【New Feature】W4afp8 supports per group quantization (#4272)

* w4afp8 支持per group

* code style

* 精度完成

* revert append attn utils

* ffn1 动态量化

* ffn2 支持动态量化

* code style

* code style

* 修改单测

* 修改单测

* fix bug

* Implement conditional parameter creation for layers

Add parameter creation for up_gate_proj_in_scale when ep_size > 1.

* code style

* fix conflict

* code style

* code style

* 修复w4aint8 精度

* fix ci

---------

Co-authored-by: yuanxiaolan <yuanxiaolan01@baidu.com>

fastdeploy/model_executor/layers/quantization/mix_quant.py

@@ -39,6 +39,7 @@ class MixQuantConfig(QuantConfigBase):
         is_permuted: bool = True,
         is_quantized: bool = False,
         hadamard_block_size: int = 128,
+        moe_dynamic_quant: bool = False,
     ) -> None:
         super().__init__()
         self.dense_quant_type = dense_quant_type
@@ -57,6 +58,7 @@ class MixQuantConfig(QuantConfigBase):
         self.is_checkpoint_bf16 = not is_quantized
         self.is_quantized = is_quantized
         self.hadamard_block_size = hadamard_block_size
+        self.moe_dynamic_quant = moe_dynamic_quant
 
     def name(self) -> str:
         return "mix_quant"
@@ -73,6 +75,7 @@ class MixQuantConfig(QuantConfigBase):
             config.get("is_permuted", True),
             config.get("is_quantized", False),
             config.get("hadamard_block_size", 128),
+            config.get("moe_dynamic_quant", False),
         )
 
     def get_quant_method(self, layer) -> Optional[QuantMethodBase]: