Skip to content

[Spec] Multi-layer mamba scatter cleanup; fix positional call bug #25030

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
hnyls2002 merged 1 commit into from
May 12, 2026
+22 −24
Merged

[Spec] Multi-layer mamba scatter cleanup; fix positional call bug #25030

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
46 changes: 22 additions & 24 deletions python/sglang/srt/speculative/multi_layer_eagle_worker.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -561,13 +561,10 @@ def verify(self, batch: ScheduleBatch):
logits_output.hidden_states = logits_output.hidden_states[res.accept_indices]

if self.target_worker.model_runner.hybrid_gdn_config is not None:
num_accept_tokens = (
torch.tensor(
res.num_correct_drafts_per_req_cpu,
device=logits_output.hidden_states.device,
dtype=torch.int64,
)
+ 1
num_correct_drafts = torch.tensor(
res.num_correct_drafts_per_req_cpu,
device=logits_output.hidden_states.device,
dtype=torch.int64,
)

# If topk > 1, we need to use retrieve_next_token and retrieve_next_sibling to handle the eagle tree custom attention mask
Expand All @@ -577,28 +574,29 @@ def verify(self, batch: ScheduleBatch):
# first_token_indices_per_req=prepend(0, accept_indices[cumulative_num_accept_tokens[:-1]]) = [0, 5, 10]
# last_token_indices_per_req=accept_indices[cumulative_num_accept_tokens - 1] = [4, 9, 11] (last token ID of each req)
# last_correct_step_indices = [4,4,1]; those are the per-req spec-decoding step offsets that contain the correct mamba caches
cumulative_num_accept_tokens = torch.cumsum(num_accept_tokens, dim=0)
req_start_positions = torch.cat(
[
torch.zeros(
1,
dtype=cumulative_num_accept_tokens.dtype,
device=cumulative_num_accept_tokens.device,
),
cumulative_num_accept_tokens[:-1],
]
# equivalent: last_correct_step_indices = last_token_indices_per_req - first_token_indices_per_req;
# `accepted_indices_offset` equals `first_token_indices_per_req` because the first accepted slot of each req is its "current token" at logical position i * draft_token_num.
cumulative_num_accept_tokens = torch.cumsum(
num_correct_drafts + 1, dim=0
)
accepted_indices_offset = torch.arange(
0,
len(batch.seq_lens) * self.speculative_num_draft_tokens,
step=self.speculative_num_draft_tokens,
dtype=num_correct_drafts.dtype,
device=num_correct_drafts.device,
)
first_token_indices_per_req = res.accept_indices[req_start_positions]
last_token_indices_per_req = res.accept_indices[
cumulative_num_accept_tokens - 1
]
last_correct_step_indices = (
last_token_indices_per_req - first_token_indices_per_req
res.accept_indices[cumulative_num_accept_tokens - 1]
- accepted_indices_offset
)
else:
last_correct_step_indices = num_accept_tokens - 1
last_correct_step_indices = num_correct_drafts
self.target_worker.model_runner.attn_backend.update_mamba_state_after_mtp_verify(
last_correct_step_indices, self.target_worker.model_runner.model
last_correct_step_indices=last_correct_step_indices,
mamba_track_indices=None,
mamba_steps_to_track=None,
model=self.target_worker.model_runner.model,
)

if batch.return_logprob:
Expand Down
Loading