[Misc] Merge diffusion forward context #582
Conversation
![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.
ℹ️ About Codex in GitHub
Codex has been enabled to automatically 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 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| yield | ||
| finally: | ||
| _forward_context = prev_context | ||
| get_cached_compilation_config.cache_clear() |
There was a problem hiding this comment.
override_forward_context calls get_cached_compilation_config.cache_clear() when exiting (lines 72-75), but get_cached_compilation_config is now a plain function without an lru_cache, so it has no cache_clear attribute. Any use of set_forward_context/override_forward_context will raise AttributeError on exit, aborting model initialization and tests immediately. Please restore caching or drop the cache_clear call.
Useful? React with 👍 / 👎.
|
cc @wtomin |
|
resolve the conflict please since #597 merged |
done. |
Signed-off-by: iwzbi <[email protected]>
3 participants
Purpose
There are 2 omni_diffusion_config context now, simplify into one. Remove
_current_omni_diffusion_configindata.pyand use_forward_contextinforward_context.pyfor all.However, there is an exception: when using vllm components, and vllm internally calls
get_current_vllm_config, but vllm-omni does not explicitly callset_current_vllm_config, therefore internally vllm cannot obtain the correct config.