Skip to content

[Fix] Fixes speculative decode indexing and unpad condition for attention metadata #5626

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
yiz-liu merged 2 commits into from
Jan 8, 2026
Merged

[Fix] Fixes speculative decode indexing and unpad condition for attention metadata #5626

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
6f31ae6
Fixes speculative decode indexing and unpad condition
yiz-liu Jan 6, 2026
1c99bc4
Preserves full token mappings to avoid shape mismatch
yiz-liu Jan 8, 2026
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
7 changes: 5 additions & 2 deletions vllm_ascend/attention/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -117,8 +117,11 @@ def unpadded(self, num_actual_tokens: int,
num_actual_tokens=num_actual_tokens,
max_query_len=self.max_query_len,
decode_token_per_req=self.decode_token_per_req,
block_table_tensor=self.block_table_tensor[:num_actual_reqs],
slot_mapping=self.slot_mapping[:num_actual_tokens],
# NOTE: keep all tokens for block_table_tensor and slot_mapping otherwise
# there will be error about shape mismatch during reshape and cache.
# This is really strange since vLLM slices them as well
block_table_tensor=self.block_table_tensor,
slot_mapping=self.slot_mapping,
causal=self.causal,
actual_seq_lengths_q=self.actual_seq_lengths_q[:num_actual_tokens],
positions=self.positions[:num_actual_tokens],
Expand Down
13 changes: 3 additions & 10 deletions vllm_ascend/spec_decode/eagle_proposer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -692,13 +692,6 @@ def prepare_inputs_padded(
num_reqs = common_attn_metadata.num_reqs
device = valid_sampled_tokens_count.device

if num_reqs != spec_decode_metadata.cu_num_draft_tokens.shape[0]:
# TODO: This is a serious issue and should be taken care of ASAP
# In short, why input_batch.num_reqs != attn_metadata.num_reqs?
# Previously in #4963, we modified `query_start_loc`, but this
# problem remains unsolved.
num_reqs = spec_decode_metadata.cu_num_draft_tokens.shape[0]

token_indices_to_sample = torch.empty((num_reqs, ),
dtype=torch.int32,
device=device)
Expand Down Expand Up @@ -730,9 +723,9 @@ def prepare_inputs_padded(
torch.zeros_like(num_draft_tokens_gpu),
)

query_start_loc = common_attn_metadata.query_start_loc[
1:1 + num_rejected_tokens_gpu.shape[0]]
token_indices_to_sample = query_start_loc - 1 - num_rejected_tokens_gpu
token_indices_to_sample = (
common_attn_metadata.query_start_loc[1:] - 1 -
num_rejected_tokens_gpu)

query_start_loc_cpu = common_attn_metadata.query_start_loc_cpu

Expand Down
4 changes: 1 addition & 3 deletions vllm_ascend/worker/model_runner_v1.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1013,9 +1013,7 @@ def _prepare_inputs(
if self.speculative_config and \
self.spec_decode_common_attn_metadata is None:
self.spec_decode_common_attn_metadata = common_attn_metadata
if self.speculative_config.method in ("eagle", "eagle3") and \
(self.vllm_config.speculative_config.enforce_eager \
or self.use_async_scheduling):
if num_reqs != base_num_reqs or total_num_scheduled_tokens != num_input_tokens:
self.spec_decode_common_attn_metadata = \
self.spec_decode_common_attn_metadata.unpadded(
total_num_scheduled_tokens, base_num_reqs)
Expand Down
Loading