Error: context deadline exceeded

[aole]

What is the issue?

I keep getting "Error: context deadline exceeded" for a lot of models that I download from hf.

ollama run hf.co/DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF:Q4_K_M
pulling manifest
pulling 8440f2a076f1: 100% ▕██████████████████████████████████████████████████████████▏ 18 GB
pulling eacf610d1ee4: 100% ▕██████████████████████████████████████████████████████████▏ 927 MB
Error: context deadline exceeded

Relevant log output

time=2026-07-30T10:05:15.022-04:00 level=INFO source=routes.go:1947 msg="server config" env="map[CUDA_VISIBLE_DEVICES: GGML_VK_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: LLAMA_ARG_FIT: LLAMA_ARG_FIT_TARGET: NO_PROXY: OLLAMA_CONTEXT_LENGTH:32768 OLLAMA_DEBUG:INFO OLLAMA_DEBUG_LOG_REQUESTS:false OLLAMA_EDITOR: OLLAMA_FLASH_ATTENTION:false OLLAMA_GO_TEMPLATE:true OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://127.0.0.1:11434 OLLAMA_IGPU_ENABLE: OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE: OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MAX_TRANSFER_STREAMS:4 OLLAMA_MODELS:D:\\OllamaModels OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NO_CLOUD:true OLLAMA_NUM_PARALLEL:1 OLLAMA_ORIGINS:[* http://localhost https://localhost http://localhost:* https://localhost:* http://127.0.0.1 https://127.0.0.1 http://127.0.0.1:* https://127.0.0.1:* http://0.0.0.0 https://0.0.0.0 http://0.0.0.0:* https://0.0.0.0:* app://* file://* tauri://* vscode-webview://* vscode-file://*] OLLAMA_REMOTES:[ollama.com] OLLAMA_SCHED_SPREAD:false OLLAMA_VULKAN:true ROCR_VISIBLE_DEVICES:]"
time=2026-07-30T10:05:15.022-04:00 level=INFO source=routes.go:1949 msg="Ollama cloud disabled: true"
time=2026-07-30T10:05:15.067-04:00 level=INFO source=images.go:883 msg="total blobs: 77"
time=2026-07-30T10:05:15.073-04:00 level=INFO source=images.go:890 msg="total unused blobs removed: 0"
time=2026-07-30T10:05:15.075-04:00 level=INFO source=routes.go:2004 msg="Listening on 127.0.0.1:11434 (version 0.32.5)"
time=2026-07-30T10:05:15.075-04:00 level=INFO source=model_recommendations.go:177 msg="model recommendations cache sleep scheduled" wait=4h6m45.218682378s consecutive_failures=0
time=2026-07-30T10:05:15.085-04:00 level=INFO source=runner.go:60 msg="discovering available GPUs..."
time=2026-07-30T10:05:17.274-04:00 level=INFO source=model_list_cache.go:112 msg="model list cache hydration complete" models=22 failures=0 elapsed=2.1988268s
time=2026-07-30T10:05:22.044-04:00 level=INFO source=types.go:32 msg="inference compute" id=0 filter_id=0 library=CUDA compute=8.9 name=CUDA0 description="NVIDIA GeForce RTX 4060 Ti" libdirs=ollama,cuda_v13 driver=13.3 pci_id=0000:01:00.0 type=discrete total="16.0 GiB" available="14.9 GiB"
time=2026-07-30T10:05:22.045-04:00 level=INFO source=types.go:32 msg="inference compute" id=1 filter_id=1 library=CUDA compute=7.5 name=CUDA1 description="NVIDIA GeForce RTX 2080 SUPER" libdirs=ollama,cuda_v13 driver=13.3 pci_id=0000:02:00.0 type=discrete total="8.0 GiB" available="6.9 GiB"
time=2026-07-30T10:05:22.045-04:00 level=INFO source=routes.go:2054 msg="vram-based default context" total_vram="24.0 GiB" default_num_ctx=32768
[GIN] 2026/07/30 - 10:05:22 | 200 |            0s |       127.0.0.1 | GET      "/api/version"
time=2026-07-30T14:12:00.295-04:00 level=INFO source=model_recommendations.go:177 msg="model recommendations cache sleep scheduled" wait=3h38m57.720840545s consecutive_failures=0
time=2026-07-30T17:50:58.016-04:00 level=INFO source=model_recommendations.go:177 msg="model recommendations cache sleep scheduled" wait=3h26m34.305424689s consecutive_failures=0
[GIN] 2026/07/30 - 19:49:16 | 200 |            0s |       127.0.0.1 | HEAD     "/"
[GIN] 2026/07/30 - 19:49:16 | 404 |    138.7296ms |       127.0.0.1 | POST     "/api/show"
time=2026-07-30T19:49:19.456-04:00 level=INFO source=download.go:181 msg="downloading 8440f2a076f1 in 19 1 GB part(s)"
time=2026-07-30T19:54:21.537-04:00 level=INFO source=download.go:181 msg="downloading eacf610d1ee4 in 10 100 MB part(s)"
time=2026-07-30T19:55:26.196-04:00 level=INFO source=download.go:181 msg="downloading 2a30fe37ad16 in 1 479 B part(s)"
time=2026-07-30T19:55:56.197-04:00 level=INFO source=images.go:1263 msg="request failed: Get \"https://hf.co/v2/DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF/blobs/sha256:2a30fe37ad16e9f7decd571f88282b1d59338c45b7bff0173a9e14476eaf7742\": context deadline exceeded"
time=2026-07-30T19:55:56.197-04:00 level=WARN source=download.go:260 msg="failed to get direct URL; backing off and retrying" err="Get \"https://hf.co/v2/DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF/blobs/sha256:2a30fe37ad16e9f7decd571f88282b1d59338c45b7bff0173a9e14476eaf7742\": context deadline exceeded"
[GIN] 2026/07/30 - 19:55:56 | 200 |         6m39s |       127.0.0.1 | POST     "/api/pull"

OS

Windows

GPU

Nvidia

CPU

Intel

Ollama version

0.32.5

[aole]

I am able to download the 1kb file (as mentioned in the logs) separately and put it alongside the other blobs. The model then works.

[sammyvoncheese]

have the same issue.

[adarshsm]

I think the logs point at a specific structural issue rather than just a slow network. The failure is in getDirectURL resolution (server/download.go):

directURL, err := func() (*url.URL, error) {
    ctx, cancel := context.WithTimeout(ctx, 30*time.Second)
    defer cancel()

    backoff := newBackoff(10 * time.Second)
    for {
        // ...
        resp, err := makeRequestWithRetry(ctx, http.MethodGet, requestURL, nil, nil, newOpts)
        if err != nil {
            slog.Warn("failed to get direct URL; backing off and retrying", "err", err)
            if err := backoff(ctx); err != nil {
                return nil, err
            }
            continue
        }
        // ...
    }
}()

The whole retry-with-backoff loop shares a single 30s context, and the underlying request has no timeout of its own — makeRequest builds its client as &http.Client{CheckRedirect: ...} with no Timeout. So when the direct-URL GET to hf.co hangs, the first attempt consumes the entire 30s budget on its own. The context then expires, backoff(ctx) sees ctx.Err() != nil and returns immediately, and the loop never gets a real second attempt. That matches the log timing exactly — the request starts at 19:55:26 and fails at 19:55:56, 30s later, with the "backing off and retrying" line appearing only once.

That also explains why it's the tiny 479 B blob and why @aole's workaround (fetching the same file manually) works: it isn't a download-size issue, it's the CDN redirect-URL resolution transiently hanging, and fetching seconds later succeeds. The retry loop is meant to absorb exactly that, but as written a single hang defeats it because one attempt can eat the whole budget.

A fix that would make the existing retry actually resilient: give each attempt its own bounded timeout (e.g. wrap each makeRequestWithRetry call in its own context.WithTimeout) inside a larger overall budget, so one hang is capped and the loop can genuinely retry. Happy to put up a PR along those lines if that direction sounds right — I can drive a hung-server case through the existing testMakeRequestDialContext hook for a regression test.

[realAbitbol]

I'm having the same problem

[HiddenHari]

I am able to download the 1kb file (as mentioned in the logs) separately and put it alongside the other blobs. The model then works.

If this doesn't work, another temporary bypass is to just use ollama create xxx -f modelfile, and point to the downloaded large blob file (size of the actual model) in the modelfile

FROM \.ollama\models\blobs\sha256-asdfasdf

[jbs123]

same issue

[akai-hana]

having this same issue, bump

[akai-hana]

For anyone going through this thread looking for a quick fix; to expand on what @aole said, here's a concrete walkthrough:

---

1. Run ollama serve on a terminal, then ollama pull on another.
2. Have it fail, and then go back to the terminal where you ran ollama serve, looking for output similar to this one (this is what I got in my case):

time=2026-08-07T21:49:59.299+02:00 level=INFO source=images.go:1263 msg="request failed: Get \"https://hf.co/v2/DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF/blobs/sha256:2a30fe37ad16e9f7decd571f88282b1d59338c45b7bff0173a9e14476eaf7742\": context deadline exceeded"
time=2026-08-07T21:49:59.299+02:00 level=WARN source=download.go:260 msg="failed to get direct URL; backing off and retrying" err="Get \"https://hf.co/v2/DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF/blobs/sha256:2a30fe37ad16e9f7decd571f88282b1d59338c45b7bff0173a9e14476eaf7742\": context deadline exceeded"

3. Notice the parts where it says msg="request failed: Get \"<URL>\": context deadline exceeded": grab the , and wget it, or download it through a browser or however else. In my case, I got a file called sha256:<stuff>.
4. Go to ~/.ollama/models/blobs (or your equivalent if you're running it differently), and ls for the part: see whether there's any already existing files that match it (in my case I found 3, some that said "partial", and one that didn't). Delete these files.
5. Drop the file you downloaded here, and rename it from sha256:<stuff> to sha256-<stuff>, replacing the : with -.
6. Profit! It should work.

[RerankerGuo]

A focused fix is available in #17551, based on @adarshsm's root-cause analysis above.

It preserves the existing 30-second overall budget while limiting each direct URL attempt to 10 seconds, allowing the intended retry to occur. The regression test fails on main with one request and passes with the patch using two requests.

If anyone still reproducing this on Windows can test the branch, that result would help validate the fix against the reported environment.

[vortexau]

I've downloaded the extra file, and renamed it as above - however it won't install/run the model if I re-run the ollama run command. I get the same context deadline exceeded error, and the client exits. If I pull tinyllama, I get the verifying sha256 digest steps, writing manafest, success.

$ ollama run hf.co/DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF:Q4_K_M
pulling manifest 
pulling 8440f2a076f1: 100% ▕████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▏  18 GB                         
pulling eacf610d1ee4: 100% ▕████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▏ 927 MB                         
Error: context deadline exceeded

$ ollama list
NAME    ID    SIZE    MODIFIED 

-rw-rw-r-- 1 vortex vortex         479 Aug 16 18:21 sha256-2a30fe37ad16e9f7decd571f88282b1d59338c45b7bff0173a9e14476eaf7742
-rw-r--r-- 1 vortex vortex 18047255072 Aug 16 18:18 sha256-8440f2a076f149f65bf16975f3a08359936ba2658fc1827b51a30b717de49acb
-rw-r--r-- 1 vortex vortex   927607360 Aug 16 18:18 sha256-eacf610d1ee4bd5ed0197a0777dd8f4fceb8eefa27009067c7d496cb68fbde45

$ sha256sum sha256-2a30fe37ad16e9f7decd571f88282b1d59338c45b7bff0173a9e14476eaf7742
2a30fe37ad16e9f7decd571f88282b1d59338c45b7bff0173a9e14476eaf7742  sha256-2a30fe37ad16e9f7decd571f88282b1d59338c45b7bff0173a9e14476eaf7742

Are there any other steps I can take? I don't rully understand the steps in the comment above.

[Master-Antonio]

Same error, cant install from huggingface.

[jasonhejiahuan]

I am able to download the 1kb file (as mentioned in the logs) separately and put it alongside the other blobs. The model then works.

Same error, it will also cause unused model file been removed automatically, logged total unused blobs removed: 1 and the full model file need to be downloaded again. OLLAMA_NOPRUNE=1parameter need to be set after manually install the small file