[GH-ISSUE #15298] Crash: Importing Gemma4 from Safetensor #9787

Closed
opened 2026-04-12 22:40:06 -05:00 by GiteaMirror · 4 comments
Owner

Originally created by @chigkim on GitHub (Apr 3, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/15298

What is the issue?

When I tried to import Gemma4 from safetensor, it just crashes during converting model.

Relevant log output

GOMAXPROCS=1 ollama create gemma4:26b-a4b-heretic -f gemma4-26b.modelfile -q q8_0
gathering model components
copying file sha256:501e322eb3735099d8d32cbc8cb8f606e40001180efb27fe6f8cca0cf28e3bfd 100%
...
converting model
time=2026-04-03T12:44:32.374-04:00 level=ERROR source=server.go:201 msg="ollama exited" err="signal: killed"

OS

No response

GPU

Apple

CPU

Apple

Ollama version

0.20.0

Originally created by @chigkim on GitHub (Apr 3, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/15298 ### What is the issue? When I tried to import Gemma4 from safetensor, it just crashes during converting model. ### Relevant log output ```shell GOMAXPROCS=1 ollama create gemma4:26b-a4b-heretic -f gemma4-26b.modelfile -q q8_0 gathering model components copying file sha256:501e322eb3735099d8d32cbc8cb8f606e40001180efb27fe6f8cca0cf28e3bfd 100% ... converting model time=2026-04-03T12:44:32.374-04:00 level=ERROR source=server.go:201 msg="ollama exited" err="signal: killed" ``` ### OS _No response_ ### GPU Apple ### CPU Apple ### Ollama version 0.20.0
GiteaMirror added the bug label 2026-04-12 22:40:06 -05:00
Author
Owner

@rick-github commented on GitHub (Apr 3, 2026):

GOMAXPROCS needs to be set in the server environment.

<!-- gh-comment-id:4184303286 --> @rick-github commented on GitHub (Apr 3, 2026): `GOMAXPROCS` needs to be set in the server environment.
Author
Owner

@chigkim commented on GitHub (Apr 3, 2026):

Oh, that did it! Thanks!
BTW, is there a way to convert to mlx quant and import?
If I specify -q mxfp8, I get Error: unsupported quantization type MXFP8 - supported types are F32, F16, Q4_K_S, Q4_K_M, Q8_0.

<!-- gh-comment-id:4184405739 --> @chigkim commented on GitHub (Apr 3, 2026): Oh, that did it! Thanks! BTW, is there a way to convert to mlx quant and import? If I specify -q mxfp8, I get `Error: unsupported quantization type MXFP8 - supported types are F32, F16, Q4_K_S, Q4_K_M, Q8_0`.
Author
Owner

@rick-github commented on GitHub (Apr 3, 2026):

ollama create --experimental
<!-- gh-comment-id:4184411115 --> @rick-github commented on GitHub (Apr 3, 2026): ``` ollama create --experimental ```
Author
Owner

@chigkim commented on GitHub (Apr 3, 2026):

Thanks! It said successful, but when I try to use it it throws:
mlx runner failed: Error: unsupported architecture: Gemma4ForConditionalGeneration (exit: exit status 1) (status code: 500)

<!-- gh-comment-id:4184566015 --> @chigkim commented on GitHub (Apr 3, 2026): Thanks! It said successful, but when I try to use it it throws: mlx runner failed: Error: unsupported architecture: Gemma4ForConditionalGeneration (exit: exit status 1) (status code: 500)
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#9787