Debug

Signed-off-by: wangxiyuan <[email protected]>

vllm/attention/layer.py

@@ -9,12 +9,15 @@
 from vllm.attention.selector import backend_name_to_enum, get_attn_backend
 from vllm.config import CacheConfig, get_current_vllm_config
 from vllm.forward_context import ForwardContext, get_forward_context
+from vllm.logger import init_logger
 from vllm.model_executor.layers.quantization.base_config import (
     QuantizationConfig)
 from vllm.model_executor.layers.quantization.kv_cache import BaseKVCacheMethod
 from vllm.platforms import _Backend, current_platform
 from vllm.utils import direct_register_custom_op
 
+logger = init_logger(__name__)
+
 
 class Attention(nn.Module):
     """Attention layer.
@@ -307,6 +310,9 @@ def unified_attention_with_output_fake(
     return
 
 
+logger.info("====================current platform===========: ", current_platform.dispatch_key)
+
+
 direct_register_custom_op(
     op_name="unified_attention_with_output",
     op_func=unified_attention_with_output,