Skip to content

[PD] Respect sampling_params.max_new_tokens when PD disaggregation is activated #7598

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
Jun 28, 2025
Merged

[PD] Respect sampling_params.max_new_tokens when PD disaggregation is activated #7598

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c6d4fbb
[PD] Respect sampling_params.max_new_tokens when PD disaggregation is…
ShangmingCai Jun 27, 2025
090b6d0
Remove import
ShangmingCai Jun 28, 2025
17644e2
Merge branch 'main' into fix_pd_max_new_tokens
ShangmingCai Jun 28, 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
14 changes: 13 additions & 1 deletion python/sglang/srt/disaggregation/decode.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -604,9 +604,21 @@ def pop_transferred(self) -> List[Req]:
: decode_req.req.top_logprobs_num
].tolist()
)

if hasattr(decode_req.kv_receiver, "clear"):
decode_req.kv_receiver.clear()
transferred_reqs.append(decode_req.req)

# special handling for sampling_params.max_new_tokens == 1
if decode_req.req.sampling_params.max_new_tokens == 1:
# finish immediately
decode_req.req.check_finished()
self.scheduler.stream_output(
[decode_req.req], decode_req.req.return_logprob
)
self.tree_cache.cache_finished_req(decode_req.req)
Comment on lines +612 to +618
Copy link
Contributor

@gemini-code-assist gemini-code-assist bot Jun 27, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

medium

Consider extracting the code block that starts at line 618 and ends at line 624 into a separate function to improve code modularity and readability. This new function could be named something like handle_max_new_tokens_equals_one.

def handle_max_new_tokens_equals_one(decode_req):
    decode_req.req.check_finished()
    self.scheduler.stream_output(
        [decode_req.req], decode_req.req.return_logprob
    )
    self.tree_cache.cache_finished_req(decode_req.req)

if decode_req.req.sampling_params.max_new_tokens == 1:
    # finish immediately
    handle_max_new_tokens_equals_one(decode_req)

else:
transferred_reqs.append(decode_req.req)

indices_to_remove.add(i)
elif poll in [
KVPoll.Bootstrapping,
Expand Down
Loading