Missed one

vllm-project · Sep 12, 2024 · e457298 · e457298
1 parent 7074367
commit e457298
Showing 1 changed file with 2 additions and 1 deletion.
diff --git a/vllm/model_executor/models/utils.py b/vllm/model_executor/models/utils.py
@@ -10,6 +10,7 @@
 from vllm.attention.backends.blocksparse_attn import (
     BlocksparseFlashAttentionMetadata)
 from vllm.attention.backends.flash_attn import FlashAttentionMetadata
+from vllm.attention.backends.flashinfer import FlashInferMetadata
 from vllm.attention.backends.rocm_flash_attn import ROCmFlashAttentionMetadata
 from vllm.attention.backends.xformers import XFormersMetadata
 from vllm.config import (CacheConfig, LoRAConfig, MultiModalConfig,
@@ -152,7 +153,7 @@ def merge_partial_multimodal_embeddings(
     if not isinstance(
             prefill_metadata,
         (FlashAttentionMetadata, XFormersMetadata, ROCmFlashAttentionMetadata,
-         BlocksparseFlashAttentionMetadata)):
+         BlocksparseFlashAttentionMetadata, FlashInferMetadata)):
         raise ValueError(
             f"Unsupported Attention backend ({type(prefill_metadata)}) for "
             "partial embedding replacement.")