skip is_v32

vllm/model_executor/models/config.py

@@ -410,8 +410,8 @@ class DeepseekV32ForCausalLM(VerifyAndUpdateConfig):
         hf_config = vllm_config.model_config.hf_config
 
         # Mirror the check in vllm/model_executor/models/deepseek_v2.py
-        is_v32 = hasattr(hf_config, "index_topk")
-        assert is_v32
+        # is_v32 = hasattr(hf_config, "index_topk")
+        # assert is_v32
 
         # For DeepSeekV3.2, we use a custom fp8 format as default (i.e.
         #   "auto")

vllm/model_executor/models/deepseek_mtp.py

@@ -54,7 +54,8 @@ class DeepSeekMultiTokenPredictorLayer(nn.Module):
                                  config.hidden_size,
                                  bias=False)
 
-        self.is_v32 = hasattr(config, "index_topk")
+        # self.is_v32 = hasattr(config, "index_topk")
+        self.is_v32 = False
         if self.is_v32:
             topk_tokens = config.index_topk
             topk_indices_buffer = torch.empty(