686.manylinux_2_5_i686.whl", hash = "sha256:07b271772c100085dd28b74fa0cd81c8fb1a3ba18b21e03d7c27f3436a10606b", size = 212446 },
     { url = "https://files.pythonhosted.org/packages/a9/f5/a2c23eb03b61a0b8747f211eb716446c826ad66818ddc7810cc2cc19b3f2/cffi-2.0.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:d48a880098c96020b02d5a1f7d9251308510ce8858940e6fa99ece33f610838b", size = 220101 },
+    { url = "https://files.pythonhosted.org/packages/f2/7f/e6647792fc5850d634695bc0e6ab4111ae88e89981d35ac269956605feba/cffi-2.0.0-cp313-cp313-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:f93fd8e5c8c0a4aa1f424d6173f14a892044054871c771f8566e4008eaa359d2", size = 207948 },
+    { url = "https://files.pythonhosted.org/packages/cb/1e/a5a1bd6f1fb30f22573f76533de12a00bf274abcdc55c8edab639078abb6/cffi-2.0.0-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:dd4f05f54a52fb558f1ba9f528228066954fee3ebe629fc1660d874d040ae5a3", size = 206422 },
     { url = "https://files.pythonhosted.org/packages/98/df/0a1755e750013a2081e863e7cd37e0cdd02664372c754e5560099eb7aa44/cffi-2.0.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:c8d3b5532fc71b7a77c09192b4a5a200ea992702734a2e9279a37f2478236f26", size = 219499 },
     { url = "https://files.pythonhosted.org/packages/50/e1/a969e687fcf9ea58e6e2a928ad5e2dd88cc12f6f0ab477e9971f2309b57c/cffi-2.0.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:d9b29c1f0ae438d5ee9acb31cadee00a58c46cc9c0b2f9038c6b0b3470877a8c", size = 222928 },
     { url = "https://files.pythonhosted.org/packages/36/54/0362578dd2c9e557a28ac77698ed67323ed5b9775ca9d3fe73fe191bb5d8/cffi-2.0.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:6d50360be4546678fc1b79ffe7a66265e28667840010348dd69a314145807a1b", size = 221302 },
     { url = "https://files.pythonhosted.org/packages/d6/43/0e822876f87ea8a4ef95442c3d766a06a51fc5298823f884ef87aaad168c/cffi-2.0.0-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:24b6f81f1983e6df8db3adc38562c83f7d4a0c36162885ec7f7b77c7dcbec97b", size = 220049 },
+    { url = "https://files.pythonhosted.org/packages/b4/89/76799151d9c2d2d1ead63c2429da9ea9d7aac304603de0c6e8764e6e8e70/cffi-2.0.0-cp314-cp314-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:12873ca6cb9b0f0d3a0da705d6086fe911591737a59f28b7936bdfed27c0d47c", size = 207793 },
+    { url = "https://files.pythonhosted.org/packages/bb/dd/3465b14bb9e24ee24cb88c9e3730f6de63111fffe513492bf8c808a3547e/cffi-2.0.0-cp314-cp314-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:d9b97165e8aed9272a6bb17c01e3cc5871a594a446ebedc996e2397a1c1ea8ef", size = 206300 },
     { url = "https://files.pythonhosted.org/packages/47/d9/d83e293854571c877a92da46fdec39158f8d7e68da75bf73581225d28e90/cffi-2.0.0-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:afb8db5439b81cf9c9d0c80404b60c3cc9c3add93e114dcae767f1477cb53775", size = 219244 },
     { url = "https://files.pythonhosted.org/packages/2b/0f/1f177e3683aead2bb00f7679a16451d302c436b5cbf2505f0ea8146ef59e/cffi-2.0.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:737fe7d37e1a1bffe70bd5754ea763a62a066dc5913ca57e957824b72a85e205", size = 222828 },
     { url = "https://files.pythonhosted.org/packages/c6/0f/cafacebd4b040e3119dcb32fed8bdef8dfe94da653155f9d0b9dc660166e/cffi-2.0.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:38100abb9d1b1435bc4cc340bb4489635dc2f0da7456590877030c9b3d40b0c1", size = 220926 },
     { url = "https://files.pythonhosted.org/packages/be/b4/c56878d0d1755cf9caa54ba71e5d049479c52f9e4afc230f06822162ab2f/cffi-2.0.0-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:7cc09976e8b56f8cebd752f7113ad07752461f48a58cbba644139015ac24954c", size = 221593 },
+    { url = "https://files.pythonhosted.org/packages/e0/0d/eb704606dfe8033e7128df5e90fee946bbcb64a04fcdaa97321309004000/cffi-2.0.0-cp314-cp314t-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:92b68146a71df78564e4ef48af17551a5ddd142e5190cdf2c5624d0c3ff5b2e8", size = 209354 },
+    { url = "https://files.pythonhosted.org/packages/d8/19/3c435d727b368ca475fb8742ab97c9cb13a0de600ce86f62eab7fa3eea60/cffi-2.0.0-cp314-cp314t-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:b1e74d11748e7e98e2f426ab176d4ed720a64412b6a15054378afdb71e0f37dc", size = 208480 },
     { url = "https://files.pythonhosted.org/packages/d0/44/681604464ed9541673e486521497406fadcc15b5217c3e326b061696899a/cffi-2.0.0-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:28a3a209b96630bca57cce802da70c266eb08c6e97e5afd61a75611ee6c64592", size = 221584 },
     { url = "https://files.pythonhosted.org/packages/25/8e/342a504ff018a2825d395d44d63a767dd8ebc927ebda557fecdaca3ac33a/cffi-2.0.0-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:7553fb2090d71822f02c629afe6042c299edf91ba1bf94951165613553984512", size = 224443 },
     { url = "https://files.pythonhosted.org/packages/e1/5e/b666bacbbc60fbf415ba9988324a132c9a7a0448a9a8f125074671c0f2c3/cffi-2.0.0-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:6c6c373cfc5c83a975506110d17457138c8c63016b563cc9ed6e056a82f13ce4", size = 223437 },
@@ -305,8 +311,10 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/5c/49/498c86566a1d80e978b42f0d702795f69887005548c041636df6ae1ca64c/cryptography-46.0.3-cp311-abi3-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:01ca9ff2885f3acc98c29f1860552e37f6d7c7d013d7334ff2a9de43a449315d", size = 4450807 },
     { url = "https://files.pythonhosted.org/packages/4b/0a/863a3604112174c8624a2ac3c038662d9e59970c7f926acdcfaed8d61142/cryptography-46.0.3-cp311-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:6eae65d4c3d33da080cff9c4ab1f711b15c1d9760809dad6ea763f3812d254cb", size = 4299615 },
     { url = "https://files.pythonhosted.org/packages/64/02/b73a533f6b64a69f3cd3872acb6ebc12aef924d8d103133bb3ea750dc703/cryptography-46.0.3-cp311-abi3-manylinux_2_28_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:e5bf0ed4490068a2e72ac03d786693adeb909981cc596425d09032d372bcc849", size = 4016800 },
+    { url = "https://files.pythonhosted.org/packages/25/d5/16e41afbfa450cde85a3b7ec599bebefaef16b5c6ba4ec49a3532336ed72/cryptography-46.0.3-cp311-abi3-manylinux_2_28_ppc64le.whl", hash = "sha256:5ecfccd2329e37e9b7112a888e76d9feca2347f12f37918facbb893d7bb88ee8", size = 4984707 },
     { url = "https://files.pythonhosted.org/packages/c9/56/e7e69b427c3878352c2fb9b450bd0e19ed552753491d39d7d0a2f5226d41/cryptography-46.0.3-cp311-abi3-manylinux_2_28_x86_64.whl", hash = "sha256:a2c0cd47381a3229c403062f764160d57d4d175e022c1df84e168c6251a22eec", size = 4482541 },
     { url = "https://files.pythonhosted.org/packages/78/f6/50736d40d97e8483172f1bb6e698895b92a223dba513b0ca6f06b2365339/cryptography-46.0.3-cp311-abi3-manylinux_2_34_aarch64.whl", hash = "sha256:549e234ff32571b1f4076ac269fcce7a808d3bf98b76c8dd560e42dbc66d7d91", size = 4299464 },
+    { url = "https://files.pythonhosted.org/packages/00/de/d8e26b1a855f19d9994a19c702fa2e93b0456beccbcfe437eda00e0701f2/cryptography-46.0.3-cp311-abi3-manylinux_2_34_ppc64le.whl", hash = "sha256:c0a7bb1a68a5d3471880e264621346c48665b3bf1c3759d682fc0864c540bd9e", size = 4950838 },
     { url = "https://files.pythonhosted.org/packages/8f/29/798fc4ec461a1c9e9f735f2fc58741b0daae30688f41b2497dcbc9ed1355/cryptography-46.0.3-cp311-abi3-manylinux_2_34_x86_64.whl", hash = "sha256:10b01676fc208c3e6feeb25a8b83d81767e8059e1fe86e1dc62d10a3018fa926", size = 4481596 },
     { url = "https://files.pythonhosted.org/packages/15/8d/03cd48b20a573adfff7652b76271078e3045b9f49387920e7f1f631d125e/cryptography-46.0.3-cp311-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:0abf1ffd6e57c67e92af68330d05760b7b7efb243aab8377e583284dbab72c71", size = 4426782 },
     { url = "https://files.pythonhosted.org/packages/fa/b1/ebacbfe53317d55cf33165bda24c86523497a6881f339f9aae5c2e13e57b/cryptography-46.0.3-cp311-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:a04bee9ab6a4da801eb9b51f1b708a1b5b5c9eb48c03f74198464c66f0d344ac", size = 4698381 },
@@ -314,8 +322,10 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/c5/fd/bc1daf8230eaa075184cbbf5f8cd00ba9db4fd32d63fb83da4671b72ed8a/cryptography-46.0.3-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:39b6755623145ad5eff1dab323f4eae2a32a77a7abef2c5089a04a3d04366715", size = 4435078 },
     { url = "https://files.pythonhosted.org/packages/82/98/d3bd5407ce4c60017f8ff9e63ffee4200ab3e23fe05b765cab805a7db008/cryptography-46.0.3-cp314-cp314t-manylinux_2_28_aarch64.whl", hash = "sha256:db391fa7c66df6762ee3f00c95a89e6d428f4d60e7abc8328f4fe155b5ac6e54", size = 4293460 },
     { url = "https://files.pythonhosted.org/packages/26/e9/e23e7900983c2b8af7a08098db406cf989d7f09caea7897e347598d4cd5b/cryptography-46.0.3-cp314-cp314t-manylinux_2_28_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:78a97cf6a8839a48c49271cdcbd5cf37ca2c1d6b7fdd86cc864f302b5e9bf459", size = 3995237 },
+    { url = "https://files.pythonhosted.org/packages/91/15/af68c509d4a138cfe299d0d7ddb14afba15233223ebd933b4bbdbc7155d3/cryptography-46.0.3-cp314-cp314t-manylinux_2_28_ppc64le.whl", hash = "sha256:dfb781ff7eaa91a6f7fd41776ec37c5853c795d3b358d4896fdbb5df168af422", size = 4967344 },
     { url = "https://files.pythonhosted.org/packages/ca/e3/8643d077c53868b681af077edf6b3cb58288b5423610f21c62aadcbe99f4/cryptography-46.0.3-cp314-cp314t-manylinux_2_28_x86_64.whl", hash = "sha256:6f61efb26e76c45c4a227835ddeae96d83624fb0d29eb5df5b96e14ed1a0afb7", size = 4466564 },
     { url = "https://files.pythonhosted.org/packages/0e/43/c1e8726fa59c236ff477ff2b5dc071e54b21e5a1e51aa2cee1676f1c986f/cryptography-46.0.3-cp314-cp314t-manylinux_2_34_aarch64.whl", hash = "sha256:23b1a8f26e43f47ceb6d6a43115f33a5a37d57df4ea0ca295b780ae8546e8044", size = 4292415 },
+    { url = "https://files.pythonhosted.org/packages/42/f9/2f8fefdb1aee8a8e3256a0568cffc4e6d517b256a2fe97a029b3f1b9fe7e/cryptography-46.0.3-cp314-cp314t-manylinux_2_34_ppc64le.whl", hash = "sha256:b419ae593c86b87014b9be7396b385491ad7f320bde96826d0dd174459e54665", size = 4931457 },
     { url = "https://files.pythonhosted.org/packages/79/30/9b54127a9a778ccd6d27c3da7563e9f2d341826075ceab89ae3b41bf5be2/cryptography-46.0.3-cp314-cp314t-manylinux_2_34_x86_64.whl", hash = "sha256:50fc3343ac490c6b08c0cf0d704e881d0d660be923fd3076db3e932007e726e3", size = 4466074 },
     { url = "https://files.pythonhosted.org/packages/ac/68/b4f4a10928e26c941b1b6a179143af9f4d27d88fe84a6a3c53592d2e76bf/cryptography-46.0.3-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:22d7e97932f511d6b0b04f2bfd818d73dcd5928db509460aaf48384778eb6d20", size = 4420569 },
     { url = "https://files.pythonhosted.org/packages/a3/49/3746dab4c0d1979888f125226357d3262a6dd40e114ac29e3d2abdf1ec55/cryptography-46.0.3-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:d55f3dffadd674514ad19451161118fd010988540cee43d8bc20675e775925de", size = 4681941 },
@@ -323,8 +333,10 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/26/42/fa8389d4478368743e24e61eea78846a0006caffaf72ea24a15159215a14/cryptography-46.0.3-cp38-abi3-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:15ab9b093e8f09daab0f2159bb7e47532596075139dd74365da52ecc9cb46c5d", size = 4440029 },
     { url = "https://files.pythonhosted.org/packages/5f/eb/f483db0ec5ac040824f269e93dd2bd8a21ecd1027e77ad7bdf6914f2fd80/cryptography-46.0.3-cp38-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:46acf53b40ea38f9c6c229599a4a13f0d46a6c3fa9ef19fc1a124d62e338dfa0", size = 4297222 },
     { url = "https://files.pythonhosted.org/packages/fd/cf/da9502c4e1912cb1da3807ea3618a6829bee8207456fbbeebc361ec38ba3/cryptography-46.0.3-cp38-abi3-manylinux_2_28_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:10ca84c4668d066a9878890047f03546f3ae0a6b8b39b697457b7757aaf18dbc", size = 4012280 },
+    { url = "https://files.pythonhosted.org/packages/6b/8f/9adb86b93330e0df8b3dcf03eae67c33ba89958fc2e03862ef1ac2b42465/cryptography-46.0.3-cp38-abi3-manylinux_2_28_ppc64le.whl", hash = "sha256:36e627112085bb3b81b19fed209c05ce2a52ee8b15d161b7c643a7d5a88491f3", size = 4978958 },
     { url = "https://files.pythonhosted.org/packages/d1/a0/5fa77988289c34bdb9f913f5606ecc9ada1adb5ae870bd0d1054a7021cc4/cryptography-46.0.3-cp38-abi3-manylinux_2_28_x86_64.whl", hash = "sha256:1000713389b75c449a6e979ffc7dcc8ac90b437048766cef052d4d30b8220971", size = 4473714 },
     { url = "https://files.pythonhosted.org/packages/14/e5/fc82d72a58d41c393697aa18c9abe5ae1214ff6f2a5c18ac470f92777895/cryptography-46.0.3-cp38-abi3-manylinux_2_34_aarch64.whl", hash = "sha256:b02cf04496f6576afffef5ddd04a0cb7d49cf6be16a9059d793a30b035f6b6ac", size = 4296970 },
+    { url = "https://files.pythonhosted.org/packages/78/06/5663ed35438d0b09056973994f1aec467492b33bd31da36e468b01ec1097/cryptography-46.0.3-cp38-abi3-manylinux_2_34_ppc64le.whl", hash = "sha256:71e842ec9bc7abf543b47cf86b9a743baa95f4677d22baa4c7d5c69e49e9bc04", size = 4940236 },
     { url = "https://files.pythonhosted.org/packages/fc/59/873633f3f2dcd8a053b8dd1d38f783043b5fce589c0f6988bf55ef57e43e/cryptography-46.0.3-cp38-abi3-manylinux_2_34_x86_64.whl", hash = "sha256:402b58fc32614f00980b66d6e56a5b4118e6cb362ae8f3fda141ba4689bd4506", size = 4472642 },
     { url = "https://files.pythonhosted.org/packages/3d/39/8e71f3930e40f6877737d6f69248cf74d4e34b886a3967d32f919cc50d3b/cryptography-46.0.3-cp38-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:ef639cb3372f69ec44915fafcd6698b6cc78fbe0c2ea41be867f6ed612811963", size = 4423126 },
     { url = "https://files.pythonhosted.org/packages/cd/c7/f65027c2810e14c3e7268353b1681932b87e5a48e65505d8cc17c99e36ae/cryptography-46.0.3-cp38-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:3b51b8ca4f1c6453d8829e1eb7299499ca7f313900dd4d89a24b8b87c0a780d4", size = 4686573 },
@@ -403,7 +415,7 @@ requires-dist = [
     { name = "mflux", specifier = "==0.15.4" },
     { name = "mlx", marker = "sys_platform == 'darwin'", specifier = "==0.30.4" },
     { name = "mlx", extras = ["cpu"], marker = "sys_platform == 'linux'", specifier = "==0.30.4" },
-    { name = "mlx-lm", specifier = "==0.30.5" },
+    { name = "mlx-lm", git = "https://github.com/ml-explore/mlx-lm?branch=main" },
     { name = "openai-harmony", specifier = ">=0.0.8" },
     { name = "pillow", specifier = ">=11.0,<12.0" },
     { name = "psutil", specifier = ">=7.0.0" },
@@ -1061,7 +1073,7 @@ wheels = [
 [[package]]
 name = "mlx-lm"
 version = "0.30.5"
-source = { registry = "https://pypi.org/simple" }
+source = { git = "https://github.com/ml-explore/mlx-lm?branch=main#96699e6dadb13b82b28285bb131a0741997d19ae" }
 dependencies = [
     { name = "jinja2", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
     { name = "mlx", marker = "sys_platform == 'darwin'" },
@@ -1071,10 +1083,6 @@ dependencies = [
     { name = "sentencepiece", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
     { name = "transformers", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/0b/90/4469d9f75f196e6255f59a89441abe0079925d30a001462e1c1c4bc4e6a1/mlx_lm-0.30.5.tar.gz", hash = "sha256:9e6cb258c65b766c6af25cb90958aef40acab67139f05839eef19864cb3154f6", size = 262367 }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/89/ba/66db6e1e5f1ef506655b562932f6bd8f72600116d5f31f92d71c1f200b3f/mlx_lm-0.30.5-py3-none-any.whl", hash = "sha256:a80bc8e3efdebe81813b0f6eb403fb66a7a15071e256f4e7102ada986acb75bb", size = 366716 },
-]
 
 [[package]]
 name = "mlx-metal"
@@ -1495,9 +1503,6 @@ version = "11.3.0"
 source = { registry = "https://pypi.org/simple" }
 sdist = { url = "https://files.pythonhosted.org/packages/f3/0d/d0d6dea55cd152ce3d6767bb38a8fc10e33796ba4ba210cbab9354b6d238/pillow-11.3.0.tar.gz", hash = "sha256:3828ee7586cd0b2091b6209e5ad53e20d0649bbe87164a459d0676e035e8f523", size = 47113069 }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/1e/93/0952f2ed8db3a5a4c7a11f91965d6184ebc8cd7cbb7941a260d5f018cd2d/pillow-11.3.0-cp313-cp313-ios_13_0_arm64_iphoneos.whl", hash = "sha256:1c627742b539bba4309df89171356fcb3cc5a9178355b2727d1b74a6cf155fbd", size = 2128328 },
-    { url = "https://files.pythonhosted.org/packages/4b/e8/100c3d114b1a0bf4042f27e0f87d2f25e857e838034e98ca98fe7b8c0a9c/pillow-11.3.0-cp313-cp313-ios_13_0_arm64_iphonesimulator.whl", hash = "sha256:30b7c02f3899d10f13d7a48163c8969e4e653f8b43416d23d13d1bbfdc93b9f8", size = 2170652 },
-    { url = "https://files.pythonhosted.org/packages/aa/86/3f758a28a6e381758545f7cdb4942e1cb79abd271bea932998fc0db93cb6/pillow-11.3.0-cp313-cp313-ios_13_0_x86_64_iphonesimulator.whl", hash = "sha256:7859a4cc7c9295f5838015d8cc0a9c215b77e43d07a25e460f35cf516df8626f", size = 2227443 },
     { url = "https://files.pythonhosted.org/packages/01/f4/91d5b3ffa718df2f53b0dc109877993e511f4fd055d7e9508682e8aba092/pillow-11.3.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:ec1ee50470b0d050984394423d96325b744d55c701a439d2bd66089bff963d3c", size = 5278474 },
     { url = "https://files.pythonhosted.org/packages/f9/0e/37d7d3eca6c879fbd9dba21268427dffda1ab00d4eb05b32923d4fbe3b12/pillow-11.3.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:7db51d222548ccfd274e4572fdbf3e810a5e66b00608862f947b163e613b67dd", size = 4686038 },
     { url = "https://files.pythonhosted.org/packages/ff/b0/3426e5c7f6565e752d81221af9d3676fdbb4f352317ceafd42899aaf5d8a/pillow-11.3.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:2d6fcc902a24ac74495df63faad1884282239265c6839a0a6416d33faedfae7e", size = 5864407 },
@@ -2273,7 +2278,7 @@ wheels = [
 
 [[package]]
 name = "transformers"
-version = "5.0.0rc3"
+version = "5.0.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "filelock", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
@@ -2282,15 +2287,14 @@ dependencies = [
     { name = "packaging", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
     { name = "pyyaml", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
     { name = "regex", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
-    { name = "requests", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
     { name = "safetensors", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
     { name = "tokenizers", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
     { name = "tqdm", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
     { name = "typer-slim", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/3f/a3/7c116a8d85f69ea7749cf4c2df79e64c35d028e5fc7ea0168f299d03b8c7/transformers-5.0.0rc3.tar.gz", hash = "sha256:a0315b92b7e087617ade42ec9e6e92ee7620541cc5d6a3331886c52cbe306f5c", size = 8388520 }
+sdist = { url = "https://files.pythonhosted.org/packages/bc/79/845941711811789c85fb7e2599cea425a14a07eda40f50896b9d3fda7492/transformers-5.0.0.tar.gz", hash = "sha256:5f5634efed6cf76ad068cc5834c7adbc32db78bbd6211fb70df2325a9c37dec8", size = 8424830 }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/1e/f2/ae2b8968764253bdf38a48dee3c299b8d0bedf7c8ffbe3449fca9bd95338/transformers-5.0.0rc3-py3-none-any.whl", hash = "sha256:383fad27f4f73092d330e45fae384681e5c8521e1dc1cf6cb1a297780e68bf2d", size = 10107087 },
+    { url = "https://files.pythonhosted.org/packages/52/f3/ac976fa8e305c9e49772527e09fbdc27cc6831b8a2f6b6063406626be5dd/transformers-5.0.0-py3-none-any.whl", hash = "sha256:587086f249ce64c817213cf36afdb318d087f790723e9b3d4500b97832afd52d", size = 10142091 },
 ]
 
 [[package]]