diff --git a/vllm/v1/worker/gpu/attn_utils.py b/vllm/v1/worker/gpu/attn_utils.py index 227c51bb47..4f9e5e99a9 100644 --- a/vllm/v1/worker/gpu/attn_utils.py +++ b/vllm/v1/worker/gpu/attn_utils.py @@ -218,7 +218,7 @@ def _reshape_attention_kv_cache( kv_cache = ( kv_raw_tensor.view(-1, block_stride)[:, offset : offset + page_bytes] .view(dtype) - .view(kv_cache_shape) + .view(permuted_kv_cache_shape) ) elif kv_cache_spec.page_size_padded is not None: # Use a strided view to skip the padding between physical pages.