[ Kernel ] Enable fp8-marlin for fbgemm-fp8 models (#6606)

.buildkite/lm-eval-harness/configs/Meta-Llama-3-70B-Instruct-FBGEMM-nonuniform.yaml

@@ -0,0 +1,11 @@
+# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-hf-baseline.sh -m nm-testing/Meta-Llama-3-70B-Instruct-FBGEMM-nonuniform -b auto -l 1000 -f 5
+model_name: "nm-testing/Meta-Llama-3-70B-Instruct-FBGEMM-nonuniform"
+tasks:
+- name: "gsm8k"
+  metrics:
+  - name: "exact_match,strict-match"
+    value: 0.905
+  - name: "exact_match,flexible-extract"
+    value: 0.905
+limit: 1000
+num_fewshot: 5

.buildkite/lm-eval-harness/configs/models-large.txt

@@ -1,3 +1,4 @@
+Meta-Llama-3-70B-Instruct-FBGEMM-nonuniform.yaml
 Meta-Llama-3-70B-Instruct.yaml
 Mixtral-8x7B-Instruct-v0.1.yaml
 Qwen2-57B-A14-Instruct.yaml

vllm/model_executor/layers/quantization/fbgemm_fp8.py

@@ -9,9 +9,12 @@ from vllm.model_executor.layers.linear import (LinearBase, LinearMethodBase,
                                                UnquantizedLinearMethod)
 from vllm.model_executor.layers.quantization.base_config import (
     QuantizationConfig, QuantizeMethodBase)
+from vllm.model_executor.layers.quantization.utils.marlin_utils_fp8 import (
+    apply_fp8_marlin_linear, prepare_fp8_layer_for_marlin)
 from vllm.model_executor.layers.quantization.utils.w8a8_utils import (
     apply_fp8_linear, create_per_channel_scale_param)
 from vllm.model_executor.utils import set_weight_attrs
+from vllm.platforms import current_platform
 
 logger = init_logger(__name__)
 
@@ -31,6 +34,12 @@ class FBGEMMFp8Config(QuantizationConfig):
         self.ignore_list = ignore_list
         self.input_scale_ub = input_scale_ub
 
+        # For GPUs that lack FP8 hardware support, we can leverage the Marlin
+        # kernel for fast weight-only FP8 quantization
+        capability = current_platform.get_device_capability()
+        capability = capability[0] * 10 + capability[1]
+        self.use_marlin = capability < 89
+
     @classmethod
     def get_name(cls) -> str:
         return "fbgemm_fp8"
@@ -41,7 +50,7 @@ class FBGEMMFp8Config(QuantizationConfig):
 
     @classmethod
     def get_min_capability(cls) -> int:
-        return 89
+        return 80
 
     @classmethod
     def get_config_filenames(cls) -> List[str]:
@@ -143,11 +152,26 @@ class FBGEMMFp8LinearMethod(LinearMethodBase):
         weight = layer.weight
         layer.weight = Parameter(weight.t(), requires_grad=False)
 
+        if self.quant_config.use_marlin:
+            prepare_fp8_layer_for_marlin(layer)
+            # Activations not quantized for marlin.
+            del layer.input_scale_ub
+
     def apply(self,
               layer: torch.nn.Module,
               x: torch.Tensor,
               bias: Optional[torch.Tensor] = None) -> torch.Tensor:
 
+        if self.quant_config.use_marlin:
+            return apply_fp8_marlin_linear(
+                input=x,
+                weight=layer.weight,
+                weight_scale=layer.weight_scale,
+                workspace=layer.workspace,
+                size_n=layer.output_size_per_partition,
+                size_k=layer.input_size_per_partition,
+                bias=bias)
+
         return apply_fp8_linear(input=x,
                                 weight=layer.weight,
                                 weight_scale=layer.weight_scale,

vllm/model_executor/layers/quantization/utils/marlin_utils_fp8.py

@@ -76,8 +76,13 @@ def prepare_fp8_layer_for_marlin(layer: torch.nn.Module) -> None:
     # WEIGHT SCALES
     # Currently Marlin doesn't support per-tensor scales, so we
     # expand it to channelwise
-    scales = layer.weight_scale.repeat(1, part_size_n).to(
+    is_channelwise = layer.weight_scale.shape[0] == part_size_n
-        layer.orig_dtype).to(device)
+    if is_channelwise:
+        scales = layer.weight_scale
+    else:
+        scales = layer.weight_scale.repeat(1, part_size_n)
+    scales = scales.to(layer.orig_dtype).to(device)
+
     # Permute scales
     marlin_scales = marlin_permute_scales(s=scales,
                                           size_k=part_size_k,