[Bugfix] Update starcoder2 to remap k/v scale names for kv_cache quantization (#11148)

vllm/model_executor/models/starcoder2.py

@@ -37,7 +37,8 @@ from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.sampler import SamplerOutput, get_sampler
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     DEFAULT_VOCAB_PADDING_SIZE, ParallelLMHead, VocabParallelEmbedding)
-from vllm.model_executor.model_loader.weight_utils import default_weight_loader
+from vllm.model_executor.model_loader.weight_utils import (
+    default_weight_loader, maybe_remap_kv_scale_name)
 from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 
@@ -345,6 +346,10 @@ class Starcoder2ForCausalLM(nn.Module, SupportsPP):
                 weight_loader(param, loaded_weight, shard_id)
                 break
             else:
+                name = maybe_remap_kv_scale_name(name, params_dict)
+                if name is None:
+                    continue
+
                 if self.config.tie_word_embeddings and "lm_head.weight" in name:
                     continue
                 if is_pp_missing_parameter(name, self):