rebase

Signed-off-by: shen-shanshan <[email protected]>

Author: shen-shanshan

vllm/v1/worker/gpu_model_runner.py

@@ -53,7 +53,7 @@
 from vllm.sequence import IntermediateTensors, PoolerOutput
 from vllm.tasks import GenerationTask, PoolingTask, SupportedTask
 from vllm.utils import (STR_DTYPE_TO_TORCH_DTYPE, DeviceMemoryProfiler,
-                        GiB_bytes, LazyLoader, check_use_alibi, get_dtype_size,
+                        GiB_bytes, check_use_alibi, get_dtype_size,
                         is_pin_memory_available, round_up, supports_dynamo)
 from vllm.v1.attention.backends.gdn_attn import GDNAttentionMetadataBuilder
 from vllm.v1.attention.backends.utils import (