[bug] Fix "Current vLLM config is not set." warnings when FlashInfer attention is used (#30241)
Signed-off-by: Po-Han Huang <pohanh@nvidia.com>
This commit is contained in:
@@ -429,6 +429,7 @@ class FlashInferMetadataBuilder(AttentionMetadataBuilder[FlashInferMetadata]):
|
||||
super().__init__(kv_cache_spec, layer_names, vllm_config, device)
|
||||
self.cache_config = vllm_config.cache_config
|
||||
self.model_config = vllm_config.model_config
|
||||
self.attention_config = vllm_config.attention_config
|
||||
self._workspace_buffer = None
|
||||
self._prefill_wrapper: (
|
||||
BatchPrefillWithPagedKVCacheWrapper | BatchDCPPrefillWrapper | None
|
||||
@@ -779,6 +780,7 @@ class FlashInferMetadataBuilder(AttentionMetadataBuilder[FlashInferMetadata]):
|
||||
self.cache_dtype,
|
||||
self.q_data_type,
|
||||
is_prefill=True,
|
||||
force_use_trtllm=self.attention_config.use_trtllm_attention,
|
||||
has_sinks=self.has_sinks,
|
||||
has_spec=uses_spec_reorder,
|
||||
)
|
||||
|
||||
Reference in New Issue
Block a user