Skip to content

fix(transformers): include image entries in chat-template content for VLM #6

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
AlanKharebov wants to merge 1 commit into from
Closed

fix(transformers): include image entries in chat-template content for VLM #6

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 11 additions & 1 deletion src/multihead/adapters/transformers_adapter.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -143,7 +143,17 @@ async def generate(self, prompt: str, **kwargs: Any) -> dict[str, Any]:
# Use chat template when available (required for Qwen3, etc.)
tokenizer = self._tokenizer or self._processor
if not skip_template and tokenizer is not None and hasattr(tokenizer, "apply_chat_template"):
messages = [{"role": "user", "content": prompt}]
# VLM with images: content must be a list including image placeholders
# so the chat template emits <|vision_start|><|image_pad|><|vision_end|>
# tokens that bind to the image features in the processor() call below.
images_for_template = kwargs.get("images")
if images_for_template and self._processor is not None:
content: Any = [{"type": "image"} for _ in images_for_template] + [
{"type": "text", "text": prompt},
]
Comment on lines +149 to +153
else:
content = prompt
messages = [{"role": "user", "content": content}]
template_kwargs: dict[str, Any] = {
"tokenize": False,
"add_generation_prompt": True,
Expand Down
Loading