Skip to content

[bugfix] fix accuracy problem when chunked prefill #438

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 4 commits into from
Nov 29, 2025
Merged

[bugfix] fix accuracy problem when chunked prefill #438

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
2d140fd
fix accuracy problem when chunked prefill
qyh111 Nov 29, 2025
8c1c387
remove unnessary code for preempted request
qyh111 Nov 29, 2025
30a4246
modify dump
qyh111 Nov 29, 2025
5d90ace
fix comment
qyh111 Nov 29, 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
32 changes: 17 additions & 15 deletions ucm/integration/vllm/ucm_connector.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,6 +39,9 @@ class RequestMeta:
hbm_hit_block_num: int = 0
# local_computed_block + external_computed_block
total_hit_block_num: int = 0
num_token_ids: int = 0
vllm_block_ids: list[int] = field(default_factory=list)
token_processed: int = 0


@dataclass
Expand Down Expand Up @@ -207,7 +210,6 @@ def get_num_new_matched_tokens(
request: "Request",
num_computed_tokens: int,
) -> tuple[int, bool]:

assert num_computed_tokens % self.block_size == 0
hbm_hit_block_num = num_computed_tokens // self.block_size

Expand Down Expand Up @@ -242,13 +244,16 @@ def get_num_new_matched_tokens(
# When all the tokens are cached in ssd or hbm,
# we need to recompute the last token. This if condition will be removed
# once vLLM scheduler provides a better solution in the future.
if total_hit_block_num * self.block_size == request.num_tokens:
num_total_hit_tokens = total_hit_block_num * self.block_size
if num_total_hit_tokens == request.num_tokens:
external_hit_tokens -= 1

self.requests_meta[request.request_id] = RequestMeta(
ucm_block_ids=ucm_block_ids,
hbm_hit_block_num=hbm_hit_block_num,
total_hit_block_num=total_hit_block_num,
num_token_ids=len(request.all_token_ids),
token_processed=num_total_hit_tokens,
)

return external_hit_tokens, False
Expand Down Expand Up @@ -277,22 +282,16 @@ def _generate_dispatch_meta(
| scheduled_block_num |
"""

new_blocks_num = new_tokens // self.block_size
hbm_hit_block_num = req_meta.hbm_hit_block_num
total_hit_block_num = req_meta.total_hit_block_num
scheduled_block_num = total_hit_block_num + new_blocks_num
ucm_block_ids = req_meta.ucm_block_ids
req_meta.vllm_block_ids.extend(vllm_block_ids)

dump_ucm_block_ids = ucm_block_ids[total_hit_block_num:scheduled_block_num]
if need_load:
dump_vllm_block_ids = vllm_block_ids[
total_hit_block_num:scheduled_block_num
]
else:
dump_vllm_block_ids = vllm_block_ids

# after this round, req_meta will be updated
req_meta.total_hit_block_num = scheduled_block_num
start_idx = req_meta.token_processed // self.block_size
end_idx = (req_meta.token_processed + new_tokens) // self.block_size
dump_ucm_block_ids = ucm_block_ids[start_idx:end_idx]
dump_vllm_block_ids = req_meta.vllm_block_ids[start_idx:end_idx]
req_meta.token_processed += new_tokens

load_ucm_block_ids, load_vllm_block_ids = [], []
if need_load:
Expand Down Expand Up @@ -332,10 +331,13 @@ def build_connector_meta(
continue
req_meta = self.requests_meta.get(request_id)
if req_meta:
new_block_ids = []
if scheduled_cached_reqs.new_block_ids[i] != None:
new_block_ids = scheduled_cached_reqs.new_block_ids[i][0]
requests_dispatch_meta[request_id] = self._generate_dispatch_meta(
req_meta,
scheduler_output.num_scheduled_tokens[request_id],
scheduled_cached_reqs.new_block_ids[i][0],
new_block_ids,
scheduled_cached_reqs.resumed_from_preemption[i],
)
else:
Expand Down