github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 16:40:08 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #10125] qwen2.5-coder-cline:14b not using NVIDIA GPU #53156

New Issue
Closed
opened 2026-04-29 02:08:39 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-04-29 02:08:39 -05:00
Owner
Copy Link

Originally created by @SvenMeyer on GitHub (Apr 4, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/10125

What is the issue?

qwen2.5-coder:32b-instruct-q8_0 offloads as many layers as possible to the 8 GB GPU

qwen2.5-coder-cline:7b offloads some layers to the GPU, but actually there should be space for more

qwen2.5-coder-cline:14b (and the 32b model) do not use the GPU at all, put all layers just into CPU RAM

Something went wrong when the cline derivative of the model was created ?

I even created a new modelfile which sets mmap to true, but that did not change anything.

Relevant log output


$ ollama run --verbose maryasov/qwen2.5-coder:32b-instruct-q8_0

load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: offloading 11 repeating layers to GPU
load_tensors: offloaded 11/65 layers to GPU
load_tensors:        CUDA0 model buffer size =  5435.42 MiB
load_tensors:   CPU_Mapped model buffer size = 33202.08 MiB


$ ollama run --verbose maryasov/qwen2.5-coder-cline:7b

load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: offloading 8 repeating layers to GPU
load_tensors: offloaded 8/29 layers to GPU
load_tensors:        CUDA0 model buffer size =  1103.35 MiB
load_tensors:   CPU_Mapped model buffer size =  4460.45 MiB


$ ollama run --verbose maryasov/qwen2.5-coder-cline:14b

load_tensors: loading model tensors, this can take a while... (mmap = false)
load_tensors:          CPU model buffer size =  8566.04 MiB

OS

Linux

GPU

Nvidia

CPU

Intel

Ollama version

0.6.2

Originally created by @SvenMeyer on GitHub (Apr 4, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/10125 ### What is the issue? `qwen2.5-coder:32b-instruct-q8_0` offloads as many layers as possible to the 8 GB GPU `qwen2.5-coder-cline:7b` offloads some layers to the GPU, but actually there should be space for more `qwen2.5-coder-cline:14b` (and the `32b` model) do not use the GPU at all, put all layers just into CPU RAM Something went wrong when the `cline` derivative of the model was created ? I even created a new modelfile which sets `mmap` to true, but that did not change anything. ### Relevant log output ```shell $ ollama run --verbose maryasov/qwen2.5-coder:32b-instruct-q8_0 load_tensors: loading model tensors, this can take a while... (mmap = true) load_tensors: offloading 11 repeating layers to GPU load_tensors: offloaded 11/65 layers to GPU load_tensors: CUDA0 model buffer size = 5435.42 MiB load_tensors: CPU_Mapped model buffer size = 33202.08 MiB $ ollama run --verbose maryasov/qwen2.5-coder-cline:7b load_tensors: loading model tensors, this can take a while... (mmap = true) load_tensors: offloading 8 repeating layers to GPU load_tensors: offloaded 8/29 layers to GPU load_tensors: CUDA0 model buffer size = 1103.35 MiB load_tensors: CPU_Mapped model buffer size = 4460.45 MiB $ ollama run --verbose maryasov/qwen2.5-coder-cline:14b load_tensors: loading model tensors, this can take a while... (mmap = false) load_tensors: CPU model buffer size = 8566.04 MiB ``` ### OS Linux ### GPU Nvidia ### CPU Intel ### Ollama version 0.6.2
GiteaMirror added the bugneeds more info labels 2026-04-29 02:08:40 -05:00
GiteaMirror commented 2026-04-29 02:08:43 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Apr 7, 2025):

Modelfile?

<!-- gh-comment-id:2782998269 --> @rick-github commented on GitHub (Apr 7, 2025): Modelfile?
GiteaMirror commented 2026-04-29 02:08:44 -05:00
Author
Owner
Copy Link

@SvenMeyer commented on GitHub (Apr 8, 2025):

@rick-github As the 32b version showed the same problem, I paste that here. It was the last one I tried - unmodified, as downloaded from ollama (only reformatted here for better readability).

https://ollama.com/maryasov/qwen2.5-coder-cline

{
    "schemaVersion": 2,
    "mediaType": "application/vnd.docker.distribution.manifest.v2+json",
    "config": {
        "mediaType": "application/vnd.docker.container.image.v1+json",
        "digest": "sha256:5df93646d8858926437da52dca2ab9c23abd0391145aa087d4ae3b20cb50ecb1",
        "size": 488
    },
    "layers": [
        {
            "mediaType": "application/vnd.ollama.image.model",
            "digest": "sha256:ac3d1ba8aa77755dab3806d9024e9c385ea0d5b412d6bdf9157f8a4a7e9fc0d9",
            "size": 19851336384,
            "from": "/usr/share/ollama/.ollama/models/blobs/sha256-ac3d1ba8aa77755dab3806d9024e9c385ea0d5b412d6bdf9157f8a4a7e9fc0d9"
        },
        {
            "mediaType": "application/vnd.ollama.image.template",
            "digest": "sha256:3852d4fe7f87a6e29157b211ff0b037d2c0624023b6c849599e9b4e9148316f8",
            "size": 1899
        },
        {
            "mediaType": "application/vnd.ollama.image.license",
            "digest": "sha256:832dd9e00a68dd83b3c3fb9f5588dad7dcf337a0db50f7d9483f310cd292e92e",
            "size": 11343
        },
        {
            "mediaType": "application/vnd.ollama.image.params",
            "digest": "sha256:8b2542ec39db97e0dad14ab4fe6dd6301e127390ca064522c9ce752a26f1526e",
            "size": 119
        }
    ]
}

Then tried to use it with this Modelfile

FROM maryasov/qwen2.5-coder-cline:32b
PARAMETER use_mmap true

... which did not change anything.

<!-- gh-comment-id:2785216625 --> @SvenMeyer commented on GitHub (Apr 8, 2025): @rick-github As the 32b version showed the same problem, I paste that here. It was the last one I tried - unmodified, as downloaded from ollama (only reformatted here for better readability). [https://ollama.com/maryasov/qwen2.5-coder-cline](https://ollama.com/maryasov/qwen2.5-coder-cline) ``` { "schemaVersion": 2, "mediaType": "application/vnd.docker.distribution.manifest.v2+json", "config": { "mediaType": "application/vnd.docker.container.image.v1+json", "digest": "sha256:5df93646d8858926437da52dca2ab9c23abd0391145aa087d4ae3b20cb50ecb1", "size": 488 }, "layers": [ { "mediaType": "application/vnd.ollama.image.model", "digest": "sha256:ac3d1ba8aa77755dab3806d9024e9c385ea0d5b412d6bdf9157f8a4a7e9fc0d9", "size": 19851336384, "from": "/usr/share/ollama/.ollama/models/blobs/sha256-ac3d1ba8aa77755dab3806d9024e9c385ea0d5b412d6bdf9157f8a4a7e9fc0d9" }, { "mediaType": "application/vnd.ollama.image.template", "digest": "sha256:3852d4fe7f87a6e29157b211ff0b037d2c0624023b6c849599e9b4e9148316f8", "size": 1899 }, { "mediaType": "application/vnd.ollama.image.license", "digest": "sha256:832dd9e00a68dd83b3c3fb9f5588dad7dcf337a0db50f7d9483f310cd292e92e", "size": 11343 }, { "mediaType": "application/vnd.ollama.image.params", "digest": "sha256:8b2542ec39db97e0dad14ab4fe6dd6301e127390ca064522c9ce752a26f1526e", "size": 119 } ] } ``` Then tried to use it with this Modelfile ``` FROM maryasov/qwen2.5-coder-cline:32b PARAMETER use_mmap true ``` ... which did not change anything.
GiteaMirror commented 2026-04-29 02:08:45 -05:00
Author
Owner
Copy Link

@SvenMeyer commented on GitHub (Apr 8, 2025):

same result with version 0.6.5

<!-- gh-comment-id:2785221529 --> @SvenMeyer commented on GitHub (Apr 8, 2025): same result with version 0.6.5
GiteaMirror commented 2026-04-29 02:08:45 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Apr 8, 2025):

From the params blob:

{
    "num_ctx": 32768,
    "stop": [
        "<|im_start|>",
        "<|im_end|>",
        "<|endoftext|>"
    ],
    "temperature": 0
}

A 32K context window requires 8G, which may be too large to be supported along with the model weights and computation graph on your GPU. Server logs may show details.

<!-- gh-comment-id:2786965550 --> @rick-github commented on GitHub (Apr 8, 2025): From the `params` blob: ``` { "num_ctx": 32768, "stop": [ "<|im_start|>", "<|im_end|>", "<|endoftext|>" ], "temperature": 0 } ``` A 32K context window requires 8G, which may be too large to be supported along with the model weights and computation graph on your GPU. [Server logs](https://github.com/ollama/ollama/blob/main/docs/troubleshooting.md#how-to-troubleshoot-issues) may show details.
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label bug needs more info
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#53156
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?