xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-30 00:17:20 +08:00
Code Issues Packages Projects Releases Wiki Activity

[BUG] Qwen3-next MTP. Fix attn metadata build bug (#26564)

Browse Source 
Signed-off-by: Vadim Gimpelson <vadim.gimpelson@gmail.com>
This commit is contained in:
Vadim Gimpelson 2025-10-10 22:59:03 +04:00 committed by GitHub
parent 7c12763b24
commit e94cfd51da
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 6 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
vllm/v1/spec_decode/eagle.py
View File

@ -41,7 +41,6 @@ from vllm.v1.sample.metadata import SamplingMetadata
from vllm.v1.spec_decode.metadata import SpecDecodeMetadata from vllm.v1.spec_decode.metadata import SpecDecodeMetadata
from vllm.v1.utils import CpuGpuBuffer from vllm.v1.utils import CpuGpuBuffer
from vllm.v1.worker.gpu_input_batch import CachedRequestState, InputBatch from vllm.v1.worker.gpu_input_batch import CachedRequestState, InputBatch
from vllm.v1.worker.ubatching import dbo_current_ubatch_id
logger = init_logger(__name__) logger = init_logger(__name__)
@ -234,11 +233,11 @@ class EagleProposer:
assert self.runner is not None assert self.runner is not None
# FIXME: need to consider multiple kv_cache_groups if self.attn_metadata_builder is None:
ubatch_id = dbo_current_ubatch_id() attn_metadata_builder = self._get_attention_metadata_builder()
attn_metadata_builder = self.runner.attn_groups[0][0].metadata_builders[ else:
ubatch_id attn_metadata_builder = self.attn_metadata_builder
]
attn_metadata = attn_metadata_builder.build_for_drafting( attn_metadata = attn_metadata_builder.build_for_drafting(
common_attn_metadata=common_attn_metadata, draft_index=0 common_attn_metadata=common_attn_metadata, draft_index=0
) )
@ -1076,7 +1075,7 @@ class EagleProposer:
inputs_embeds=inputs_embeds, inputs_embeds=inputs_embeds,
) )
def _get_attention_metadata_builder(self) -> list[AttentionMetadataBuilder]: def _get_attention_metadata_builder(self) -> AttentionMetadataBuilder:
"""Find and return the attention metadata builders for EAGLE layers. """Find and return the attention metadata builders for EAGLE layers.
Returns: Returns: