[None][fix] Fix #8383 introduced TRTLLM backend python error (#9804)

jhaotingc · web-flow · commit da074be037a2 · 2025-12-08T13:31:37.000-08:00
Signed-off-by: Jhao-Ting Chen &lt;jhaotingc@nvidia.com&gt;
diff --git a/tensorrt_llm/_torch/attention_backend/trtllm.py b/tensorrt_llm/_torch/attention_backend/trtllm.py
@@ -604,6 +604,9 @@ def is_nvfp4_output_kernel_available(
             is_mla_enable,
         )
 
+    def is_sm_version_trtllm_gen_kernel(self, sm):
+        return not (sm < 100 or sm in [120, 121])
+
 
 @dataclass(kw_only=True)
 class TrtllmAttentionMetadata(AttentionMetadata):

Original file line number	Diff line number	Diff line change
`@@ -604,6 +604,9 @@ def is_nvfp4_output_kernel_available(`
`604`	`604`	`is_mla_enable,`
`605`	`605`	`)`
`606`	`606`
	`607`	`+ def is_sm_version_trtllm_gen_kernel(self, sm):`
	`608`	`+ return not (sm < 100 or sm in [120, 121])`
	`609`	`+`
`607`	`610`
`608`	`611`	`@dataclass(kw_only=True)`
`609`	`612`	`class TrtllmAttentionMetadata(AttentionMetadata):`