[AMD] [Quantization] Add override flag for attention dtype instead of using kv_cache_dtype trigger (#17331)

Signed-off-by: Randall Smith <Randall.Smith@amd.com>

vllm/attention/backends/rocm_flash_attn.py

@@ -17,6 +17,7 @@ from vllm.attention.backends.utils import (CommonAttentionState,
                                            CommonMetadataBuilder)
 from vllm.attention.ops.paged_attn import (PagedAttention,
                                            PagedAttentionMetadata)
+from vllm.config import get_current_vllm_config
 from vllm.logger import init_logger
 from vllm.platforms import current_platform
 from vllm.platforms.rocm import use_rocm_custom_paged_attention
@@ -584,6 +585,10 @@ class ROCmFlashAttentionImpl(AttentionImpl):
                 logger.debug("Using naive (SDPA) attention in ROCmBackend")
 
         self.aiter_kv_scales_initialized = False
+        self.force_fp8_attention = (
+            get_current_vllm_config() is not None
+            and get_current_vllm_config().model_config.override_attention_dtype
+            == "fp8")
 
     def repeat_kv(self, x: torch.Tensor, n_rep: int) -> torch.Tensor:
         """torch.repeat_interleave(x, dim=1, repeats=n_rep)"""
@@ -770,9 +775,12 @@ class ROCmFlashAttentionImpl(AttentionImpl):
                             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")
+                                      and (self.kv_cache_dtype == "fp8"
+                                           or self.force_fp8_attention))
+
                     full_scales = (
                         layer._q_scale.item(), layer._k_scale.item(),
                         layer._v_scale.item(),

vllm/config.py

@@ -417,6 +417,8 @@ class ModelConfig:
     available.\n
     - "vllm" will use the vLLM model implementation.\n
     - "transformers" will use the Transformers model implementation."""
+    override_attention_dtype: Optional[str] = None
+    """Override dtype for attention"""
 
     def compute_hash(self) -> str:
         """
@@ -517,6 +519,12 @@ class ModelConfig:
 
         from vllm.platforms import current_platform
 
+        if (self.override_attention_dtype is not None
+                and not current_platform.is_rocm()):
+            warnings.warn(
+                "override-attention-dtype is set but not using ROCm platform",
+                stacklevel=2)
+
         if (self.enable_sleep_mode
                 and not current_platform.is_sleep_mode_available()):
             raise ValueError(

vllm/engine/arg_utils.py

@@ -429,6 +429,7 @@ class EngineArgs:
     override_generation_config: dict[str, Any] = \
         get_field(ModelConfig, "override_generation_config")
     model_impl: str = ModelConfig.model_impl
+    override_attention_dtype: str = ModelConfig.override_attention_dtype
 
     calculate_kv_scales: bool = CacheConfig.calculate_kv_scales
 
@@ -549,6 +550,8 @@ class EngineArgs:
         model_group.add_argument("--model-impl",
                                  choices=[f.value for f in ModelImpl],
                                  **model_kwargs["model_impl"])
+        model_group.add_argument("--override-attention-dtype",
+                                 **model_kwargs["override_attention_dtype"])
 
         # Model loading arguments
         load_kwargs = get_kwargs(LoadConfig)
@@ -946,6 +949,7 @@ class EngineArgs:
             override_generation_config=self.override_generation_config,
             enable_sleep_mode=self.enable_sleep_mode,
             model_impl=self.model_impl,
+            override_attention_dtype=self.override_attention_dtype,
         )
 
     def create_load_config(self) -> LoadConfig: