feat: fused_moe fp8 monkey patch (#2174)

python/sglang/srt/layers/quantization/__init__.py

@@ -1,18 +1,19 @@
 # Adapted from https://raw.githubusercontent.com/vllm-project/vllm/v0.5.5/vllm/model_executor/layers/quantization/__init__.py
 
-from typing import Dict, Type
+from typing import Callable, Dict, Optional, Type
 
+import torch
 from vllm.model_executor.layers.quantization.aqlm import AQLMConfig
 from vllm.model_executor.layers.quantization.awq import AWQConfig
 from vllm.model_executor.layers.quantization.awq_marlin import AWQMarlinConfig
 from vllm.model_executor.layers.quantization.bitsandbytes import BitsAndBytesConfig
-from vllm.model_executor.layers.quantization.compressed_tensors.compressed_tensors import (  # noqa: E501
+from vllm.model_executor.layers.quantization.compressed_tensors.compressed_tensors import (
     CompressedTensorsConfig,
 )
 from vllm.model_executor.layers.quantization.deepspeedfp import DeepSpeedFPConfig
 from vllm.model_executor.layers.quantization.experts_int8 import ExpertsInt8Config
 from vllm.model_executor.layers.quantization.fbgemm_fp8 import FBGEMMFp8Config
-from vllm.model_executor.layers.quantization.fp8 import Fp8Config
+from vllm.model_executor.layers.quantization.fp8 import Fp8Config, Fp8MoEMethod
 from vllm.model_executor.layers.quantization.gguf import GGUFConfig
 from vllm.model_executor.layers.quantization.gptq import GPTQConfig
 from vllm.model_executor.layers.quantization.gptq_marlin import GPTQMarlinConfig
@@ -30,8 +31,6 @@
     "tpu_int8": Int8TpuConfig,
     "fp8": Fp8Config,
     "fbgemm_fp8": FBGEMMFp8Config,
-    # The order of gptq methods is important for config.py iteration over
-    # override_quantization_method(..)
     "marlin": MarlinConfig,
     "gguf": GGUFConfig,
     "gptq_marlin_24": GPTQMarlin24Config,
@@ -47,38 +46,89 @@
 
 def get_quantization_config(quantization: str) -> Type[QuantizationConfig]:
     if quantization not in QUANTIZATION_METHODS:
-        raise ValueError(f"Invalid quantization method: {quantization}")
+        raise ValueError(
+            f"Invalid quantization method: {quantization}. "
+            f"Available methods: {list(QUANTIZATION_METHODS.keys())}"
+        )
     return QUANTIZATION_METHODS[quantization]
 
 
-__all__ = [
-    "QuantizationConfig",
-    "get_quantization_config",
-    "QUANTIZATION_METHODS",
-]
+def fp8_moe_apply(
+    self,
+    layer: torch.nn.Module,
+    x: torch.Tensor,
+    router_logits: torch.Tensor,
+    top_k: int,
+    renormalize: bool,
+    use_grouped_topk: bool,
+    topk_group: Optional[int] = None,
+    num_expert_group: Optional[int] = None,
+    custom_routing_function: Optional[Callable] = None,
+) -> torch.Tensor:
+    """Enhanced apply method for FP8 MoE."""
+    from sglang.srt.layers.fused_moe_triton import FusedMoE
+    from sglang.srt.layers.fused_moe_triton.fused_moe import fused_experts
+
+    # Expert selection
+    topk_weights, topk_ids = FusedMoE.select_experts(
+        hidden_states=x,
+        router_logits=router_logits,
+        use_grouped_topk=use_grouped_topk,
+        top_k=top_k,
+        renormalize=renormalize,
+        topk_group=topk_group,
+        num_expert_group=num_expert_group,
+        custom_routing_function=custom_routing_function,
+    )
+
+    # Expert fusion with FP8 quantization
+    return fused_experts(
+        x,
+        layer.w13_weight,
+        layer.w2_weight,
+        topk_weights=topk_weights,
+        topk_ids=topk_ids,
+        inplace=True,
+        use_fp8_w8a8=True,
+        w1_scale=layer.w13_weight_scale,
+        w2_scale=layer.w2_weight_scale,
+        a1_scale=layer.w13_input_scale,
+        a2_scale=layer.w2_input_scale,
+    )
 
 
 def fp8_get_quant_method(self, layer, prefix):
+    """Enhanced get_quant_method for FP8 config."""
     from vllm.model_executor.layers.linear import LinearBase
-    from vllm.model_executor.layers.quantization.fp8 import (
-        Fp8LinearMethod,
-        Fp8MoEMethod,
-    )
+    from vllm.model_executor.layers.quantization.fp8 import Fp8LinearMethod
     from vllm.model_executor.layers.quantization.utils.quant_utils import (
         is_layer_skipped,
     )
 
     from sglang.srt.layers.fused_moe_triton.layer import FusedMoE
+    from sglang.srt.layers.linear import UnquantizedLinearMethod
 
     if isinstance(layer, LinearBase):
         if is_layer_skipped(prefix, self.ignored_layers):
-            from sglang.srt.layers.linear import UnquantizedLinearMethod
-
             return UnquantizedLinearMethod()
         return Fp8LinearMethod(self)
     elif isinstance(layer, FusedMoE):
         return Fp8MoEMethod(self)
     return None
 
 
-setattr(Fp8Config, "get_quant_method", fp8_get_quant_method)
+def apply_monkey_patches():
+    """Apply all monkey patches in one place."""
+    setattr(Fp8MoEMethod, "apply", fp8_moe_apply)
+    setattr(Fp8Config, "get_quant_method", fp8_get_quant_method)
+
+
+# Apply patches when module is imported
+apply_monkey_patches()
+
+
+__all__ = [
+    "QuantizationConfig",
+    "get_quantization_config",
+    "QUANTIZATION_METHODS",
+]