fix:

Author: wangzaijun

lightllm/models/llama/layer_infer/transformer_layer_infer.py

@@ -110,15 +110,15 @@ def _bind_attention(self):
             self._context_attention_kernel = partial(
                 LlamaTransformerLayerInfer._context_attention_kernel_ppl_int8kv, self
             )
-        elif "ppl_int8kv_flashdecoding" in self.mode:
+        elif "ppl_int8kv_flashdecoding_diverse" in self.mode:
             self._token_attention_kernel = partial(
                 LlamaTransformerLayerInfer._token_decode_attention_ppl_int8kv_flashdecoding_diverse, self
             )
             self._copy_kv_to_mem_cache = partial(LlamaTransformerLayerInfer._copy_kv_to_mem_cache_ppl_int8kv, self)
             self._context_attention_kernel = partial(
                 LlamaTransformerLayerInfer._context_attention_kernel_ppl_int8kv, self
             )
-        elif "ppl_int8kv_flashdecoding_diverse" in self.mode:
+        elif "ppl_int8kv_flashdecoding" in self.mode:
             self._token_attention_kernel = partial(
                 LlamaTransformerLayerInfer._token_decode_attention_ppl_int8kv_flashdecoding, self
             )