Merge branch 'v0.11.0-dev-ds' into 'v0.11.0-dev'

fix: 修复deepseek量化模型的若干问题

See merge request dcutoolkit/deeplearing/vllm!338

vllm/model_executor/layers/fused_moe/config.py

@@ -287,12 +287,16 @@ class FusedMoEQuantConfig:
 
     @property
     def use_int8_w8a8(self) -> bool:
-        return self.quant_dtype == torch.int8
+        return self.quant_dtype == torch.int8 and self._w1.dtype == torch.int8
 
     @property
     def use_int8_w8a16(self) -> bool:
         return (self._a1.dtype is None and self._w1.dtype == torch.int8)
 
+    @property
+    def use_int4_w4a8(self) -> bool:
+        return (self._a1.dtype == torch.int8 and self._w1.dtype == "int4")
+
     @property
     def use_int4_w4a16(self) -> bool:
         return (self._a1.dtype is None and self._w1.dtype == "int4")

vllm/model_executor/layers/fused_moe/fused_moe.py

@@ -1813,7 +1813,6 @@ def fused_experts(
     quant_config: Optional[FusedMoEQuantConfig] = None,
     allow_deep_gemm: bool = False,
     allow_cutlass_block_scaled_grouped_gemm: bool = False,
-    use_int4_w4a8: bool = False,
     use_nn_moe: Optional[bool] = False,
 ) -> torch.Tensor:
 
@@ -1873,7 +1872,7 @@ def fused_experts(
             use_int8_w8a8=quant_config.use_int8_w8a8,
             use_int8_w8a16=quant_config.use_int8_w8a16,
             use_int4_w4a16=quant_config.use_int4_w4a16,
-            use_int4_w4a8=use_int4_w4a8,
+            use_int4_w4a8=quant_config.use_int4_w4a8,
             use_mxfp4_w4a4=quant_config.use_mxfp4_w4a4,
             per_channel_quant=quant_config.per_act_token_quant,
             global_num_experts=global_num_experts,
@@ -2370,7 +2369,7 @@ class TritonExperts(mk.FusedMoEPermuteExpertsUnpermute):
             use_int8_w8a8=self.quant_config.use_int8_w8a8,
             use_int8_w8a16=self.quant_config.use_int8_w8a16,
             use_int4_w4a16=self.quant_config.use_int4_w4a16,
-            use_int4_w4a8=self.use_int4_w4a8,
+            use_int4_w4a8=self.quant_config.use_int4_w4a8,
             per_channel_quant=self.per_act_token_quant,
             block_shape=self.block_shape,
             B_bias=self.w1_bias,
@@ -2404,7 +2403,7 @@ class TritonExperts(mk.FusedMoEPermuteExpertsUnpermute):
             use_int8_w8a8=self.quant_config.use_int8_w8a8,
             use_int8_w8a16=self.quant_config.use_int8_w8a16,
             use_int4_w4a16=self.quant_config.use_int4_w4a16,
-            use_int4_w4a8=self.use_int4_w4a8,
+            use_int4_w4a8=self.quant_config.use_int4_w4a8,
             per_channel_quant=self.per_act_token_quant,
             block_shape=self.block_shape,
             B_bias=self.w2_bias,

vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_marlin.py

@@ -49,6 +49,7 @@ class SlimQuantCompressedTensorsMarlinConfig(CompressedTensorsConfig):
         sparsity_ignore_list: list[str],
         kv_cache_scheme: Optional[dict[str, Any]] = None,
         config: Optional[dict[str, Any]] = None,
+        transform_config: Optional[dict[str, Any]] = None,
     ):
         super().__init__(
             target_scheme_map,
@@ -57,7 +58,8 @@ class SlimQuantCompressedTensorsMarlinConfig(CompressedTensorsConfig):
             sparsity_scheme_map,
             sparsity_ignore_list,
             kv_cache_scheme,
-            config
+            config,
+            transform_config
         )
     @classmethod
     def override_quantization_method(

vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe_marlin.py

@@ -10,10 +10,11 @@ from vllm.logger import init_logger
 from torch.nn.parameter import Parameter
 from vllm.model_executor.layers.fused_moe import (
     FusedMoE, FusedMoEActivationFormat, FusedMoEMethodBase,
-    FusedMoeWeightScaleSupported)
+    FusedMoeWeightScaleSupported, FusedMoEConfig)
 from vllm.model_executor.utils import set_weight_attrs
 from vllm.model_executor.layers.quantization.utils.w8a8_utils import(
     get_w8a8_int8_marlin_weights)
+from vllm.model_executor.layers.fused_moe.config import (FusedMoEQuantConfig)
 try:
     from lmslim.layers.fused_moe.fuse_moe_int8_marlin import fused_experts_impl_int8_marlin
 except Exception:
@@ -27,6 +28,9 @@ __all__ = [
 
 
 class CompressedTensorsMarlinMoEMethod(FusedMoEMethodBase):
+    def __init_(self, moe: FusedMoEConfig):
+        super().__init__(moe)
+
     @staticmethod
     def get_moe_method(
         quant_config: "SlimQuantCompressedTensorsMarlinConfig",  # type: ignore # noqa E501
@@ -38,17 +42,17 @@ class CompressedTensorsMarlinMoEMethod(FusedMoEMethodBase):
         input_quant = quant_config.target_scheme_map["Linear"].get(
             "input_activations")
         if quant_config._is_dynamic_token_w8a8(weight_quant, input_quant):
-            return CompressedTensorsW8A8Int8MarlinMoEMethod(quant_config)
+            return CompressedTensorsW8A8Int8MarlinMoEMethod(quant_config, layer.moe_config)
         else:
             raise RuntimeError(
                 f"Slimquant_marlin does not support the FusedMoe scheme: {weight_quant}, {input_quant}")
 
 class CompressedTensorsW8A8Int8MarlinMoEMethod(CompressedTensorsMarlinMoEMethod):
-    def __init__(
-            self,
-            quant_config: "CompressedTensorsMarlinConfig"  # type: ignore # noqa E501
-    ):
+    def __init__(self,
+                 quant_config: "CompressedTensorsConfig",
+                 moe: FusedMoEConfig):
         self.quant_config = quant_config
+        super().__init__(moe)
         self.weight_quant = self.quant_config.target_scheme_map["Linear"].get(
             "weights")
         self.input_quant = self.quant_config.target_scheme_map["Linear"].get(
@@ -69,7 +73,10 @@ class CompressedTensorsW8A8Int8MarlinMoEMethod(CompressedTensorsMarlinMoEMethod)
                 "For INT8 Fused MoE layers, we require channelwise, "
                 "dynamic per token quantization. Found static input scales.")
 
-        
+    def get_fused_moe_quant_config(
+            self, layer: torch.nn.Module) -> Optional[FusedMoEQuantConfig]:
+        return None
+
     def create_weights(self, layer: torch.nn.Module, num_experts: int,
                        hidden_size: int, intermediate_size_per_partition: int,
                        params_dtype: torch.dtype, **extra_weight_attrs):
@@ -171,7 +178,7 @@ class CompressedTensorsW8A8Int8MarlinMoEMethod(CompressedTensorsMarlinMoEMethod)
                 "`CompressedTensorsW8A8Int8MoEMethod` yet.")
 
 
-        topk_weights, topk_ids = FusedMoE.select_experts(
+        topk_weights, topk_ids, _ = FusedMoE.select_experts(
             hidden_states=x,
             router_logits=router_logits,
             use_grouped_topk=use_grouped_topk,

vllm/model_executor/layers/quantization/slimquant_w4a8.py

@@ -13,11 +13,13 @@ from vllm.model_executor.parameter import (BasevLLMParameter,
                                            ChannelQuantScaleParameter,
                                            ModelWeightParameter,
                                            PerTensorScaleParameter)
+from vllm.model_executor.layers.fused_moe.config import FusedMoEQuantConfig
+from vllm.model_executor.layers.fused_moe.modular_kernel import (
+    FusedMoEModularKernel)
 from lmslim.layers.gemm.int8_utils import (
     per_token_group_quant_int8,
     per_token_quant_int8)
 from vllm.utils import W8a8GetCacheJSON
-
 import os
 from vllm import _custom_ops as ops
 from vllm import envs
@@ -79,7 +81,7 @@ class SlimQuantW4A8Int8Config(QuantizationConfig):
         if isinstance(layer, LinearBase):
             return SlimQuantW4A8Int8LinearMethod(self)
         elif isinstance(layer, FusedMoE):
-            return SlimQuantW4A8Int8MoEMethod(self)
+            return SlimQuantW4A8Int8MoEMethod(self,layer.moe_config)
         return None
 
     def get_scaled_act_names(self) -> List[str]:
@@ -92,7 +94,7 @@ class SlimQuantW4A8Int8LinearMethod(LinearMethodBase):
         self.quantization_config = quantization_config
         self.tritonsingleton= W8a8GetCacheJSON()
         self.w8a8_strategy=int(os.getenv('W8A8_SUPPORT_METHODS', '1'))
-        
+
     def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
         n=layer.weight.shape[0]
         k=layer.weight.shape[1]
@@ -254,9 +256,29 @@ class SlimQuantW4A8Int8MoEMethod:
             return obj
         return super().__new__(cls)
 
-    def __init__(self, quant_config):
+    def __init__(self, quant_config, moe):
+        self.moe = moe
         self.quant_config = quant_config
         self.tritonsingleton= W8a8GetCacheJSON()
+        self.moe_quant_config: Optional[FusedMoEQuantConfig] = None
+        self.fused_experts: Optional[FusedMoEModularKernel] = None
+        self.topk_indices_dtype = None
+
+    def get_fused_moe_quant_config(
+            self, layer: torch.nn.Module)-> Optional[FusedMoEQuantConfig]:
+        self.moe_quant_config = FusedMoEQuantConfig.make(
+            torch.int8,
+            w1_scale=layer.w13_weight_scale,
+            w2_scale=layer.w2_weight_scale,
+            a1_scale=layer.w13_input_scale,
+            a2_scale=layer.w2_input_scale,
+            per_act_token_quant=True,
+            per_out_ch_quant=False,
+            block_shape=None,
+        )
+        self.moe_quant_config._w1.dtype="int4"
+        self.moe_quant_config._w1.dtype="int4"
+        return self.moe_quant_config
 
     def create_weights(
         self,
@@ -365,7 +387,7 @@ class SlimQuantW4A8Int8MoEMethod:
             raise NotImplementedError(
                 "EPLB not supported for `SlimQuantW4A8Int8MoEMethod` yet.")   
         # Expert selection
-        topk_weights, topk_ids = FusedMoE.select_experts(
+        topk_weights, topk_ids, _ = FusedMoE.select_experts(
             hidden_states=x,
             router_logits=router_logits,
             use_grouped_topk=use_grouped_topk,
@@ -387,15 +409,10 @@ class SlimQuantW4A8Int8MoEMethod:
             topk_weights=topk_weights,
             topk_ids=topk_ids,
             inplace=True,
-            use_int4_w4a8=True,
-            per_channel_quant=True,
             activation=activation,
             expert_map=expert_map,
             apply_router_weight_on_input=apply_router_weight_on_input,
             global_num_experts=global_num_experts,
-            w1_scale=(layer.w13_weight_scale),
-            w2_scale=(layer.w2_weight_scale),
-            a1_scale=layer.w13_input_scale,
-            a2_scale=layer.w2_input_scale,
+            quant_config=self.moe_quant_config,
             use_nn_moe=use_nn_moe,
         )

vllm/model_executor/layers/quantization/slimquant_w4a8_marlin.py

@@ -13,8 +13,9 @@ from vllm.model_executor.layers.quantization.base_config import (QuantizationCon
 from vllm.model_executor.layers.quantization.utils.w4a8_utils import w4a8_weight_repack_impl
 from vllm.model_executor.layers.fused_moe import (FusedMoE, FusedMoEMethodBase,
                                                   FusedMoeWeightScaleSupported)
-from vllm.model_executor.parameter import (ChannelQuantScaleParameter,
-                                           ModelWeightParameter)
+from vllm.model_executor.layers.fused_moe.config import FusedMoEQuantConfig
+from vllm.model_executor.layers.fused_moe.modular_kernel import (
+    FusedMoEModularKernel)
 from vllm.model_executor.layers.quantization.slimquant_w4a8 import SlimQuantW4A8Int8LinearMethod
 
 try:
@@ -110,7 +111,7 @@ class SlimQuantW4A8Int8MarlinConfig(QuantizationConfig):
         if isinstance(layer, LinearBase):
             return SlimQuantW4A8Int8LinearMethod(self)
         elif isinstance(layer, FusedMoE):
-            return SlimQuantW4A8Int8MarlinMoEMethod(self)
+            return SlimQuantW4A8Int8MarlinMoEMethod(self,layer.moe_config)
         return None
 
     def get_scaled_act_names(self) -> List[str]:
@@ -142,8 +143,16 @@ class SlimQuantW4A8Int8MarlinMoEMethod:
             return obj
         return super().__new__(cls)
 
-    def __init__(self, quant_config):
+    def __init__(self, quant_config, moe):
+        self.moe = moe
         self.quant_config = quant_config
+        self.moe_quant_config: Optional[FusedMoEQuantConfig] = None
+        self.fused_experts: Optional[FusedMoEModularKernel] = None
+        self.topk_indices_dtype = None
+
+    def get_fused_moe_quant_config(
+            self, layer: torch.nn.Module) :
+        return None
 
     def create_weights(
         self,
@@ -235,7 +244,7 @@ class SlimQuantW4A8Int8MarlinMoEMethod:
             raise NotImplementedError(
                 "EPLB not supported for `SlimQuantW4A8Int8MarlinMoEMethod` yet.")
         # Expert selection
-        topk_weights, topk_ids = FusedMoE.select_experts(
+        topk_weights, topk_ids, _ = FusedMoE.select_experts(
             hidden_states=x,
             router_logits=router_logits,
             use_grouped_topk=use_grouped_topk,