[Quantization][FP8] Add support for FP8 models with input_scale for output projection and QK quantization (vllm-project#15734)

Signed-off-by: Randall Smith <Randall.Smith@amd.com>
Signed-off-by: Luka Govedič <lgovedic@redhat.com>
Co-authored-by: Luka Govedič <lgovedic@redhat.com>
Signed-off-by: Agata Dobrzyniewicz <adobrzyniewicz@habana.ai>

Author: 2 people

vllm/attention/backends/abstract.py

@@ -237,6 +237,7 @@ class AttentionLayer(Protocol):
     _v_scale: torch.Tensor
     _k_scale_float: float
     _v_scale_float: float
+    _prob_scale: torch.Tensor
 
     def forward(
         self,

vllm/attention/backends/rocm_flash_attn.py

@@ -766,6 +766,12 @@ def forward(
                             query.dtype,
                             seq_lens,
                             make_attn_mask=causal_mask)  # type: ignore
+                    use_fp8_scales = (layer._q_scale and layer._k_scale
+                                      and layer._v_scale and layer._prob_scale
+                                      and self.kv_cache_dtype == "fp8")
+                    full_scales = (
+                        layer._q_scale, layer._k_scale, layer._v_scale,
+                        layer._prob_scale) if use_fp8_scales else None
                     self.triton_attn_func(
                         query,
                         key,
@@ -779,6 +785,7 @@ def forward(
                         self.scale,
                         attn_masks[0][None]
                         if attn_masks is not None else None,
+                        full_scales,
                     )
                 elif self.use_naive_attn:
                     if self.num_kv_heads != self.num_heads:

vllm/attention/layer.py

@@ -90,6 +90,7 @@ def __init__(
         # FlashAttn doesn't support quantizing the kv-cache only
         # but requires q to be quantized as well.
         self._q_scale = torch.tensor(1.0, dtype=torch.float32)
+        self._prob_scale = torch.tensor(1.0, dtype=torch.float32)
 
         # We also keep the float32 versions of k/v_scale for attention
         # backends that don't support tensors (Flashinfer)

vllm/config.py

@@ -3767,6 +3767,17 @@ def _get_quantization_config(
             return quant_config
         return None
 
+    @staticmethod
+    def get_quantization_config(
+            model_config: ModelConfig,
+            load_config: LoadConfig) -> Optional[QuantizationConfig]:
+        import copy
+
+        # For some reason, the _ version of this modifies the model_config
+        # object, so using deepcopy to avoid this problem.
+        return VllmConfig._get_quantization_config(copy.deepcopy(model_config),
+                                                   load_config)
+
     def with_hf_config(
         self,
         hf_config: PretrainedConfig,

vllm/engine/arg_utils.py

@@ -1368,6 +1368,23 @@ def _is_v1_supported_oracle(self, model_config: ModelConfig) -> bool:
                                recommend_to_remove=False)
             return False
 
+        if current_platform.is_rocm():
+            from vllm.model_executor.layers.quantization.fp8 import Fp8Config
+            load_config = self.create_load_config()
+            quantization_config = VllmConfig.get_quantization_config(
+                model_config, load_config)
+            if isinstance(quantization_config, Fp8Config):
+                _raise_or_fallback(feature_name="fp8 for ROCm",
+                                   recommend_to_remove=False)
+                return False
+            from vllm.model_executor.layers.quantization.quark.quark import (
+                QuarkConfig)
+
+            if isinstance(quantization_config, QuarkConfig
+                          ) and quantization_config.has_fp8_layer_weights():
+                _raise_or_fallback(feature_name="Quark fp8 for ROCm",
+                                   recommend_to_remove=False)
+
         # No Fp8 KV cache so far.
         if self.kv_cache_dtype != "auto":
             fp8_attention = self.kv_cache_dtype.startswith("fp8")

vllm/model_executor/layers/quantization/fp8.py

@@ -140,6 +140,11 @@ def get_cache_scale(self, name: str) -> Optional[str]:
             return name.replace(".k_proj.output_scale", ".attn.k_scale")
         if name.endswith(".output_scale") and ".v_proj" in name:
             return name.replace(".v_proj.output_scale", ".attn.v_scale")
+        if name.endswith(".output_scale") and ".q_proj" in name:
+            return name.replace(".q_proj.output_scale", ".attn.q_scale")
+        if name.endswith("self_attn.prob_output_scale"):
+            return name.replace(".prob_output_scale", ".attn.prob_scale")
+        # If no matches, return None
         return None
 
 

vllm/model_executor/layers/quantization/kv_cache.py

@@ -38,6 +38,9 @@ def create_weights(self, layer: torch.nn.Module):
                                            requires_grad=False)
         layer.v_scale = torch.nn.Parameter(torch.tensor(-1.0),
                                            requires_grad=False)
+        # Initialize P = softmax(QK^T) scales
+        layer.prob_scale = torch.nn.Parameter(torch.tensor(-1.0),
+                                              requires_grad=False)
 
     def apply(self, layer: torch.nn.Module) -> torch.Tensor:
         raise RuntimeError(
@@ -97,5 +100,38 @@ def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
                     "may cause accuracy issues. Please make sure k/v_scale "
                     "scaling factors are available in the fp8 checkpoint.")
 
+        if layer.q_scale > 0.0:
+            q_scale = layer.q_scale
+            if current_platform.is_fp8_fnuz():
+                q_scale *= 2
+            layer.calculate_kv_scales = False
+        else:
+            q_scale = 1.0
+        if layer.prob_scale > 0.0:
+            prob_scale = layer.prob_scale
+            if current_platform.is_fp8_fnuz():
+                prob_scale *= 2
+        else:
+            prob_scale = 1.0
+
+        is_singleton_float = lambda x: isinstance(x, float) or isinstance(
+            x, torch.Tensor) and x.numel() == 1 and x.is_floating_point()
+        if not is_singleton_float(q_scale) or not is_singleton_float(
+                prob_scale):
+            raise ValueError("Only support per-tensor scaling factor"
+                             "for fp8-quantized Q/prob")
+
+        # These are used in the final Attention.forward()
+        layer._q_scale.copy_(q_scale)
+        layer._prob_scale.copy_(prob_scale)
+        if q_scale == 1.0 or prob_scale == 1.0:
+            logger.warning_once(
+                f"Using Q scale {q_scale} and prob scale {prob_scale} "
+                "with fp8 attention. This may cause accuracy issues. "
+                "Please make sure Q/prob scaling factors are "
+                "available in the fp8 checkpoint.")
+
         del layer.k_scale
         del layer.v_scale
+        del layer.q_scale
+        del layer.prob_scale

vllm/model_executor/layers/quantization/quark/quark.py

@@ -1,7 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
 
 import fnmatch
-import re
 from typing import Any, Dict, List, Optional, cast
 
 import torch
@@ -125,6 +124,13 @@ def from_config(cls, config: Dict[str, Any]) -> "QuarkConfig":
             for q_config in q_configs:
                 q_config["output_tensors"] = None
 
+            # In case q_proj output is also quantized, remove the configuration
+            # to keep qkv consistency.
+            q_proj_q_config = cast(Dict[str, Any],
+                                   layer_quant_config.get("*q_proj"))
+            if q_proj_q_config is not None:
+                q_proj_q_config["output_tensors"] = None
+
         return cls(quant_config=config,
                    kv_cache_group=kv_cache_group,
                    kv_cache_config=kv_cache_config,
@@ -289,29 +295,30 @@ def get_cache_scale(self, name: str) -> Optional[str]:
         :param name: param name
         :return: matching param name for KV cache scale in vLLM
         """
-        if self.kv_cache_group is None or len(self.kv_cache_group) == 0:
-            return None
-
-        kv_proj_names = [
-            re.split(r"[*.]", kv_cache)[-1] for kv_cache in self.kv_cache_group
-        ]
-        if name.endswith(".output_scale"):
-            if len(kv_proj_names) == 1 and kv_proj_names[0] in name:
-                kv_output_scale_name = "." + kv_proj_names[0] + ".output_scale"
-                return name.replace(kv_output_scale_name, ".attn.k_scale")
-
-            elif len(kv_proj_names) == 2:
-                for kv_proj_name in kv_proj_names:
-                    if kv_proj_name in name and kv_proj_name == "k_proj":
-                        return name.replace(".k_proj.output_scale",
-                                            ".attn.k_scale")
-                    elif kv_proj_name in name and kv_proj_name == "v_proj":
-                        return name.replace(".v_proj.output_scale",
-                                            ".attn.v_scale")
+        if name.endswith(".output_scale") and ".k_proj" in name:
+            return name.replace(".k_proj.output_scale", ".attn.k_scale")
+        if name.endswith(".output_scale") and ".v_proj" in name:
+            return name.replace(".v_proj.output_scale", ".attn.v_scale")
+        if name.endswith(".output_scale") and ".q_proj" in name:
+            return name.replace(".q_proj.output_scale", ".attn.q_scale")
+        if name.endswith("self_attn.prob_output_scale"):
+            return name.replace(".prob_output_scale", ".attn.prob_scale")
 
         # If no matches, return None
         return None
 
+    def has_fp8_layer_weights(self):
+        layer_quant_config = self.quant_config.get("layer_quant_config")
+        to_dict = lambda obj: cast(Dict[str, Any], obj) or {}
+        return any([
+            'fp8' in cast(
+                str,
+                to_dict(
+                    to_dict(to_dict(layer_quant_config).get(layer_name)).get(
+                        "weight")).get("dtype"))
+            for layer_name in ["*v_proj", "*k_proj", "*q_proj"]
+        ])
+
 
 class QuarkLinearMethod(LinearMethodBase):