iHore/vLLM
8
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[Bugfix] FLASHINFER_MLA_SPARSE_SM120 compatibility with GLM-5 NVFP4 (#46506)

Browse Source
This commit is contained in:
Gabriel Wu
2026-06-25 09:12:00 -07:00
committed by GitHub
parent e45b279928
commit 8fa36fbbeb
1 changed files with 6 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
vllm/v1/attention/backends/mla/flashinfer_mla_sparse_sm120.py
+6 -4
View File
@@ -81,11 +81,13 @@ class FlashInferMLASparseSM120Impl(SparseMLAAttentionImpl[FlashInferMLASparseMet
)
self.kv_scale_format = _kv_scale_format_for_model(model_type)
assert indexer is not None, (
"FLASHINFER_MLA_SPARSE_SM120 requires a sparse-MLA indexer "
"(model with index_topk in its config)."
# Skip-topk layers are built with indexer=None and get the shared
# buffer via mla_args instead (cf. FLASHMLA_SPARSE).
self.topk_indices_buffer: torch.Tensor | None = (
indexer.topk_indices_buffer
if indexer is not None
else mla_args.get("topk_indices_buffer")
)
self.topk_indices_buffer: torch.Tensor | None = indexer.topk_indices_buffer
from vllm.utils.flashinfer import has_flashinfer_sparse_mla_sm120
if not has_flashinfer_sparse_mla_sm120():