Skip to content
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

Jump to bottom

Simplify chunked prefill #1667

Merged
merged 2 commits into from
Oct 14, 2024
Merged

Simplify chunked prefill #1667

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
81f58e0
update
merrymercy Oct 14, 2024
605e78a
update
merrymercy Oct 14, 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
35 changes: 17 additions & 18 deletions python/sglang/srt/managers/scheduler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -456,7 +456,12 @@ def get_next_batch_to_run(self):
and not self.last_batch.is_empty()
):
if self.current_inflight_req:
self.last_batch.filter_batch(self.current_inflight_req)
self.last_batch.filter_batch(
current_inflight_req=self.current_inflight_req
)
self.tree_cache.cache_unfinished_req(self.current_inflight_req)
# Inflight request keeps its rid but will get a new req_pool_idx.
self.req_to_token_pool.free(self.current_inflight_req.req_pool_idx)
self.batch_is_full = False
if not self.last_batch.is_empty():
if self.running_batch is None:
Expand Down Expand Up @@ -728,27 +733,24 @@ def process_batch_result_prefill(self, batch: ScheduleBatch, result):
# Check finish conditions
logprob_pt = 0
for i, req in enumerate(batch.reqs):
if req is not self.current_inflight_req:
if req.is_inflight_req > 0:
req.is_inflight_req -= 1
else:
# Inflight reqs' prefill is not finished
req.completion_tokens_wo_jump_forward += 1
req.output_ids.append(next_token_ids[i])
req.check_finished()

if req.finished():
self.tree_cache.cache_finished_req(req)
elif not batch.decoding_reqs or req not in batch.decoding_reqs:
self.tree_cache.cache_unfinished_req(req)

if req.regex_fsm is not None:
req.regex_fsm_state = req.regex_fsm.get_next_state(
req.regex_fsm_state, next_token_ids[i]
)

if req.finished():
self.tree_cache.cache_finished_req(req)
elif not batch.decoding_reqs or req not in batch.decoding_reqs:
self.tree_cache.cache_unfinished_req(req)

if req.is_inflight_req > 0:
# Inflight request would get a new req idx
req.is_inflight_req -= 1
self.req_to_token_pool.free(req.req_pool_idx)

if req.return_logprob:
logprob_pt += self.add_logprob_return_values(
i, req, logprob_pt, next_token_ids, logits_output
Expand All @@ -760,7 +762,9 @@ def process_batch_result_prefill(self, batch: ScheduleBatch, result):
# Check finish conditions
for i, req in enumerate(batch.reqs):
req.embedding = embeddings[i]
if req is not self.current_inflight_req:
if req.is_inflight_req > 0:
req.is_inflight_req -= 1
else:
# Inflight reqs' prefill is not finished
# dummy output token for embedding models
req.output_ids.append(0)
Expand All @@ -771,11 +775,6 @@ def process_batch_result_prefill(self, batch: ScheduleBatch, result):
else:
self.tree_cache.cache_unfinished_req(req)

if req.is_inflight_req > 0:
# Inflight request would get a new req idx
req.is_inflight_req -= 1
self.req_to_token_pool.free(req.req_pool_idx)

self.stream_output(batch)

def process_batch_result_decode(self, batch: ScheduleBatch, result):
Expand Down
Loading