[QUANT] Add GPTQModel Dynamic Quantization + lm_head Quantization (#3790)

Signed-off-by: ZX-ModelCloud <zx@modelcloud.ai>
Co-authored-by: ZX-ModelCloud <zx@modelcloud.ai>

python/sglang/srt/models/torch_native_llama.py

@@ -64,6 +64,7 @@ from sglang.srt.layers.vocab_parallel_embedding import (
 )
 from sglang.srt.model_executor.forward_batch_info import ForwardBatch
 from sglang.srt.model_loader.weight_utils import default_weight_loader
+from sglang.srt.utils import add_prefix
 
 tp_size = get_tensor_model_parallel_world_size()
 tp_rank = get_tensor_model_parallel_rank()
@@ -294,14 +295,14 @@ class LlamaDecoderLayer(nn.Module):
             rope_is_neox_style=rope_is_neox_style,
             max_position_embeddings=max_position_embeddings,
             quant_config=quant_config,
-            prefix=f"{prefix}.self_attn",
+            prefix=add_prefix("self_attn", prefix),
         )
         self.mlp = LlamaMLP(
             hidden_size=self.hidden_size,
             intermediate_size=config.intermediate_size,
             hidden_act=config.hidden_act,
             quant_config=quant_config,
-            prefix=f"{prefix}.mlp",
+            prefix=add_prefix("mlp", prefix),
         )
         self.input_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
         self.post_attention_layernorm = RMSNorm(