[Perf]支持DSA架构模型开启MTP后运行full graph模式

8c96d505 · 王敏 · 6efaf21a · 8c96d505
Commit 8c96d505 authored Apr 17, 2026 by 王敏
Show whitespace changes
Inline Side-by-side

Showing with 1 addition and 1 deletion

vllm/v1/attention/backends/mla/indexer.py vllm/v1/attention/backends/mla/indexer.py +1 -1

No files found.
--- a/vllm/v1/attention/backends/mla/indexer.py
+++ b/vllm/v1/attention/backends/mla/indexer.py
@@ -201,7 +201,7 @@ def get_max_prefill_buffer_size(vllm_config: VllmConfig):

 class DeepseekV32IndexerMetadataBuilder(AttentionMetadataBuilder):
    _cudagraph_support: ClassVar[AttentionCGSupport] = (
-        AttentionCGSupport.UNIFORM_SINGLE_TOKEN_DECODE
+        AttentionCGSupport.UNIFORM_BATCH
    )

    reorder_batch_threshold: int = 1