update weights_not_loaded and flash_mla_with_kvcache

vllm/model_executor/model_loader/default_loader.py

@@ -272,6 +272,7 @@ class DefaultModelLoader(BaseModelLoader):
         # that have loaded weights tracking currently.
         if model_config.quantization is None and loaded_weights is not None:
             weights_not_loaded = weights_to_load - loaded_weights
+            weights_not_loaded = {k for k in weights_not_loaded if not k.endwith("indexer.weights_proj.bias")}
             if weights_not_loaded:
                 raise ValueError("Following weights were not initialized from "
                                  f"checkpoint: {weights_not_loaded}")

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

@@ -210,6 +210,8 @@ class FlashMLAImpl(MLACommonImpl[FlashMLAMetadata]):
             causal=True,
             descale_q=layer._q_scale.reshape(1),
             descale_k=layer._k_scale.reshape(1),
+            is_fp8_kvcache=False,
+            indices= None,
         )
 
         return o, lse