Where does ollama-python saves the pulled model? #107

New Issue

2026-02-15T16:28:01-05:00

yindo commented

2026-02-15 16:28:01 -05:00

Originally created by @Lopa07 on GitHub (Jul 1, 2024).

Where does ollama-python saves the pulled model using ollama.pull('llava')? I tried to set the environment variable OLLAMA_MODELS similar to Ollama-cli. But it is not using the path provided in OLLAMA_MODELS.

Originally created by @Lopa07 on GitHub (Jul 1, 2024). Where does ollama-python saves the pulled model using `ollama.pull('llava')`? I tried to set the environment variable `OLLAMA_MODELS` similar to Ollama-cli. But it is not using the path provided in `OLLAMA_MODELS`.

yindo commented

2026-02-15 16:28:01 -05:00

@anthonywu commented on GitHub (Jul 31, 2024):

I tried to set the environment variable

common mistake is not have export OLLAMA_MODELS=<value>ed the variable when you set it, I'd double check that.

The model dir on macOS is ~/.ollama/models, should be similar on Linux systems, and should be %USERPROFILE% on Windows (I don't dev on Windows, cannot confirm for you)

On Mac, the model files are stored in chunks in ~/.ollama/models/blobs but they are sha256--prefixed directory names that are not readable. To match which one matches your model you need to first check the manifest which looks like this for today's llama3:

{
  "schemaVersion": 2,
  "mediaType": "application/vnd.docker.distribution.manifest.v2+json",
  "config": {
    "mediaType": "application/vnd.docker.container.image.v1+json",
    "digest": "sha256:3f8eb4da87fa7a3c9da615036b0dc418d31fef2a30b115ff33562588b32c691d",
    "size": 485
  },
  "layers": [
    {
      "mediaType": "application/vnd.ollama.image.model",
      "digest": "sha256:6a0746a1ec1aef3e7ec53868f220ff6e389f6f8ef87a01d77c96807de94ca2aa",
      "size": 4661211424
    },
    {
      "mediaType": "application/vnd.ollama.image.license",
      "digest": "sha256:4fa551d4f938f68b8c1e6afa9d28befb70e3f33f75d0753248d530364aeea40f",
      "size": 12403
    },
    {
      "mediaType": "application/vnd.ollama.image.template",
      "digest": "sha256:8ab4849b038cf0abc5b1c9b8ee1443dca6b93a045c2272180d985126eb40bf6f",
      "size": 254
    },
    {
      "mediaType": "application/vnd.ollama.image.params",
      "digest": "sha256:577073ffcc6ce95b9981eacc77d1039568639e5638e83044994560d9ef82ce1b",
      "size": 110
    }
  ]
}

note the layers items where mediaType is application/vnd.ollama.image.model, the digest is sha256:6a0746a1ec1aef3e7ec53868f220ff6e389f6f8ef87a01d77c96807de94ca2aa. You convert that : to - and you get the blobs/ subdirectory name.

To piece this altogether, if you have jq (or similar json parser installed), you can confirm this is a big model file previously downloaded:

# this assumes the first `layers` item represents `image.model`
du -h ~/.ollama/models/blobs/$(jq -r '.layers[0].digest' ~/.ollama/models/manifests/registry.ollama.ai/library/llama3/latest | sed s/:/-/)
4.3G	/Users/yourusername/.ollama/models/blobs/sha256-6a0746a1ec1aef3e7ec53868f220ff6e389f6f8ef87a01d77c96807de94ca2aa

this matches the reported model size:

replace llama3/latest with your model's name and tag to get the same outcome

if you successfully set OLLAMA_MODELS to some other path, and don't want to re-download the files, I think you can just manually mv the directory to your new preferred path

hope this helps

@anthonywu commented on GitHub (Jul 31, 2024): > I tried to set the environment variable common mistake is not have `export OLLAMA_MODELS=<value>`ed the variable when you set it, I'd double check that. ----- The model dir on macOS is `~/.ollama/models`, should be similar on Linux systems, and should be `%USERPROFILE%` on Windows (I don't dev on Windows, cannot confirm for you) On Mac, the model files are stored in chunks in `~/.ollama/models/blobs` but they are `sha256-`-prefixed directory names that are not readable. To match which one matches your model you need to first check the manifest which looks like this for today's `llama3`: ```json { "schemaVersion": 2, "mediaType": "application/vnd.docker.distribution.manifest.v2+json", "config": { "mediaType": "application/vnd.docker.container.image.v1+json", "digest": "sha256:3f8eb4da87fa7a3c9da615036b0dc418d31fef2a30b115ff33562588b32c691d", "size": 485 }, "layers": [ { "mediaType": "application/vnd.ollama.image.model", "digest": "sha256:6a0746a1ec1aef3e7ec53868f220ff6e389f6f8ef87a01d77c96807de94ca2aa", "size": 4661211424 }, { "mediaType": "application/vnd.ollama.image.license", "digest": "sha256:4fa551d4f938f68b8c1e6afa9d28befb70e3f33f75d0753248d530364aeea40f", "size": 12403 }, { "mediaType": "application/vnd.ollama.image.template", "digest": "sha256:8ab4849b038cf0abc5b1c9b8ee1443dca6b93a045c2272180d985126eb40bf6f", "size": 254 }, { "mediaType": "application/vnd.ollama.image.params", "digest": "sha256:577073ffcc6ce95b9981eacc77d1039568639e5638e83044994560d9ef82ce1b", "size": 110 } ] } ``` note the `layers` items where `mediaType` is `application/vnd.ollama.image.model`, the digest is `sha256:6a0746a1ec1aef3e7ec53868f220ff6e389f6f8ef87a01d77c96807de94ca2aa`. You convert that `:` to `-` and you get the `blobs/` subdirectory name. To piece this altogether, if you have `jq` (or similar json parser installed), you can confirm this is a big model file previously downloaded: ```sh # this assumes the first `layers` item represents `image.model` du -h ~/.ollama/models/blobs/$(jq -r '.layers[0].digest' ~/.ollama/models/manifests/registry.ollama.ai/library/llama3/latest | sed s/:/-/) 4.3G /Users/yourusername/.ollama/models/blobs/sha256-6a0746a1ec1aef3e7ec53868f220ff6e389f6f8ef87a01d77c96807de94ca2aa ``` this matches the reported model size: replace `llama3/latest` with your model's name and tag to get the same outcome ----- if you successfully set `OLLAMA_MODELS` to some other path, and don't want to re-download the files, I think you can just manually `mv` the directory to your new preferred path hope this helps

yindo referenced this issue

2026-02-15 16:30:07 -05:00

[PR #107] [MERGED] Bump pytest from 7.4.4 to 8.1.1 #364

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: ollama/ollama-python#107