docs: fix broken .md links and render issues (#13550 )

amd: use GTT on iGPUs on linux (#13196 )
On Linux, look at the GTT memory information for iGPUs.
2025-12-23 23:59:11 -05:00 · 2025-12-23 12:44:55 -05:00 · 2025-12-23 09:30:05 -08:00 · 2025-12-19 15:48:15 -08:00 · 2025-12-18 13:21:29 -08:00
23 changed files with 190 additions and 60 deletions
--- a/api/types.go
+++ b/api/types.go
@@ -554,6 +554,9 @@ type CreateRequest struct {
 	Renderer string `json:"renderer,omitempty"`
 	Parser   string `json:"parser,omitempty"`

+	// Requires is the minimum version of Ollama required by the model.
+	Requires string `json:"requires,omitempty"`
+
 	// Info is a map of additional information for the model
 	Info map[string]any `json:"info,omitempty"`

@@ -604,6 +607,7 @@ type ShowResponse struct {
 	Tensors       []Tensor           `json:"tensors,omitempty"`
 	Capabilities  []model.Capability `json:"capabilities,omitempty"`
 	ModifiedAt    time.Time          `json:"modified_at,omitempty"`
+	Requires      string             `json:"requires,omitempty"`
 }

 // CopyRequest is the request passed to [Client.Copy].
--- a/cmd/cmd.go
+++ b/cmd/cmd.go
@@ -943,6 +943,9 @@ func showInfo(resp *api.ShowResponse, verbose bool, w io.Writer) error {
 			rows = append(rows, []string{"", "parameters", resp.Details.ParameterSize})
 		}
 		rows = append(rows, []string{"", "quantization", resp.Details.QuantizationLevel})
+		if resp.Requires != "" {
+			rows = append(rows, []string{"", "requires", resp.Requires})
+		}
 		return
 	})

--- a/cmd/cmd_test.go
+++ b/cmd/cmd_test.go
@@ -291,6 +291,31 @@ Weigh anchor!
 			t.Errorf("unexpected output (-want +got):\n%s", diff)
 		}
 	})
+
+	t.Run("min version", func(t *testing.T) {
+		var b bytes.Buffer
+		if err := showInfo(&api.ShowResponse{
+			Details: api.ModelDetails{
+				Family:            "test",
+				ParameterSize:     "7B",
+				QuantizationLevel: "FP16",
+			},
+			Requires: "0.14.0",
+		}, false, &b); err != nil {
+			t.Fatal(err)
+		}
+
+		expect := `  Model
+    architecture    test      
+    parameters      7B        
+    quantization    FP16      
+    requires        0.14.0    
+
+`
+		if diff := cmp.Diff(expect, b.String()); diff != "" {
+			t.Errorf("unexpected output (-want +got):\n%s", diff)
+		}
+	})
 }

 func TestDeleteHandler(t *testing.T) {
--- a/docs/faq.mdx
+++ b/docs/faq.mdx
@@ -14,11 +14,11 @@ curl -fsSL https://ollama.com/install.sh | sh

 ## How can I view the logs?

-Review the [Troubleshooting](./troubleshooting.md) docs for more about using logs.
+Review the [Troubleshooting](./troubleshooting) docs for more about using logs.

 ## Is my GPU compatible with Ollama?

-Please refer to the [GPU docs](./gpu.md).
+Please refer to the [GPU docs](./gpu).

 ## How can I specify the context window size?

--- a/docs/gpu.mdx
+++ b/docs/gpu.mdx
@@ -33,7 +33,7 @@ Check your compute compatibility to see if your card is supported:
 | 5.0                | GeForce GTX         | `GTX 750 Ti` `GTX 750` `NVS 810`                                                                                               |
 |                    | Quadro              | `K2200` `K1200` `K620` `M1200` `M520` `M5000M` `M4000M` `M3000M` `M2000M` `M1000M` `K620M` `M600M` `M500M`                     |

-For building locally to support older GPUs, see [developer.md](./development.md#linux-cuda-nvidia)
+For building locally to support older GPUs, see [developer](./development#linux-cuda-nvidia)

 ### GPU Selection

@@ -54,7 +54,7 @@ sudo modprobe nvidia_uvm`

 Ollama supports the following AMD GPUs via the ROCm library:

-> [!NOTE]
+> **NOTE:**
 > Additional AMD GPU support is provided by the Vulkan Library - see below.


@@ -132,9 +132,9 @@ Ollama supports GPU acceleration on Apple devices via the Metal API.

 ## Vulkan GPU Support

-> [!NOTE]
+> **NOTE:**
 > Vulkan is currently an Experimental feature.  To enable, you must set OLLAMA_VULKAN=1 for the Ollama server as
-described in the [FAQ](faq.md#how-do-i-configure-ollama-server)
+described in the [FAQ](faq#how-do-i-configure-ollama-server)

 Additional GPU support on Windows and Linux is provided via
 [Vulkan](https://www.vulkan.org/). On Windows most GPU vendors drivers come
@@ -161,6 +161,6 @@ sudo setcap cap_perfmon+ep /usr/local/bin/ollama

 To select specific Vulkan GPU(s), you can set the environment variable
 `GGML_VK_VISIBLE_DEVICES` to one or more numeric IDs on the Ollama server as
-described in the [FAQ](faq.md#how-do-i-configure-ollama-server). If you
+described in the [FAQ](faq#how-do-i-configure-ollama-server). If you
 encounter any problems with Vulkan based GPUs, you can disable all Vulkan GPUs
 by setting `GGML_VK_VISIBLE_DEVICES=-1` 
--- a/docs/modelfile.mdx
+++ b/docs/modelfile.mdx
@@ -41,6 +41,7 @@ INSTRUCTION arguments
 | [`ADAPTER`](#adapter)               | Defines the (Q)LoRA adapters to apply to the model.            |
 | [`LICENSE`](#license)               | Specifies the legal license.                                   |
 | [`MESSAGE`](#message)               | Specify message history.                                       |
+| [`REQUIRES`](#requires)             | Specify the minimum version of Ollama required by the model.   |

 ## Examples

@@ -248,6 +249,16 @@ MESSAGE user Is Ontario in Canada?
 MESSAGE assistant yes
 ```

+### REQUIRES
+
+The `REQUIRES` instruction allows you to specify the minimum version of Ollama required by the model.
+
+```
+REQUIRES <version>
+```
+
+The version should be a valid Ollama version (e.g. 0.14.0).
+
 ## Notes

 - the **`Modelfile` is not case sensitive**. In the examples, uppercase instructions are used to make it easier to distinguish it from arguments.
--- a/docs/troubleshooting.mdx
+++ b/docs/troubleshooting.mdx
@@ -87,7 +87,7 @@ When Ollama starts up, it takes inventory of the GPUs present in the system to d

 ### Linux NVIDIA Troubleshooting

-If you are using a container to run Ollama, make sure you've set up the container runtime first as described in [docker.md](./docker.md)
+If you are using a container to run Ollama, make sure you've set up the container runtime first as described in [docker](./docker)

 Sometimes the Ollama can have difficulties initializing the GPU. When you check the server logs, this can show up as various error codes, such as "3" (not initialized), "46" (device unavailable), "100" (no device), "999" (unknown), or others. The following troubleshooting techniques may help resolve the problem

--- a/go.mod
+++ b/go.mod
@@ -15,8 +15,8 @@ require (
 	github.com/spf13/cobra v1.7.0
 	github.com/stretchr/testify v1.9.0
 	github.com/x448/float16 v0.8.4
-	golang.org/x/sync v0.12.0
-	golang.org/x/sys v0.36.0
+	golang.org/x/sync v0.17.0
+	golang.org/x/sys v0.37.0
 )

 require (
@@ -29,7 +29,8 @@ require (
 	github.com/pdevine/tensor v0.0.0-20240510204454-f88f4562727c
 	github.com/tkrajina/typescriptify-golang-structs v0.2.0
 	golang.org/x/image v0.22.0
-	golang.org/x/tools v0.30.0
+	golang.org/x/mod v0.30.0
+	golang.org/x/tools v0.38.0
 	gonum.org/v1/gonum v0.15.0
 )

@@ -76,11 +77,11 @@ require (
 	github.com/twitchyliquid64/golang-asm v0.15.1 // indirect
 	github.com/ugorji/go/codec v1.2.12 // indirect
 	golang.org/x/arch v0.8.0 // indirect
-	golang.org/x/crypto v0.36.0
+	golang.org/x/crypto v0.43.0
 	golang.org/x/exp v0.0.0-20250218142911-aa4b98e5adaa // indirect
-	golang.org/x/net v0.38.0 // indirect
-	golang.org/x/term v0.30.0
-	golang.org/x/text v0.23.0
+	golang.org/x/net v0.46.0 // indirect
+	golang.org/x/term v0.36.0
+	golang.org/x/text v0.30.0
 	google.golang.org/protobuf v1.34.1
 	gopkg.in/yaml.v3 v3.0.1 // indirect
 )
--- a/go.sum
+++ b/go.sum
@@ -224,8 +224,8 @@ golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACk
 golang.org/x/crypto v0.0.0-20190510104115-cbcb75029529/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
 golang.org/x/crypto v0.0.0-20191011191535-87dc89f01550/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
 golang.org/x/crypto v0.0.0-20200622213623-75b288015ac9/go.mod h1:LzIPMQfyMNhhGPhUkYOs5KpL4U8rLKemX1yGLhDgUto=
-golang.org/x/crypto v0.36.0 h1:AnAEvhDddvBdpY+uR+MyHmuZzzNqXSe/GvuDeob5L34=
-golang.org/x/crypto v0.36.0/go.mod h1:Y4J0ReaxCR1IMaabaSMugxJES1EpwhBHhv2bDHklZvc=
+golang.org/x/crypto v0.43.0 h1:dduJYIi3A3KOfdGOHX8AVZ/jGiyPa3IbBozJ5kNuE04=
+golang.org/x/crypto v0.43.0/go.mod h1:BFbav4mRNlXJL4wNeejLpWxB7wMbc79PdRGhWKncxR0=
 golang.org/x/exp v0.0.0-20180321215751-8460e604b9de/go.mod h1:CJ0aWSM057203Lf6IL+f9T1iT9GByDxfZKAQTCR3kQA=
 golang.org/x/exp v0.0.0-20180807140117-3d87b88a115f/go.mod h1:CJ0aWSM057203Lf6IL+f9T1iT9GByDxfZKAQTCR3kQA=
 golang.org/x/exp v0.0.0-20190121172915-509febef88a4/go.mod h1:CJ0aWSM057203Lf6IL+f9T1iT9GByDxfZKAQTCR3kQA=
@@ -255,6 +255,8 @@ golang.org/x/mod v0.1.1-0.20191105210325-c90efee705ee/go.mod h1:QqPTAvyqsEbceGzB
 golang.org/x/mod v0.2.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
 golang.org/x/mod v0.3.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
 golang.org/x/mod v0.4.2/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
+golang.org/x/mod v0.30.0 h1:fDEXFVZ/fmCKProc/yAXXUijritrDzahmwwefnjoPFk=
+golang.org/x/mod v0.30.0/go.mod h1:lAsf5O2EvJeSFMiBxXDki7sCgAxEUcZHXoXMKT4GJKc=
 golang.org/x/net v0.0.0-20180724234803-3673e40ba225/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
 golang.org/x/net v0.0.0-20180826012351-8a410e7b638d/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
 golang.org/x/net v0.0.0-20190108225652-1e06a53dbb7e/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
@@ -267,8 +269,8 @@ golang.org/x/net v0.0.0-20200822124328-c89045814202/go.mod h1:/O7V0waA8r7cgGh81R
 golang.org/x/net v0.0.0-20201021035429-f5854403a974/go.mod h1:sp8m0HH+o8qH0wwXwYZr8TS3Oi6o0r6Gce1SSxlDquU=
 golang.org/x/net v0.0.0-20210405180319-a5a99cb37ef4/go.mod h1:p54w0d4576C0XHj96bSt6lcn1PtDYWL6XObtHCRCNQM=
 golang.org/x/net v0.0.0-20210614182718-04defd469f4e/go.mod h1:9nx3DQGgdP8bBQD5qxJ1jj9UTztislL4KSBs9R2vV5Y=
-golang.org/x/net v0.38.0 h1:vRMAPTMaeGqVhG5QyLJHqNDwecKTomGeqbnfZyKlBI8=
-golang.org/x/net v0.38.0/go.mod h1:ivrbrMbzFq5J41QOQh0siUuly180yBYtLp+CKbEaFx8=
+golang.org/x/net v0.46.0 h1:giFlY12I07fugqwPuWJi68oOnpfqFnJIJzaIIm2JVV4=
+golang.org/x/net v0.46.0/go.mod h1:Q9BGdFy1y4nkUwiLvT5qtyhAnEHgnQ/zd8PfU6nc210=
 golang.org/x/oauth2 v0.0.0-20180821212333-d2e6202438be/go.mod h1:N/0e6XlmueqKjAGxoOufVs8QHGRruUQn6yWY3a++T0U=
 golang.org/x/oauth2 v0.0.0-20200107190931-bf48bf16ab8d/go.mod h1:gOpvHmFTYa4IltrdGE7lF6nIHvwfUNPOp7c8zoXwtLw=
 golang.org/x/sync v0.0.0-20180314180146-1d60e4601c6f/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
@@ -278,8 +280,8 @@ golang.org/x/sync v0.0.0-20190423024810-112230192c58/go.mod h1:RxMgew5VJxzue5/jJ
 golang.org/x/sync v0.0.0-20190911185100-cd5d95a43a6e/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sync v0.0.0-20201020160332-67f06af15bc9/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sync v0.0.0-20210220032951-036812b2e83c/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
-golang.org/x/sync v0.12.0 h1:MHc5BpPuC30uJk597Ri8TV3CNZcTLu6B6z4lJy+g6Jw=
-golang.org/x/sync v0.12.0/go.mod h1:1dzgHSNfp02xaA81J2MS99Qcpr2w7fw1gpm99rleRqA=
+golang.org/x/sync v0.17.0 h1:l60nONMj9l5drqw6jlhIELNv9I0A4OFgRsG9k2oT9Ug=
+golang.org/x/sync v0.17.0/go.mod h1:9KTHXmSnoGruLpwFjVSX0lNNA75CykiMECbovNTZqGI=
 golang.org/x/sys v0.0.0-20180830151530-49385e6e1522/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20190312061237-fead79001313/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
@@ -295,17 +297,17 @@ golang.org/x/sys v0.0.0-20210510120138-977fb7262007/go.mod h1:oPkhp1MJrh7nUepCBc
 golang.org/x/sys v0.0.0-20210630005230-0f9fa26af87c/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.5.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.36.0 h1:KVRy2GtZBrk1cBYA7MKu5bEZFxQk4NIDV6RLVcC8o0k=
-golang.org/x/sys v0.36.0/go.mod h1:OgkHotnGiDImocRcuBABYBEXf8A9a87e/uXjp9XT3ks=
+golang.org/x/sys v0.37.0 h1:fdNQudmxPjkdUTPnLn5mdQv7Zwvbvpaxqs831goi9kQ=
+golang.org/x/sys v0.37.0/go.mod h1:OgkHotnGiDImocRcuBABYBEXf8A9a87e/uXjp9XT3ks=
 golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1/go.mod h1:bj7SfCRtBDWHUb9snDiAeCFNEtKQo2Wmx5Cou7ajbmo=
-golang.org/x/term v0.30.0 h1:PQ39fJZ+mfadBm0y5WlL4vlM7Sx1Hgf13sMIY2+QS9Y=
-golang.org/x/term v0.30.0/go.mod h1:NYYFdzHoI5wRh/h5tDMdMqCqPJZEuNqVR5xJLd/n67g=
+golang.org/x/term v0.36.0 h1:zMPR+aF8gfksFprF/Nc/rd1wRS1EI6nDBGyWAvDzx2Q=
+golang.org/x/term v0.36.0/go.mod h1:Qu394IJq6V6dCBRgwqshf3mPF85AqzYEzofzRdZkWss=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 golang.org/x/text v0.3.3/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
 golang.org/x/text v0.3.5/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
 golang.org/x/text v0.3.6/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
-golang.org/x/text v0.23.0 h1:D71I7dUrlY+VX0gQShAThNGHFxZ13dGLBHQLVl1mJlY=
-golang.org/x/text v0.23.0/go.mod h1:/BLNzu4aZCJ1+kcD0DNRotWKage4q2rGVAg4o22unh4=
+golang.org/x/text v0.30.0 h1:yznKA/E9zq54KzlzBEAWn1NXSQ8DIp/NYMy88xJjl4k=
+golang.org/x/text v0.30.0/go.mod h1:yDdHFIX9t+tORqspjENWgzaCVXgk0yYnYuSZ8UzzBVM=
 golang.org/x/tools v0.0.0-20180525024113-a5b4c53f6e8b/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
 golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
 golang.org/x/tools v0.0.0-20190114222345-bf090417da8b/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
@@ -319,8 +321,8 @@ golang.org/x/tools v0.0.0-20200130002326-2f3ba24bd6e7/go.mod h1:TB2adYChydJhpapK
 golang.org/x/tools v0.0.0-20200619180055-7c47624df98f/go.mod h1:EkVYQZoAsY45+roYkvgYkIh4xh/qjgUK9TdY2XT94GE=
 golang.org/x/tools v0.0.0-20210106214847-113979e3529a/go.mod h1:emZCQorbCU4vsT4fOWvOPXz4eW1wZW4PmDk9uLelYpA=
 golang.org/x/tools v0.1.4/go.mod h1:o0xws9oXOQQZyjljx8fwUC0k7L1pTE6eaCbjGeHmOkk=
-golang.org/x/tools v0.30.0 h1:BgcpHewrV5AUp2G9MebG4XPFI1E2W41zU1SaqVA9vJY=
-golang.org/x/tools v0.30.0/go.mod h1:c347cR/OJfw5TI+GfX7RUPNMdDRRbjvYTS0jPyvsVtY=
+golang.org/x/tools v0.38.0 h1:Hx2Xv8hISq8Lm16jvBZ2VQf+RLmbd7wVUsALibYI/IQ=
+golang.org/x/tools v0.38.0/go.mod h1:yEsQ/d/YK8cjh0L6rZlY8tgtlKiBNTL14pGDJPJpYQs=
 golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
--- a/llama/patches/0024-GPU-discovery-enhancements.patch
+++ b/llama/patches/0024-GPU-discovery-enhancements.patch
@@ -20,10 +20,10 @@ fix vulkan PCI ID and ID handling
 ggml/src/ggml-cuda/vendors/hip.h     |   3 +
 ggml/src/ggml-impl.h                 |   8 +
 ggml/src/ggml-metal/ggml-metal.cpp   |   2 +
- ggml/src/ggml-vulkan/ggml-vulkan.cpp | 169 ++++++++-
- ggml/src/mem_hip.cpp                 | 529 +++++++++++++++++++++++++++
- ggml/src/mem_nvml.cpp                | 209 +++++++++++
- 9 files changed, 976 insertions(+), 17 deletions(-)
+ ggml/src/ggml-vulkan/ggml-vulkan.cpp | 169 +++++++-
+ ggml/src/mem_hip.cpp                 | 558 +++++++++++++++++++++++++++
+ ggml/src/mem_nvml.cpp                | 209 ++++++++++
+ 9 files changed, 1005 insertions(+), 17 deletions(-)
 create mode 100644 ggml/src/mem_hip.cpp
 create mode 100644 ggml/src/mem_nvml.cpp

@@ -58,7 +58,7 @@ index d55aed348..99ae293cc 100644
 
 set_target_properties(ggml-base PROPERTIES
 diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
-index 6852d2e20..48cdb1dcf 100644
+index 6852d2e20..334a30135 100644
 --- a/ggml/src/ggml-cuda/ggml-cuda.cu
 +++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -267,6 +267,16 @@ static ggml_cuda_device_info ggml_cuda_init() {
@@ -109,7 +109,7 @@ index 6852d2e20..48cdb1dcf 100644
 +
 +#if defined(GGML_USE_HIP)
 +    if (ggml_hip_mgmt_init() == 0) {
-+        int status = ggml_hip_get_device_memory(ctx->pci_bus_id.c_str(), free, total);
+        int status = ggml_hip_get_device_memory(ctx->pci_bus_id.c_str(), free, total, ctx->integrated != 0);
 +        if (status == 0) {
 +            GGML_LOG_DEBUG("%s device %s utilizing AMD specific memory reporting free: %zu total: %zu\n", __func__, ctx->pci_bus_id.c_str(), *free, *total);
 +            ggml_hip_mgmt_release();
@@ -204,7 +204,7 @@ index 4e162258d..d89e35a8e 100644
 #define cudaErrorPeerAccessAlreadyEnabled hipErrorPeerAccessAlreadyEnabled
 #define cudaErrorPeerAccessNotEnabled hipErrorPeerAccessNotEnabled
 diff --git a/ggml/src/ggml-impl.h b/ggml/src/ggml-impl.h
-index fe57d4c58..1c07e767a 100644
+index fe57d4c58..dba8f4695 100644
 --- a/ggml/src/ggml-impl.h
 +++ b/ggml/src/ggml-impl.h
@@ -677,6 +677,14 @@ static inline bool ggml_can_fuse_subgraph(const struct ggml_cgraph * cgraph,
@@ -216,7 +216,7 @@ index fe57d4c58..1c07e767a 100644
 +GGML_API int ggml_nvml_get_device_memory(const char *uuid, size_t *free, size_t *total);
 +GGML_API void ggml_nvml_release();
 +GGML_API int ggml_hip_mgmt_init();
-+GGML_API int ggml_hip_get_device_memory(const char *id, size_t *free, size_t *total);
+GGML_API int ggml_hip_get_device_memory(const char *id, size_t *free, size_t *total, bool is_integrated_gpu);
 +GGML_API void ggml_hip_mgmt_release();
 +
 #ifdef __cplusplus
@@ -243,7 +243,7 @@ index ba95b4acc..f6f8f7a10 100644
         /* .async                 = */ true,
         /* .host_buffer           = */ false,
 diff --git a/ggml/src/ggml-vulkan/ggml-vulkan.cpp b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
-index 5349bce24..d43d46d1d 100644
+index 5349bce24..0103fd03a 100644
 --- a/ggml/src/ggml-vulkan/ggml-vulkan.cpp
 +++ b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
@@ -236,6 +236,7 @@ class vk_memory_logger;
@@ -334,7 +334,7 @@ index 5349bce24..d43d46d1d 100644
 +        switch (props2.properties.vendorID) {
 +        case VK_VENDOR_ID_AMD:
 +            if (ggml_hip_mgmt_init() == 0) {
-+                int status = ggml_hip_get_device_memory(ctx->pci_id != "" ? ctx->pci_id.c_str() : ctx->uuid.c_str(), free, total);
+                int status = ggml_hip_get_device_memory(ctx->pci_id != "" ? ctx->pci_id.c_str() : ctx->uuid.c_str(), free, total, ctx->is_integrated_gpu);
 +                if (status == 0) {
 +                    GGML_LOG_DEBUG("%s device %s utilizing AMD specific memory reporting free: %zu total: %zu\n", __func__, ctx->pci_id != "" ? ctx->pci_id.c_str() : ctx->uuid.c_str(), *free, *total);
 +                    ggml_hip_mgmt_release();
@@ -505,10 +505,10 @@ index 5349bce24..d43d46d1d 100644
         }
 diff --git a/ggml/src/mem_hip.cpp b/ggml/src/mem_hip.cpp
 new file mode 100644
-index 000000000..c1949b899
+index 000000000..23c765806
 --- /dev/null
 +++ b/ggml/src/mem_hip.cpp
-@@ -0,0 +1,529 @@
+@@ -0,0 +1,558 @@
 +#include "ggml.h"
 +#include "ggml-impl.h"
 +
@@ -842,7 +842,7 @@ index 000000000..c1949b899
 +    if (gpus != NULL) gpus->pVtbl->Release(gpus); \
 +    if (gpu != NULL) gpu->pVtbl->Release(gpu)
 +
-+int ggml_hip_get_device_memory(const char *id, size_t *free, size_t *total) {
+int ggml_hip_get_device_memory(const char *id, size_t *free, size_t *total, bool is_integrated_gpu) {
 +    std::lock_guard<std::mutex> lock(ggml_adlx_lock);
 +    if (adlx.handle == NULL) {
 +        GGML_LOG_INFO("%s ADLX was not initialized\n", __func__);
@@ -966,13 +966,16 @@ index 000000000..c1949b899
 +    return 0;
 +}
 +void ggml_hip_mgmt_release() {}
-+int ggml_hip_get_device_memory(const char *id, size_t *free, size_t *total) {
+int ggml_hip_get_device_memory(const char *id, size_t *free, size_t *total, bool is_integrated_gpu) {
 +    GGML_LOG_INFO("%s searching for device %s\n", __func__, id);
 +    const std::string drmDeviceGlob = "/sys/class/drm/card*/device/uevent";
 +    const std::string drmTotalMemoryFile = "mem_info_vram_total";
 +    const std::string drmUsedMemoryFile = "mem_info_vram_used";
+    const std::string drmGTTTotalMemoryFile = "mem_info_gtt_total";
+    const std::string drmGTTUsedMemoryFile = "mem_info_gtt_used";
 +    const std::string drmUeventPCISlotLabel = "PCI_SLOT_NAME=";
 +
+
 +    glob_t glob_result;
 +    glob(drmDeviceGlob.c_str(), GLOB_NOSORT, NULL, &glob_result);
 +
@@ -1006,7 +1009,6 @@ index 000000000..c1949b899
 +
 +                    uint64_t memory;
 +                    totalFileStream >> memory;
-+                    *total = memory;
 +
 +                    std::string usedFile = dir + "/" + drmUsedMemoryFile;
 +                    std::ifstream usedFileStream(usedFile.c_str());
@@ -1019,6 +1021,33 @@ index 000000000..c1949b899
 +
 +                    uint64_t memoryUsed;
 +                    usedFileStream >> memoryUsed;
+
+                    if (is_integrated_gpu) {
+                        std::string totalFile = dir + "/" + drmGTTTotalMemoryFile;
+                        std::ifstream totalFileStream(totalFile.c_str());
+                        if (!totalFileStream.is_open()) {
+                            GGML_LOG_DEBUG("%s Failed to read sysfs node %s\n", __func__, totalFile.c_str());
+                            file.close();
+                            globfree(&glob_result);
+                            return 1;
+                        }
+                        uint64_t gtt;
+                        totalFileStream >> gtt;
+                        std::string usedFile = dir + "/" + drmGTTUsedMemoryFile;
+                        std::ifstream usedFileStream(usedFile.c_str());
+                        if (!usedFileStream.is_open()) {
+                            GGML_LOG_DEBUG("%s Failed to read sysfs node %s\n", __func__, usedFile.c_str());
+                            file.close();
+                            globfree(&glob_result);
+                            return 1;
+                        }
+                        uint64_t gttUsed;
+                        usedFileStream >> gttUsed;
+                        memory += gtt;
+                        memoryUsed += gttUsed;
+                    }
+
+                    *total = memory;
 +                    *free = memory - memoryUsed;
 +
 +                    file.close();
--- a/llama/patches/0028-Add-memory-detection-using-DXGI-PDH.patch
+++ b/llama/patches/0028-Add-memory-detection-using-DXGI-PDH.patch
@@ -24,12 +24,12 @@ index 99ae293cc..9a134b7af 100644
 
 set_target_properties(ggml-base PROPERTIES
 diff --git a/ggml/src/ggml-impl.h b/ggml/src/ggml-impl.h
-index 1c07e767a..0da3e065b 100644
+index dba8f4695..7e17032c7 100644
 --- a/ggml/src/ggml-impl.h
 +++ b/ggml/src/ggml-impl.h
@@ -684,6 +684,9 @@ GGML_API void ggml_nvml_release();
 GGML_API int ggml_hip_mgmt_init();
- GGML_API int ggml_hip_get_device_memory(const char *id, size_t *free, size_t *total);
+ GGML_API int ggml_hip_get_device_memory(const char *id, size_t *free, size_t *total, bool is_integrated_gpu);
 GGML_API void ggml_hip_mgmt_release();
 +GGML_API int ggml_dxgi_pdh_init();
 +GGML_API int ggml_dxgi_pdh_get_device_memory(const char* luid, size_t *free, size_t *total, bool is_integrated_gpu);
@@ -38,7 +38,7 @@ index 1c07e767a..0da3e065b 100644
 #ifdef __cplusplus
 }
 diff --git a/ggml/src/ggml-vulkan/ggml-vulkan.cpp b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
-index d43d46d1d..df79f9f79 100644
+index 0103fd03a..9cc4ebdef 100644
 --- a/ggml/src/ggml-vulkan/ggml-vulkan.cpp
 +++ b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
@@ -74,6 +74,7 @@ DispatchLoaderDynamic & ggml_vk_default_dispatcher();
--- a/llama/patches/0029-ggml-cuda-skip-large-batches.patch
+++ b/llama/patches/0029-ggml-cuda-skip-large-batches.patch
@@ -10,7 +10,7 @@ fallback to cpu
 1 file changed, 3 insertions(+)

 diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
-index 48cdb1dcf..3102d7ea7 100644
+index 334a30135..5c9dfd032 100644
 --- a/ggml/src/ggml-cuda/ggml-cuda.cu
 +++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -4633,6 +4633,9 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
--- a/llama/patches/0030-fix-bakllava-regression.patch
+++ b/llama/patches/0030-fix-bakllava-regression.patch
--- a/llama/patches/0031-win-exit-instead-of-abort.patch
+++ b/llama/patches/0031-win-exit-instead-of-abort.patch
--- a/llm/server.go
+++ b/llm/server.go
@@ -524,8 +524,13 @@ func (s *llamaServer) Load(ctx context.Context, systemInfo ml.SystemInfo, system
 	// Use the size of one layer as a buffer
 	layers := s.ggml.Tensors().GroupLayers()
 	if blk0, ok := layers["blk.0"]; ok {
+		buffer := blk0.Size() + kv[0]
 		for i := range gpus {
-			gpus[i].FreeMemory -= blk0.Size() + kv[0]
+			if gpus[i].FreeMemory > buffer {
+				gpus[i].FreeMemory -= buffer
+			} else {
+				gpus[i].FreeMemory = 0
+			}
 		}
 	} else {
 		slog.Warn("model missing blk.0 layer size")
@@ -575,7 +580,11 @@ func (s *llamaServer) Load(ctx context.Context, systemInfo ml.SystemInfo, system
 			projectorGPU = firstIntegrated
 		}

-		gpus[projectorGPU].FreeMemory -= projectorWeights
+		if gpus[projectorGPU].FreeMemory > projectorWeights {
+			gpus[projectorGPU].FreeMemory -= projectorWeights
+		} else {
+			gpus[projectorGPU].FreeMemory = 0
+		}
 	}

 	var kvTotal uint64
--- a/ml/backend/ggml/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ml/backend/ggml/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -4436,7 +4436,7 @@ static void ggml_backend_cuda_device_get_memory(ggml_backend_dev_t dev, size_t *

 #if defined(GGML_USE_HIP)
    if (ggml_hip_mgmt_init() == 0) {
-        int status = ggml_hip_get_device_memory(ctx->pci_bus_id.c_str(), free, total);
+        int status = ggml_hip_get_device_memory(ctx->pci_bus_id.c_str(), free, total, ctx->integrated != 0);
        if (status == 0) {
            GGML_LOG_DEBUG("%s device %s utilizing AMD specific memory reporting free: %zu total: %zu\n", __func__, ctx->pci_bus_id.c_str(), *free, *total);
            ggml_hip_mgmt_release();
--- a/ml/backend/ggml/ggml/src/ggml-impl.h
+++ b/ml/backend/ggml/ggml/src/ggml-impl.h
@@ -682,7 +682,7 @@ GGML_API int ggml_nvml_init();
 GGML_API int ggml_nvml_get_device_memory(const char *uuid, size_t *free, size_t *total);
 GGML_API void ggml_nvml_release();
 GGML_API int ggml_hip_mgmt_init();
-GGML_API int ggml_hip_get_device_memory(const char *id, size_t *free, size_t *total);
+GGML_API int ggml_hip_get_device_memory(const char *id, size_t *free, size_t *total, bool is_integrated_gpu);
 GGML_API void ggml_hip_mgmt_release();
 GGML_API int ggml_dxgi_pdh_init();
 GGML_API int ggml_dxgi_pdh_get_device_memory(const char* luid, size_t *free, size_t *total, bool is_integrated_gpu);
--- a/ml/backend/ggml/ggml/src/ggml-vulkan/ggml-vulkan.cpp
+++ b/ml/backend/ggml/ggml/src/ggml-vulkan/ggml-vulkan.cpp
@@ -13710,7 +13710,7 @@ void ggml_backend_vk_get_device_memory(ggml_backend_vk_device_context *ctx, size
        switch (props2.properties.vendorID) {
        case VK_VENDOR_ID_AMD:
            if (ggml_hip_mgmt_init() == 0) {
-                int status = ggml_hip_get_device_memory(ctx->pci_id != "" ? ctx->pci_id.c_str() : ctx->uuid.c_str(), free, total);
+                int status = ggml_hip_get_device_memory(ctx->pci_id != "" ? ctx->pci_id.c_str() : ctx->uuid.c_str(), free, total, ctx->is_integrated_gpu);
                if (status == 0) {
                    GGML_LOG_DEBUG("%s device %s utilizing AMD specific memory reporting free: %zu total: %zu\n", __func__, ctx->pci_id != "" ? ctx->pci_id.c_str() : ctx->uuid.c_str(), *free, *total);
                    ggml_hip_mgmt_release();
--- a/ml/backend/ggml/ggml/src/mem_hip.cpp
+++ b/ml/backend/ggml/ggml/src/mem_hip.cpp
@@ -331,7 +331,7 @@ void ggml_hip_mgmt_release() {
    if (gpus != NULL) gpus->pVtbl->Release(gpus); \
    if (gpu != NULL) gpu->pVtbl->Release(gpu)

-int ggml_hip_get_device_memory(const char *id, size_t *free, size_t *total) {
+int ggml_hip_get_device_memory(const char *id, size_t *free, size_t *total, bool is_integrated_gpu) {
    std::lock_guard<std::mutex> lock(ggml_adlx_lock);
    if (adlx.handle == NULL) {
        GGML_LOG_INFO("%s ADLX was not initialized\n", __func__);
@@ -455,13 +455,16 @@ int ggml_hip_mgmt_init() {
    return 0;
 }
 void ggml_hip_mgmt_release() {}
-int ggml_hip_get_device_memory(const char *id, size_t *free, size_t *total) {
+int ggml_hip_get_device_memory(const char *id, size_t *free, size_t *total, bool is_integrated_gpu) {
    GGML_LOG_INFO("%s searching for device %s\n", __func__, id);
    const std::string drmDeviceGlob = "/sys/class/drm/card*/device/uevent";
    const std::string drmTotalMemoryFile = "mem_info_vram_total";
    const std::string drmUsedMemoryFile = "mem_info_vram_used";
+    const std::string drmGTTTotalMemoryFile = "mem_info_gtt_total";
+    const std::string drmGTTUsedMemoryFile = "mem_info_gtt_used";
    const std::string drmUeventPCISlotLabel = "PCI_SLOT_NAME=";

+
    glob_t glob_result;
    glob(drmDeviceGlob.c_str(), GLOB_NOSORT, NULL, &glob_result);

@@ -495,7 +498,6 @@ int ggml_hip_get_device_memory(const char *id, size_t *free, size_t *total) {

                    uint64_t memory;
                    totalFileStream >> memory;
-                    *total = memory;

                    std::string usedFile = dir + "/" + drmUsedMemoryFile;
                    std::ifstream usedFileStream(usedFile.c_str());
@@ -508,6 +510,33 @@ int ggml_hip_get_device_memory(const char *id, size_t *free, size_t *total) {

                    uint64_t memoryUsed;
                    usedFileStream >> memoryUsed;
+
+                    if (is_integrated_gpu) {
+                        std::string totalFile = dir + "/" + drmGTTTotalMemoryFile;
+                        std::ifstream totalFileStream(totalFile.c_str());
+                        if (!totalFileStream.is_open()) {
+                            GGML_LOG_DEBUG("%s Failed to read sysfs node %s\n", __func__, totalFile.c_str());
+                            file.close();
+                            globfree(&glob_result);
+                            return 1;
+                        }
+                        uint64_t gtt;
+                        totalFileStream >> gtt;
+                        std::string usedFile = dir + "/" + drmGTTUsedMemoryFile;
+                        std::ifstream usedFileStream(usedFile.c_str());
+                        if (!usedFileStream.is_open()) {
+                            GGML_LOG_DEBUG("%s Failed to read sysfs node %s\n", __func__, usedFile.c_str());
+                            file.close();
+                            globfree(&glob_result);
+                            return 1;
+                        }
+                        uint64_t gttUsed;
+                        usedFileStream >> gttUsed;
+                        memory += gtt;
+                        memoryUsed += gttUsed;
+                    }
+
+                    *total = memory;
                    *free = memory - memoryUsed;

                    file.close();
--- a/parser/parser.go
+++ b/parser/parser.go
@@ -17,6 +17,7 @@ import (
 	"strings"
 	"sync"

+	"golang.org/x/mod/semver"
 	"golang.org/x/sync/errgroup"
 	"golang.org/x/text/encoding/unicode"
 	"golang.org/x/text/transform"
@@ -104,6 +105,16 @@ func (f Modelfile) CreateRequest(relativeDir string) (*api.CreateRequest, error)
 			req.Renderer = c.Args
 		case "parser":
 			req.Parser = c.Args
+		case "requires":
+			// golang.org/x/mod/semver requires "v" prefix
+			requires := c.Args
+			if !strings.HasPrefix(requires, "v") {
+				requires = "v" + requires
+			}
+			if !semver.IsValid(requires) {
+				return nil, fmt.Errorf("requires must be a valid semver (e.g. 0.14.0)")
+			}
+			req.Requires = strings.TrimPrefix(requires, "v")
 		case "message":
 			role, msg, _ := strings.Cut(c.Args, ": ")
 			messages = append(messages, api.Message{Role: role, Content: msg})
@@ -322,7 +333,7 @@ func (c Command) String() string {
 	switch c.Name {
 	case "model":
 		fmt.Fprintf(&sb, "FROM %s", c.Args)
-	case "license", "template", "system", "adapter", "renderer", "parser":
+	case "license", "template", "system", "adapter", "renderer", "parser", "requires":
 		fmt.Fprintf(&sb, "%s %s", strings.ToUpper(c.Name), quote(c.Args))
 	case "message":
 		role, message, _ := strings.Cut(c.Args, ": ")
@@ -348,7 +359,7 @@ const (
 var (
 	errMissingFrom        = errors.New("no FROM line")
 	errInvalidMessageRole = errors.New("message role must be one of \"system\", \"user\", or \"assistant\"")
-	errInvalidCommand     = errors.New("command must be one of \"from\", \"license\", \"template\", \"system\", \"adapter\", \"renderer\", \"parser\", \"parameter\", or \"message\"")
+	errInvalidCommand     = errors.New("command must be one of \"from\", \"license\", \"template\", \"system\", \"adapter\", \"renderer\", \"parser\", \"parameter\", \"message\", or \"requires\"")
 )

 type ParserError struct {
@@ -608,7 +619,7 @@ func isValidMessageRole(role string) bool {

 func isValidCommand(cmd string) bool {
 	switch strings.ToLower(cmd) {
-	case "from", "license", "template", "system", "adapter", "renderer", "parser", "parameter", "message":
+	case "from", "license", "template", "system", "adapter", "renderer", "parser", "parameter", "message", "requires":
 		return true
 	default:
 		return false
--- a/server/create.go
+++ b/server/create.go
@@ -61,6 +61,7 @@ func (s *Server) CreateHandler(c *gin.Context) {

 	config.Renderer = r.Renderer
 	config.Parser = r.Parser
+	config.Requires = r.Requires

 	for v := range r.Files {
 		if !fs.ValidPath(v) {
@@ -120,7 +121,7 @@ func (s *Server) CreateHandler(c *gin.Context) {
 					ch <- gin.H{"error": err.Error()}
 				}

-				if err == nil && !remote && (config.Renderer == "" || config.Parser == "") {
+				if err == nil && !remote && (config.Renderer == "" || config.Parser == "" || config.Requires == "") {
 					manifest, mErr := ParseNamedManifest(fromName)
 					if mErr == nil && manifest.Config.Digest != "" {
 						configPath, pErr := GetBlobsPath(manifest.Config.Digest)
@@ -134,6 +135,9 @@ func (s *Server) CreateHandler(c *gin.Context) {
 									if config.Parser == "" {
 										config.Parser = baseConfig.Parser
 									}
+									if config.Requires == "" {
+										config.Requires = baseConfig.Requires
+									}
 								}
 								cfgFile.Close()
 							}
--- a/server/routes.go
+++ b/server/routes.go
@@ -1106,6 +1106,7 @@ func GetModelInfo(req api.ShowRequest) (*api.ShowResponse, error) {
 		Messages:     msgs,
 		Capabilities: m.Capabilities(),
 		ModifiedAt:   manifest.fi.ModTime(),
+		Requires:     m.Config.Requires,
 	}

 	if m.Config.RemoteHost != "" {
--- a/types/model/config.go
+++ b/types/model/config.go
@@ -9,6 +9,7 @@ type ConfigV2 struct {
 	FileType      string   `json:"file_type"`  // shown as Quantization Level
 	Renderer      string   `json:"renderer,omitempty"`
 	Parser        string   `json:"parser,omitempty"`
+	Requires      string   `json:"requires,omitempty"`

 	RemoteHost  string `json:"remote_host,omitempty"`
 	RemoteModel string `json:"remote_model,omitempty"`
Author	SHA1	Message	Date
Vallabh Mahajan	18fdcc94e5	docs: fix broken .md links and render issues (#13550 )	2025-12-23 12:44:55 -05:00
Daniel Hiltgen	7ad036992f	amd: use GTT on iGPUs on linux (#13196 ) On Linux, look at the GTT memory information for iGPUs.	2025-12-23 09:30:05 -08:00
Jesse Gross	172b5924af	llm: Avoid integer underflow on llama engine memory layout On the llama engine, when we compute the memory layout, we reserve a buffer to allow for some flexibility for incorrect estimates. This is subtracted from GPU free memory and on GPUs with limited memory, it may underflow. Fixes #13494	2025-12-19 15:48:15 -08:00
Jeffrey Morgan	8852220f59	add REQUIRES command to Modelfile (#13361 )	2025-12-18 13:21:29 -08:00