github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-16 05:05:37 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #13080] How to compute hash for model download verification ? #86347

New Issue
Closed
opened 2026-05-10 03:09:44 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-05-10 03:09:44 -05:00
Owner
Copy Link

Originally created by @FrouxBY on GitHub (Nov 13, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/13080

Hello,

Using Ollama to pull the models, there is no way to be sure that the weights downloaded are actually the one advertised on the Ollama website,
There seems to be hash that goes with every checkpoint, but there are no instruction on how to compute locally this hash to see wether it match the one provided on ollama.com;

Additionnaly this could ensure the model integrity on my side, in case of malicious activity,

Is there a way to ensure the integrity of the weights ?

Thanks for your help

Originally created by @FrouxBY on GitHub (Nov 13, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/13080 Hello, Using Ollama to pull the models, there is no way to be sure that the weights downloaded are actually the one advertised on the Ollama website, There seems to be hash that goes with every checkpoint, but there are no instruction on how to compute locally this hash to see wether it match the one provided on ollama.com; Additionnaly this could ensure the model integrity on my side, in case of malicious activity, Is there a way to ensure the integrity of the weights ? Thanks for your help
GiteaMirror commented 2026-05-10 03:09:48 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Nov 13, 2025):

https://sha256sum.com/

The filename of the blob is the SHA256 of the contents. You can recompute the hash to verify the contents are correct, and then compare the hash to the value in the model card in the ollama library to verify it's the same file.

<!-- gh-comment-id:3528785941 --> @rick-github commented on GitHub (Nov 13, 2025): https://sha256sum.com/ The filename of the blob is the SHA256 of the contents. You can recompute the hash to verify the contents are correct, and then compare the hash to the value in the model card in the ollama library to verify it's the same file.
GiteaMirror commented 2026-05-10 03:09:49 -05:00
Author
Owner
Copy Link

@FrouxBY commented on GitHub (Nov 17, 2025):

Thank you,
I was missing that the hash on the website is the hash of the manifest file for the model,

Thanks for your help:)

<!-- gh-comment-id:3540730006 --> @FrouxBY commented on GitHub (Nov 17, 2025): Thank you, I was missing that the hash on the website is the hash of the manifest file for the model, Thanks for your help:)
GiteaMirror commented 2026-05-10 03:09:50 -05:00
Author
Owner
Copy Link

@fallrisk-ai commented on GitHub (May 5, 2026):

@FrouxBY I built Trustfall Lite, an open-source CLI that scans local Hugging Face and Ollama model caches, computes artifact hashes, and checks them against a signed public registry.

Install:

pipx install fallrisk-trustfall
trustfall scan

It reports four states:

verified – artifact hash matches a signed registry record
unknown_variant – model name appears in the registry, but local bytes differ
not_enrolled – no registry record exists
pilot_available – queued for enrollment

The important boundary: Trustfall Lite verifies artifact identity, not runtime identity. It answers “do the bytes on disk match a signed record?” It does not claim the running process is structurally the same model. That is a separate layer.

The registry is public and signed:
https://attest.fallrisk.ai/registry.json

Source:
https://github.com/fallrisk-ai/trustfall-lite

PyPI:
https://pypi.org/project/fallrisk-trustfall/

Research / evidence boundary:
https://fallrisk.ai/research/

Founder scan:
https://fallrisk.ai/articles/founder-scan

There is also a local-only mode if you do not want hash lookups sent to the API:

trustfall registry --refresh
trustfall scan --local-only

I’m especially interested in feedback from people running local model fleets, internal model caches, or AI supply-chain/security tooling.

<!-- gh-comment-id:4382635885 --> @fallrisk-ai commented on GitHub (May 5, 2026): @FrouxBY I built Trustfall Lite, an open-source CLI that scans local Hugging Face and Ollama model caches, computes artifact hashes, and checks them against a signed public registry. Install: > pipx install fallrisk-trustfall > trustfall scan It reports four states: > verified – artifact hash matches a signed registry record > unknown_variant – model name appears in the registry, but local bytes differ > not_enrolled – no registry record exists > pilot_available – queued for enrollment The important boundary: Trustfall Lite verifies artifact identity, not runtime identity. It answers “do the bytes on disk match a signed record?” It does not claim the running process is structurally the same model. That is a separate layer. > The registry is public and signed: > https://attest.fallrisk.ai/registry.json > > Source: > https://github.com/fallrisk-ai/trustfall-lite > > PyPI: > https://pypi.org/project/fallrisk-trustfall/ > > Research / evidence boundary: > https://fallrisk.ai/research/ > > Founder scan: > https://fallrisk.ai/articles/founder-scan There is also a local-only mode if you do not want hash lookups sent to the API: > trustfall registry --refresh > trustfall scan --local-only I’m especially interested in feedback from people running local model fleets, internal model caches, or AI supply-chain/security tooling.
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
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#86347
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?