Merge branch 'v0.9.2-dev-wm-1110' into 'v0.9.2-dev'

[fix]修复mtp中的笔误

See merge request dcutoolkit/deeplearing/vllm!249

vllm/v1/attention/backends/mla/common.py

@@ -402,7 +402,7 @@ class MLACommonMetadataBuilder(AttentionMetadataBuilder[M]):
         self.decode_token_num_threshold = 1  
         vllm_config = get_current_vllm_config()
         speculative_config = vllm_config.speculative_config
-        if speculative_config and speculative_config.num_speculative_tokens > 1:
+        if speculative_config and speculative_config.num_speculative_tokens > 0:
             self.use_spec_decode = True
             self.decode_token_num_threshold = 1 + speculative_config.num_speculative_tokens