Skip to content

fix deploy openai compat #2101

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

Merged
merged 3 commits into from
Sep 23, 2024
Merged

fix deploy openai compat #2101

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
9690dfe
update
Jintao-Huang Sep 23, 2024
0cd1ad5
Merge branch 'main' into fix_deploy_openai_compat
Jintao-Huang Sep 23, 2024
60f97df
lint pass
Jintao-Huang Sep 23, 2024
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
3 changes: 1 addition & 2 deletions swift/llm/deploy.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -189,8 +189,7 @@ async def _prepare_request(request: Union[ChatCompletionRequest, CompletionReque
f'the model `{model_or_engine.model_type}` is in text generation format. '
'Please use the `completions` API.')
messages = request.messages
if _args.is_multimodal:
compat_openai(messages, request)
compat_openai(messages, request)
# For agent, check if response is endwith observations and join tool observation
messages_join_observation(messages)
example = messages_to_history(messages)
Expand Down
11 changes: 6 additions & 5 deletions swift/llm/utils/template.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3379,19 +3379,20 @@ def data_collator(self, batch: List[Dict[str, Any]], padding_to: Optional[int] =
media_offset = []
cusum_offset = 0


for bi,b in enumerate(batch):
for bi, b in enumerate(batch):
if 'media_offset' in b:
max_sequence_length = res['input_ids'].shape[1]
curr_media_offset = b['media_offset']
if curr_media_offset.shape[1]<max_sequence_length:
padding = curr_media_offset[:,-1:,:].expand(curr_media_offset.shape[0], max_sequence_length-curr_media_offset.shape[1], curr_media_offset.shape[2])
if curr_media_offset.shape[1] < max_sequence_length:
padding = curr_media_offset[:, -1:, :].expand(curr_media_offset.shape[0],
max_sequence_length - curr_media_offset.shape[1],
curr_media_offset.shape[2])
curr_media_offset = torch.concat([curr_media_offset, padding], dim=1)
media_offset.append(curr_media_offset + cusum_offset)
cusum_offset += num_images[bi]

# media_offset = [b['media_offset'] for b in batch if 'media_offset' in b]

if media_offset:
res['media_offset'] = torch.concat(media_offset)
return res
Expand Down
Loading