Add conversion overrides for Llama, Qwen3, and Gemma 4 models so they roundtrip properly#677
Add conversion overrides for Llama, Qwen3, and Gemma 4 models so they roundtrip properly#677finbarrtimbers wants to merge 4 commits into
Conversation
finbarrtimbers
changed the title
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: f1b69d796a
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| hf_config = AutoConfig.from_pretrained(model_id) | ||
| hf_model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float32) |
There was a problem hiding this comment.
These helpers are called by the new Qwen3/Gemma tests without any HF_TOKEN/network skip, so a normal pytest src/test/nn/hf/convert_test.py now tries to download Qwen/Qwen3-0.6B and google/gemma-3-270m during unit tests. In offline CI or developer environments this fails before testing the converter, and even with network it pulls hundreds of MB; the existing golden tests gate the same Hub-backed coverage behind HF_TOKEN, so these should use a local/tiny config or be similarly skipped/marked.
Useful? React with 👍 / 👎.
2 participants
No description provided.