Skip to content

Support InternVL3.5-Flash #3952

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 8 commits into from
Sep 17, 2025
+383 −29
Merged

Support InternVL3.5-Flash #3952

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
075cd45
support internvl flash
CUHKSZzxy Sep 9, 2025
18e4abb
clean
CUHKSZzxy Sep 9, 2025
02cfea4
fix
CUHKSZzxy Sep 10, 2025
ab0adfc
fix context update for multi requests
CUHKSZzxy Sep 11, 2025
9740225
dropout to identity, remove clone, fix type
CUHKSZzxy Sep 11, 2025
de968e7
fix acc, explicit dtype, optimize
CUHKSZzxy Sep 16, 2025
c610ff3
get seqlen from context, pass context in post update
CUHKSZzxy Sep 17, 2025
4447ae0
remove self.model_metas
CUHKSZzxy Sep 17, 2025
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
9 changes: 7 additions & 2 deletions lmdeploy/pytorch/engine/model_agent.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -238,7 +238,9 @@ def model_forward(
context=context,
)
output = model(**input_dict)
return dict(hidden_states=output, model_metas=model_metas)
seq_length = ctx_mgr.current_context().q_seqlens

return dict(hidden_states=output, model_metas=model_metas, seq_length=seq_length)


@record_function('stopping_criteria')
Expand Down Expand Up @@ -502,7 +504,10 @@ async def __long_context_single_forward(new_inputs, max_seqlen: int):
if not is_long_context:
ret = await __forward(inputs)
if not return_logits and not inputs.is_decoding:
last_token_loc = inputs.seq_length.cumsum(0) - 1
# fetch seq_length from the returned context, since models may change it (e.g. InternVL-Flash)
seq_length = ret.get('seq_length', None)
last_token_loc = seq_length.cumsum(0) - 1

ret['hidden_states'] = ret['hidden_states'][:, last_token_loc]
else:
ret = await __long_context_single_forward(inputs, max_seqlen)
Expand Down
Loading