github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #6629] Fail to Convert Huggingface Llama3.1 with ollama create #66209

New Issue
Closed
opened 2026-05-04 00:50:08 -05:00 by GiteaMirror · 6 comments
GiteaMirror commented 2026-05-04 00:50:08 -05:00
Owner
Copy Link

Originally created by @YueChenkkk on GitHub (Sep 4, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/6629

Originally assigned to: @pdevine on GitHub.

What is the issue?

I downloaded the meta-llama-3.1-8b model from huggingface. [https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct]

And I installed ollama-linux-amd64 (version==0.3.6) manually.

When I start the service with ollama serve &, it seems nothing goes wrong.

But after I build a Modelfile with only FROM . in the huggingface model directory and try to create a ollama model with ollama create my-llama-model, I got the following error message:

usr-yc@lm-machine:~/models/vanilla-llama-3.1-8b$ ollama create my-llama-model
[GIN] 2024/09/04 - 15:12:28 | 200 |      23.352µs |       127.0.0.1 | HEAD     "/"
transferring model data ⠙ [GIN] 2024/09/04 - 15:13:00 | 200 |     168.198µs |       127.0.0.1 | POST     "/api/blobs/sha256:xxxxxxx"
[GIN] 2024/09/04 - 15:13:00 | 200 |    5.022104ms |       127.0.0.1 | POST     "/api/create"
transferring model data 100%
converting model
Error: proto: cannot parse invalid wire-format data

OS

Linux

GPU

Nvidia

CPU

Intel

Ollama version

0.3.6

Originally created by @YueChenkkk on GitHub (Sep 4, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/6629 Originally assigned to: @pdevine on GitHub. ### What is the issue? I downloaded the meta-llama-3.1-8b model from huggingface. [https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct] And I installed ollama-linux-amd64 (version==0.3.6) manually. When I start the service with `ollama serve &`, it seems nothing goes wrong. But after I build a Modelfile with only `FROM .` in the huggingface model directory and try to create a ollama model with `ollama create my-llama-model`, I got the following error message: ``` usr-yc@lm-machine:~/models/vanilla-llama-3.1-8b$ ollama create my-llama-model [GIN] 2024/09/04 - 15:12:28 | 200 | 23.352µs | 127.0.0.1 | HEAD "/" transferring model data ⠙ [GIN] 2024/09/04 - 15:13:00 | 200 | 168.198µs | 127.0.0.1 | POST "/api/blobs/sha256:xxxxxxx" [GIN] 2024/09/04 - 15:13:00 | 200 | 5.022104ms | 127.0.0.1 | POST "/api/create" transferring model data 100% converting model Error: proto: cannot parse invalid wire-format data ``` ### OS Linux ### GPU Nvidia ### CPU Intel ### Ollama version 0.3.6
GiteaMirror added the bug label 2026-05-04 00:50:09 -05:00
GiteaMirror commented 2026-05-04 00:50:25 -05:00
Author
Owner
Copy Link

@YueChenkkk commented on GitHub (Sep 4, 2024):

I've tried to first convert the hf model checkpoint into a gguf file in bf16 and ollama create again.

This time I hit the following error:

usr-yc@lm-machine:~/models/vanilla-llama-3.1-8b-gguf$ ollama create my-llama-model
[GIN] 2024/09/04 - 16:35:30 | 200 |      32.033µs |       127.0.0.1 | HEAD     "/"
transferring model data 99% ⠼ [GIN] 2024/09/04 - 16:36:15 | 201 | 28.804626423s |       127.0.0.1 | POST     "/api/blobs/xxxxxxxx"
transferring model data 100% ⠦ [GIN] 2024/09/04 - 16:36:15 | 200 |   82.992457ms |       127.0.0.1 | POST     "/api/create"
transferring model data 100%
Error: invalid file magic
<!-- gh-comment-id:2328285909 --> @YueChenkkk commented on GitHub (Sep 4, 2024): I've tried to first convert the hf model checkpoint into a gguf file in bf16 and `ollama create` again. This time I hit the following error: ``` usr-yc@lm-machine:~/models/vanilla-llama-3.1-8b-gguf$ ollama create my-llama-model [GIN] 2024/09/04 - 16:35:30 | 200 | 32.033µs | 127.0.0.1 | HEAD "/" transferring model data 99% ⠼ [GIN] 2024/09/04 - 16:36:15 | 201 | 28.804626423s | 127.0.0.1 | POST "/api/blobs/xxxxxxxx" transferring model data 100% ⠦ [GIN] 2024/09/04 - 16:36:15 | 200 | 82.992457ms | 127.0.0.1 | POST "/api/create" transferring model data 100% Error: invalid file magic ```
GiteaMirror commented 2026-05-04 00:50:30 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Sep 10, 2024):

Hey @YueChenkkk There's a bug here where Meta includes two copies of the weights, one in safetensors format and one in pytorch format which is sitting in the directory called original/. As a workaround for now, if you remove the original/ directory it should work.

The second error you got when trying to convert w/ llama.cpp's scripts is due to an invalid conversion. That one will be harder to debug.

<!-- gh-comment-id:2339438185 --> @pdevine commented on GitHub (Sep 10, 2024): Hey @YueChenkkk There's a bug here where Meta includes two copies of the weights, one in _safetensors_ format and one in _pytorch_ format which is sitting in the directory called `original/`. As a workaround for now, if you remove the `original/` directory it should work. The second error you got when trying to convert w/ llama.cpp's scripts is due to an invalid conversion. That one will be harder to debug.
GiteaMirror commented 2026-05-04 00:50:34 -05:00
Author
Owner
Copy Link

@YueChenkkk commented on GitHub (Sep 10, 2024):

Hey @YueChenkkk There's a bug here where Meta includes two copies of the weights, one in safetensors format and one in pytorch format which is sitting in the directory called original/. As a workaround for now, if you remove the original/ directory it should work.

The second error you got when trying to convert w/ llama.cpp's scripts is due to an invalid conversion. That one will be harder to debug.

Thank you so much for the reply!

<!-- gh-comment-id:2339627619 --> @YueChenkkk commented on GitHub (Sep 10, 2024): > Hey @YueChenkkk There's a bug here where Meta includes two copies of the weights, one in _safetensors_ format and one in _pytorch_ format which is sitting in the directory called `original/`. As a workaround for now, if you remove the `original/` directory it should work. > > The second error you got when trying to convert w/ llama.cpp's scripts is due to an invalid conversion. That one will be harder to debug. Thank you so much for the reply!
GiteaMirror commented 2026-05-04 00:50:42 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Sep 10, 2024):

@YueChenkkk this should be fixed as of 0.3.7. You'll need to update ollama and you can check ollama --version to make sure the the client is at least at 0.3.7 (but preferably just upgrade everything to 0.3.10 which is the current latest version).

<!-- gh-comment-id:2341755968 --> @pdevine commented on GitHub (Sep 10, 2024): @YueChenkkk this should be fixed as of `0.3.7`. You'll need to update ollama and you can check `ollama --version` to make sure the the _client_ is at least at `0.3.7` (but preferably just upgrade everything to `0.3.10` which is the current latest version).
GiteaMirror commented 2026-05-04 00:50:50 -05:00
Author
Owner
Copy Link

@samuelvisser commented on GitHub (Mar 24, 2025):

I have this exact same issue with ollama version 0.6.2 and llama model 3.2-3b:
2025-03-24T18:54:02.935992515Z Error: proto: cannot parse invalid wire-format data

Deleting the original folder fixed the problem like suggested (but it means the bug isnt fixed..)

<!-- gh-comment-id:2749121680 --> @samuelvisser commented on GitHub (Mar 24, 2025): I have this exact same issue with ollama version 0.6.2 and llama model 3.2-3b: `2025-03-24T18:54:02.935992515Z Error: proto: cannot parse invalid wire-format data` Deleting the original folder fixed the problem like suggested (but it means the bug isnt fixed..)
GiteaMirror commented 2026-05-04 00:50:57 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Mar 28, 2025):

I think it "unfixed" itself because of the BERT model converter.

<!-- gh-comment-id:2762392715 --> @pdevine commented on GitHub (Mar 28, 2025): I think it "unfixed" itself because of the BERT model converter.
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
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#66209
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?