[Attention] Flash Attention 3 - fp8 (#14570)

Signed-off-by: Mickael Seznec <mickael@mistral.ai>

cmake/external_projects/vllm_flash_attn.cmake

@@ -38,7 +38,7 @@ else()
   FetchContent_Declare(
           vllm-flash-attn
           GIT_REPOSITORY https://github.com/vllm-project/flash-attention.git
-          GIT_TAG 9bfa9869829d8c593527eb34c5271d0090f7ccc9 
+          GIT_TAG dc9d410b3e2d6534a4c70724c2515f4def670a22
           GIT_PROGRESS TRUE
           # Don't share the vllm-flash-attn build between build types
           BINARY_DIR ${CMAKE_BINARY_DIR}/vllm-flash-attn

tests/kernels/test_flash_attn.py

@@ -15,6 +15,7 @@ NUM_HEADS = [(4, 4), (8, 2), (16, 2)]
 HEAD_SIZES = [128, 256]
 BLOCK_SIZES = [16, 32]
 DTYPES = [torch.float16, torch.bfloat16]
+QDTYPES = [None, torch.float8_e4m3fn]
 # one value large enough to test overflow in index calculation.
 # one value small enough to test the schema op check
 NUM_BLOCKS = [32768, 2048]
@@ -85,6 +86,7 @@ def ref_paged_attn(
 @pytest.mark.parametrize("num_blocks", NUM_BLOCKS)
 @pytest.mark.parametrize("sliding_window", [None, 256])
 @pytest.mark.parametrize("fa_version", [2, 3])
+@pytest.mark.parametrize("q_dtype", QDTYPES)
 @torch.inference_mode()
 def test_flash_attn_with_paged_kv(
     use_out: bool,
@@ -97,11 +99,15 @@ def test_flash_attn_with_paged_kv(
     num_blocks: int,
     sliding_window: Optional[int],
     fa_version: int,
+    q_dtype: Optional[torch.dtype],
 ) -> None:
     torch.set_default_device("cuda")
     if not is_fa_version_supported(fa_version):
         pytest.skip(f"Flash attention version {fa_version} not supported due "
                     f"to: \"{fa_version_unsupported_reason(fa_version)}\"")
+    if q_dtype is not None and (dtype != torch.bfloat16 or fa_version == 2):
+        pytest.skip("Flash attention with quantized inputs is only "
+                    "supported on version 3 with bfloat16 base type")
 
     current_platform.seed_everything(0)
     num_seqs = len(kv_lens)
@@ -130,10 +136,28 @@ def test_flash_attn_with_paged_kv(
 
     q = query.unsqueeze(1)
     out = torch.empty_like(q) if use_out else None
+
+    maybe_quantized_query = q
+    maybe_quantized_key_cache = key_cache
+    maybe_quantized_value_cache = value_cache
+    q_descale = None
+    k_descale = None
+    v_descale = None
+    if q_dtype is not None:
+        # QKV are drawn from N(0, 1): no need for a fp8 scaling factor
+        maybe_quantized_query = query.to(q_dtype)
+        maybe_quantized_key_cache = key_cache.to(q_dtype)
+        maybe_quantized_value_cache = value_cache.to(q_dtype)
+
+        scale_shape = (num_seqs, num_kv_heads)
+        q_descale = torch.ones(scale_shape, dtype=torch.float32)
+        k_descale = torch.ones(scale_shape, dtype=torch.float32)
+        v_descale = torch.ones(scale_shape, dtype=torch.float32)
+
     output = flash_attn_with_kvcache(
-        q=q,
+        q=maybe_quantized_query,
-        k_cache=key_cache,
+        k_cache=maybe_quantized_key_cache,
-        v_cache=value_cache,
+        v_cache=maybe_quantized_value_cache,
         out=out,
         softmax_scale=scale,
         causal=True,
@@ -142,10 +166,17 @@ def test_flash_attn_with_paged_kv(
         softcap=soft_cap if soft_cap is not None else 0,
         window_size=window_size,
         fa_version=fa_version,
+        q_descale=q_descale,
+        k_descale=k_descale,
+        v_descale=v_descale,
     )
     output = output if not use_out else out
     output = output.squeeze(1)
 
+    atol, rtol = 1.5e-2, 1e-2
+    if q_dtype is not None:
+        atol, rtol = 1.5e-1, 1.5e-1
+
     ref_output = ref_paged_attn(query=query,
                                 key_cache=key_cache,
                                 value_cache=value_cache,
@@ -155,7 +186,7 @@ def test_flash_attn_with_paged_kv(
                                 scale=scale,
                                 soft_cap=soft_cap,
                                 sliding_window=sliding_window)
-    torch.testing.assert_close(output, ref_output, atol=2e-2, rtol=1e-2), \
+    torch.testing.assert_close(output, ref_output, atol=atol, rtol=rtol), \
         f"{torch.max(torch.abs(output - ref_output))}"
 
 
@@ -171,6 +202,7 @@ def test_flash_attn_with_paged_kv(
 @pytest.mark.parametrize("soft_cap", [None, 10.0, 50.0])
 @pytest.mark.parametrize("num_blocks", NUM_BLOCKS)
 @pytest.mark.parametrize("fa_version", [2, 3])
+@pytest.mark.parametrize("q_dtype", QDTYPES)
 @torch.inference_mode()
 def test_varlen_with_paged_kv(
     use_out: bool,
@@ -183,11 +215,15 @@ def test_varlen_with_paged_kv(
     soft_cap: Optional[float],
     num_blocks: int,
     fa_version: int,
+    q_dtype: Optional[torch.dtype],
 ) -> None:
     torch.set_default_device("cuda")
     if not is_fa_version_supported(fa_version):
         pytest.skip(f"Flash attention version {fa_version} not supported due "
                     f"to: \"{fa_version_unsupported_reason(fa_version)}\"")
+    if q_dtype is not None and (dtype != torch.bfloat16 or fa_version == 2):
+        pytest.skip("Flash attention with quantized inputs is only "
+                    "supported on version 3 with bfloat16 base type")
     current_platform.seed_everything(0)
     num_seqs = len(seq_lens)
     query_lens = [x[0] for x in seq_lens]
@@ -223,10 +259,28 @@ def test_varlen_with_paged_kv(
                                  dtype=torch.int32)
 
     out = torch.empty_like(query) if use_out else None
+
+    maybe_quantized_query = query
+    maybe_quantized_key_cache = key_cache
+    maybe_quantized_value_cache = value_cache
+    q_descale = None
+    k_descale = None
+    v_descale = None
+    if q_dtype is not None:
+        # QKV are drawn from N(0, 1): no need for a fp8 scaling factor
+        maybe_quantized_query = query.to(q_dtype)
+        maybe_quantized_key_cache = key_cache.to(q_dtype)
+        maybe_quantized_value_cache = value_cache.to(q_dtype)
+
+        scale_shape = (num_seqs, num_kv_heads)
+        q_descale = torch.ones(scale_shape, dtype=torch.float32)
+        k_descale = torch.ones(scale_shape, dtype=torch.float32)
+        v_descale = torch.ones(scale_shape, dtype=torch.float32)
+
     output = flash_attn_varlen_func(
-        q=query,
+        q=maybe_quantized_query,
-        k=key_cache,
+        k=maybe_quantized_key_cache,
-        v=value_cache,
+        v=maybe_quantized_value_cache,
         out=out,
         cu_seqlens_q=cu_query_lens,
         seqused_k=kv_lens,
@@ -238,6 +292,9 @@ def test_varlen_with_paged_kv(
         block_table=block_tables,
         softcap=soft_cap if soft_cap is not None else 0,
         fa_version=fa_version,
+        q_descale=q_descale,
+        k_descale=k_descale,
+        v_descale=v_descale,
     )
     output = output if not use_out else out
 
@@ -252,5 +309,8 @@ def test_varlen_with_paged_kv(
         sliding_window=sliding_window,
         soft_cap=soft_cap,
     )
-    torch.testing.assert_close(output, ref_output, atol=2e-2, rtol=1e-2), \
+    atol, rtol = 1.5e-2, 1e-2
+    if q_dtype is not None:
+        atol, rtol = 1.5e-1, 1.5e-1
+    torch.testing.assert_close(output, ref_output, atol=atol, rtol=rtol), \
         f"{torch.max(torch.abs(output - ref_output))}"

vllm/attention/__init__.py

@@ -4,12 +4,16 @@ from vllm.attention.backends.abstract import (AttentionBackend,
                                               AttentionMetadata,
                                               AttentionMetadataBuilder,
                                               AttentionState, AttentionType)
-from vllm.attention.backends.utils import get_flash_attn_version
 from vllm.attention.layer import Attention
 from vllm.attention.selector import get_attn_backend
 
 __all__ = [
-    "Attention", "AttentionBackend", "AttentionMetadata", "AttentionType",
+    "Attention",
-    "AttentionMetadataBuilder", "Attention", "AttentionState",
+    "AttentionBackend",
-    "get_attn_backend", "get_flash_attn_version"
+    "AttentionMetadata",
+    "AttentionType",
+    "AttentionMetadataBuilder",
+    "Attention",
+    "AttentionState",
+    "get_attn_backend",
 ]

vllm/attention/backends/abstract.py

@@ -232,6 +232,7 @@ class AttentionMetadataBuilder(ABC, Generic[T]):
 
 class AttentionLayer(Protocol):
 
+    _q_scale: torch.Tensor
     _k_scale: torch.Tensor
     _v_scale: torch.Tensor
     _k_scale_float: float

vllm/attention/backends/flash_attn.py

@@ -19,10 +19,10 @@ from vllm.attention.backends.abstract import (AttentionBackend, AttentionImpl,
 # yapf: enable
 from vllm.attention.backends.utils import (
     PAD_SLOT_ID, CommonAttentionState, compute_slot_mapping,
-    compute_slot_mapping_start_idx, get_flash_attn_version,
+    compute_slot_mapping_start_idx, get_num_prefill_decode_query_kv_tokens,
-    get_num_prefill_decode_query_kv_tokens, get_seq_len_block_table_args,
+    get_seq_len_block_table_args, is_all_cross_attn_metadata_set,
-    is_all_cross_attn_metadata_set, is_all_encoder_attn_metadata_set,
+    is_all_encoder_attn_metadata_set, is_block_tables_empty)
-    is_block_tables_empty)
+from vllm.fa_utils import get_flash_attn_version
 from vllm.logger import init_logger
 from vllm.multimodal import MultiModalPlaceholderMap
 from vllm.utils import async_tensor_h2d, make_tensor_with_pad
@@ -630,9 +630,11 @@ class FlashAttentionImpl(AttentionImpl):
         self.sliding_window = ((sliding_window - 1,
                                 0) if sliding_window is not None else (-1, -1))
         self.kv_cache_dtype = kv_cache_dtype
-        if is_quantized_kv_cache(self.kv_cache_dtype):
+        self.vllm_flash_attn_version = get_flash_attn_version()
+        if (is_quantized_kv_cache(self.kv_cache_dtype)
+                and self.vllm_flash_attn_version != 3):
             raise NotImplementedError(
-                "FlashAttention with FP8 KV cache not yet supported")
+                "Only FlashAttention3 supports FP8 KV cache")
         if logits_soft_cap is None:
             # In flash-attn, setting logits_soft_cap as 0 means no soft cap.
             logits_soft_cap = 0
@@ -647,7 +649,6 @@ class FlashAttentionImpl(AttentionImpl):
                 f"Head size {head_size} is not supported by FlashAttention. "
                 f"Supported head sizes are: {support_head_sizes}.")
         self.attn_type = attn_type
-        self.vllm_flash_attn_version = get_flash_attn_version()
 
     def forward(
         self,
@@ -671,13 +672,19 @@ class FlashAttentionImpl(AttentionImpl):
                 for profiling run.
             attn_metadata: Metadata for attention.
         NOTE: It in-place updates the output tensor.
+        NOTE: FP8 quantization, flash-attn expect the size of
+              {q,k,v}_descale to be (num_sequences, num_kv_heads).
+              We use torch's .expand() to avoid duplicating values
         """
-        # NOTE(woosuk): FlashAttention does not support FP8 KV cache.
-        assert layer._k_scale_float == 1.0 and layer._v_scale_float == 1.0, (
-            "key/v_scale is not supported in FlashAttention.")
-
         assert output is not None, "Output tensor must be provided."
 
+        # NOTE(woosuk): FlashAttention2 does not support FP8 KV cache.
+        if self.vllm_flash_attn_version < 3 or output.dtype != torch.bfloat16:
+            assert (
+                layer._k_scale_float == 1.0 and layer._v_scale_float == 1.0), (
+                    "key/v_scale is only supported in FlashAttention 3 with "
+                    "base dtype bfloat16")
+
         attn_type = self.attn_type
         if (attn_type == AttentionType.ENCODER
                 and (not attn_metadata.is_all_encoder_attn_metadata_set)):
@@ -694,6 +701,7 @@ class FlashAttentionImpl(AttentionImpl):
         window_size = self.sliding_window
         alibi_slopes: Optional[torch.Tensor] = self.alibi_slopes
         logits_soft_cap: Optional[float] = self.logits_soft_cap
+        fp8_attention = kv_cache_dtype.startswith("fp8")
 
         if kv_cache.numel() > 0:
             key_cache = kv_cache[0]
@@ -729,6 +737,19 @@ class FlashAttentionImpl(AttentionImpl):
                     layer._v_scale,
                 )
 
+                if fp8_attention:
+                    kv_cache = kv_cache.view(torch.float8_e4m3fn)
+                    key_cache = key_cache.view(torch.float8_e4m3fn)
+                    value_cache = value_cache.view(torch.float8_e4m3fn)
+
+        if fp8_attention:
+            num_tokens, num_heads, head_size = query.shape
+            query, _ = ops.scaled_fp8_quant(
+                query.reshape(
+                    (num_tokens, num_heads * head_size)).contiguous(),
+                layer._q_scale)
+            query = query.reshape((num_tokens, num_heads, head_size))
+
         (num_prefill_query_tokens, num_prefill_kv_tokens,
         num_decode_query_tokens) = \
             get_num_prefill_decode_query_kv_tokens(attn_metadata, attn_type)
@@ -753,6 +774,23 @@ class FlashAttentionImpl(AttentionImpl):
                 key = key[:num_prefill_kv_tokens]
                 value = value[:num_prefill_kv_tokens]
 
+                if fp8_attention:
+                    num_kv_tokens, num_kv_heads, head_size = key.shape
+
+                    key, _ = ops.scaled_fp8_quant(
+                        key.reshape((num_kv_tokens,
+                                     num_kv_heads * head_size)).contiguous(),
+                        layer._k_scale)
+                    key = key.reshape((num_kv_tokens, num_kv_heads, head_size))
+
+                    value, _ = ops.scaled_fp8_quant(
+                        value.reshape((num_kv_tokens,
+                                       num_kv_heads * head_size)).contiguous(),
+                        layer._v_scale)
+                    value = value.reshape(
+                        (num_kv_tokens, num_kv_heads, head_size))
+
+                descale_shape = (q_seq_start_loc.shape[0] - 1, key.shape[1])
                 flash_attn_varlen_func(
                     q=query,
                     k=key,
@@ -768,13 +806,19 @@ class FlashAttentionImpl(AttentionImpl):
                     softcap=logits_soft_cap,
                     out=prefill_output,
                     fa_version=self.vllm_flash_attn_version,
+                    q_descale=layer._q_scale.expand(descale_shape),
+                    k_descale=layer._k_scale.expand(descale_shape),
+                    v_descale=layer._v_scale.expand(descale_shape),
                 )
             else:
                 # prefix-enabled attention
                 assert attn_type == AttentionType.DECODER, (
                     "Only decoder-only models support prefix caching")
                 assert prefill_meta.seq_lens is not None
+                assert prefill_meta.query_start_loc is not None
                 max_seq_len = max(prefill_meta.seq_lens)
+                descale_shape = (prefill_meta.query_start_loc.shape[0] - 1,
+                                 key.shape[1])
                 flash_attn_varlen_func(  # noqa
                     q=query,
                     k=key_cache,
@@ -791,6 +835,9 @@ class FlashAttentionImpl(AttentionImpl):
                     softcap=logits_soft_cap,
                     out=prefill_output,
                     fa_version=self.vllm_flash_attn_version,
+                    q_descale=layer._q_scale.expand(descale_shape),
+                    k_descale=layer._k_scale.expand(descale_shape),
+                    v_descale=layer._v_scale.expand(descale_shape),
                 )
 
         if decode_meta := attn_metadata.decode_metadata:
@@ -804,6 +851,9 @@ class FlashAttentionImpl(AttentionImpl):
                 assert attn_type == AttentionType.DECODER, (
                     "Only decoder-only models support max_decode_query_len > 1"
                 )
+                assert decode_meta.query_start_loc is not None
+                descale_shape = (decode_meta.query_start_loc.shape[0] - 1,
+                                 key.shape[1])
                 flash_attn_varlen_func(
                     q=decode_query,
                     k=key_cache,
@@ -820,6 +870,9 @@ class FlashAttentionImpl(AttentionImpl):
                     block_table=decode_meta.block_tables,
                     out=decode_output,
                     fa_version=self.vllm_flash_attn_version,
+                    q_descale=layer._q_scale.expand(descale_shape),
+                    k_descale=layer._k_scale.expand(descale_shape),
+                    v_descale=layer._v_scale.expand(descale_shape),
                 )
             else:
                 # Use flash_attn_with_kvcache for normal decoding.
@@ -828,6 +881,7 @@ class FlashAttentionImpl(AttentionImpl):
                     _,
                     block_tables_arg,
                 ) = get_seq_len_block_table_args(decode_meta, False, attn_type)
+                descale_shape = (seq_lens_arg.shape[0], key_cache.shape[-2])
                 flash_attn_with_kvcache(
                     q=decode_query.unsqueeze(1),
                     k_cache=key_cache,
@@ -841,6 +895,9 @@ class FlashAttentionImpl(AttentionImpl):
                     softcap=logits_soft_cap,
                     out=decode_output.unsqueeze(1),
                     fa_version=self.vllm_flash_attn_version,
+                    q_descale=layer._q_scale.expand(descale_shape),
+                    k_descale=layer._k_scale.expand(descale_shape),
+                    v_descale=layer._v_scale.expand(descale_shape),
                 )
         return output
 

vllm/attention/backends/mla/common.py

@@ -203,9 +203,9 @@ from vllm.attention.backends.abstract import (AttentionBackend, AttentionLayer,
                                               AttentionState, MLAAttentionImpl)
 from vllm.attention.backends.utils import (PAD_SLOT_ID, compute_slot_mapping,
                                            compute_slot_mapping_start_idx,
-                                           get_flash_attn_version,
                                            is_block_tables_empty)
 from vllm.attention.ops.triton_merge_attn_states import merge_attn_states
+from vllm.fa_utils import get_flash_attn_version
 from vllm.model_executor.layers.linear import (ColumnParallelLinear,
                                                LinearBase, RowParallelLinear,
                                                UnquantizedLinearMethod)

vllm/attention/backends/utils.py

@@ -8,13 +8,11 @@ from typing import TYPE_CHECKING, Any, Dict, List, Tuple, Type, TypeVar, Union
 import numpy as np
 import torch
 
-from vllm import envs
 from vllm.attention import (AttentionMetadata, AttentionMetadataBuilder,
                             AttentionState)
 from vllm.attention.backends.abstract import AttentionType
 from vllm.logger import init_logger
 from vllm.multimodal import MultiModalPlaceholderMap
-from vllm.platforms import current_platform
 from vllm.utils import async_tensor_h2d, make_tensor_with_pad
 
 logger = init_logger(__name__)
@@ -585,35 +583,3 @@ def get_num_prefill_decode_query_kv_tokens(
 
     return (num_prefill_query_tokens, num_prefill_kv_tokens,
             num_decode_query_tokens)
-
-
-def get_flash_attn_version():
-    try:
-        from vllm.vllm_flash_attn.flash_attn_interface import (
-            fa_version_unsupported_reason, is_fa_version_supported)
-
-        # if hopper default to FA3, otherwise stick to FA2 for now
-        # TODO(lucas): profile FA3 on ampere to see if it makes sense to
-        #  use FA3 as default for both
-        if current_platform.get_device_capability()[0] == 9:
-            fa_version = 3 if is_fa_version_supported(3) else 2
-        else:
-            fa_version = 2
-
-        if envs.VLLM_FLASH_ATTN_VERSION is not None:
-            assert envs.VLLM_FLASH_ATTN_VERSION in [2, 3]
-            fa_version = envs.VLLM_FLASH_ATTN_VERSION
-            if (current_platform.get_device_capability()[0] == 10
-                    and envs.VLLM_FLASH_ATTN_VERSION == 3):
-                logger.warning("Cannot use FA version 3 on Blackwell platform",
-                               "defaulting to FA version 2.")
-                fa_version = 2
-
-        if not is_fa_version_supported(fa_version):
-            logger.error("Cannot use FA version %d is not supported due to %s",
-                         fa_version, fa_version_unsupported_reason(fa_version))
-
-        assert is_fa_version_supported(fa_version)
-        return fa_version
-    except (ImportError, AssertionError):
-        return None

vllm/attention/layer.py

@@ -84,6 +84,9 @@ class Attention(nn.Module):
         self.calculate_kv_scales = calculate_kv_scales
         self._k_scale = torch.tensor(1.0, dtype=torch.float32)
         self._v_scale = torch.tensor(1.0, dtype=torch.float32)
+        # 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)
 
         # We also keep the float32 versions of k/v_scale for attention
         # backends that don't support tensors (Flashinfer)
@@ -153,6 +156,7 @@ class Attention(nn.Module):
             ).parallel_config.pipeline_parallel_size)
         ]
 
+        self.q_range = torch.tensor(envs.Q_SCALE_CONSTANT, dtype=torch.float32)
         self.k_range = torch.tensor(envs.K_SCALE_CONSTANT, dtype=torch.float32)
         self.v_range = torch.tensor(envs.V_SCALE_CONSTANT, dtype=torch.float32)
 
@@ -178,7 +182,7 @@ class Attention(nn.Module):
         if self.calculate_kv_scales:
             attn_metadata = get_forward_context().attn_metadata
             if attn_metadata.enable_kv_scales_calculation:
-                self.calc_kv_scales(key, value)
+                self.calc_kv_scales(query, key, value)
         if self.use_output:
             output_shape = (output_shape
                             if output_shape is not None else query.shape)
@@ -225,7 +229,8 @@ class Attention(nn.Module):
                 return torch.ops.vllm.unified_attention(
                     query, key, value, self.layer_name)
 
-    def calc_kv_scales(self, key, value):
+    def calc_kv_scales(self, query, key, value):
+        self._q_scale.copy_(torch.abs(query).max() / self.q_range)
         self._k_scale.copy_(torch.abs(key).max() / self.k_range)
         self._v_scale.copy_(torch.abs(value).max() / self.v_range)
         self._k_scale_float = self._k_scale.item()

vllm/envs.py

@@ -78,6 +78,7 @@ if TYPE_CHECKING:
     VLLM_ENABLE_V1_MULTIPROCESSING: bool = True
     VLLM_LOG_BATCHSIZE_INTERVAL: float = -1
     VLLM_DISABLE_COMPILE_CACHE: bool = False
+    Q_SCALE_CONSTANT: int = 200
     K_SCALE_CONSTANT: int = 200
     V_SCALE_CONSTANT: int = 100
     VLLM_SERVER_DEV_MODE: bool = False
@@ -524,13 +525,17 @@ environment_variables: dict[str, Callable[[], Any]] = {
     # Pad the fp8 weights to 256 bytes for ROCm
     "VLLM_ROCM_FP8_PADDING":
     lambda: bool(int(os.getenv("VLLM_ROCM_FP8_PADDING", "1"))),
+
+    # Divisor for dynamic query scale factor calculation for FP8 KV Cache
+    "Q_SCALE_CONSTANT":
+    lambda: int(os.getenv("Q_SCALE_CONSTANT", "200")),
     # Divisor for dynamic key scale factor calculation for FP8 KV Cache
     "K_SCALE_CONSTANT":
     lambda: int(os.getenv("K_SCALE_CONSTANT", "200")),
-
     # Divisor for dynamic value scale factor calculation for FP8 KV Cache
     "V_SCALE_CONSTANT":
     lambda: int(os.getenv("V_SCALE_CONSTANT", "100")),
+
     # If set, enable multiprocessing in LLM for the V1 code path.
     "VLLM_ENABLE_V1_MULTIPROCESSING":
     lambda: bool(int(os.getenv("VLLM_ENABLE_V1_MULTIPROCESSING", "1"))),

vllm/fa_utils.py

@@ -0,0 +1,42 @@
+# SPDX-License-Identifier: Apache-2.0
+from typing import Optional
+
+from vllm import envs
+from vllm.logger import init_logger
+
+logger = init_logger(__name__)
+
+
+def get_flash_attn_version() -> Optional[int]:
+    # import here to avoid circular dependencies
+    from vllm.platforms import current_platform
+    try:
+        from vllm.vllm_flash_attn.flash_attn_interface import (
+            fa_version_unsupported_reason, is_fa_version_supported)
+        device_capability = current_platform.get_device_capability()
+
+        assert device_capability is not None
+
+        # 1. default version depending on platform
+        fa_version = 3 if (device_capability.major == 9
+                           and is_fa_version_supported(3)) else 2
+
+        # 2. override if passed by environment
+        if envs.VLLM_FLASH_ATTN_VERSION is not None:
+            assert envs.VLLM_FLASH_ATTN_VERSION in [2, 3]
+            fa_version = envs.VLLM_FLASH_ATTN_VERSION
+
+        # 3. fallback for unsupported combinations
+        if device_capability.major == 10 and fa_version == 3:
+            logger.warning("Cannot use FA version 3 on Blackwell platform",
+                           "defaulting to FA version 2.")
+            fa_version = 2
+
+        if not is_fa_version_supported(fa_version):
+            logger.error("Cannot use FA version %d is not supported due to %s",
+                         fa_version, fa_version_unsupported_reason(fa_version))
+
+        assert is_fa_version_supported(fa_version)
+        return fa_version
+    except (ImportError, AssertionError):
+        return None

vllm/model_executor/layers/quantization/kv_cache.py

@@ -26,11 +26,14 @@ class BaseKVCacheMethod(QuantizeMethodBase):
 
     def create_weights(self, layer: torch.nn.Module):
         """
-        Create "weight" (aka k_scale and v_scale) for an attention layer.
+        Create "weight" (aka q_scale, k_scale and v_scale)
+        for an attention layer.
         """
-        # Initialize the KV cache scales to -1.0, which is an invalid value.
+        # Initialize the Q and KV cache scales to -1.0, an invalid value.
-        # If the k/v_scale appears in the checkpoint, it will be
+        # If the q and k/v_scales appear in the checkpoint, it will be
         # overwritten when loading weights.
+        layer.q_scale = torch.nn.Parameter(torch.tensor(-1.0),
+                                           requires_grad=False)
         layer.k_scale = torch.nn.Parameter(torch.tensor(-1.0),
                                            requires_grad=False)
         layer.v_scale = torch.nn.Parameter(torch.tensor(-1.0),
@@ -75,6 +78,13 @@ class BaseKVCacheMethod(QuantizeMethodBase):
                 raise ValueError("Only support per-tensor scaling factor "
                                  "for fp8 KV cache")
 
+            if layer.q_scale < 0.0:
+                logger.warning_once(
+                    "Checkpoint does not provide a q scaling factor. "
+                    "Setting it to k_scale. This only matters for "
+                    "the flash-attn backend.")
+                layer._q_scale.copy_(k_scale)
+
             # These are used in the final Attention.forward()
             layer._k_scale.copy_(k_scale)
             layer._v_scale.copy_(v_scale)

vllm/platforms/cuda.py

@@ -14,6 +14,7 @@ from typing_extensions import ParamSpec
 # import custom ops, trigger op registration
 import vllm._C  # noqa
 import vllm.envs as envs
+from vllm.fa_utils import get_flash_attn_version
 from vllm.logger import init_logger
 from vllm.utils import import_pynvml
 
@@ -240,15 +241,6 @@ class CudaPlatformBase(Platform):
                 "Cannot use FlashAttention-2 backend for dtype other than "
                 "torch.float16 or torch.bfloat16.")
             target_backend = _Backend.XFORMERS
-        elif kv_cache_dtype is not None and \
-            kv_cache_dtype.startswith("fp8"):
-            logger.info(
-                "Cannot use FlashAttention-2 backend for FP8 KV cache.")
-            logger.warning(
-                "Please use FlashInfer backend with FP8 KV Cache for "
-                "better performance by setting environment variable "
-                "VLLM_ATTENTION_BACKEND=FLASHINFER")
-            target_backend = _Backend.XFORMERS
         elif block_size % 16 != 0:
             logger.info(
                 "Cannot use FlashAttention-2 backend for block size not "
@@ -270,6 +262,17 @@ class CudaPlatformBase(Platform):
                         "Cannot use FlashAttention-2 backend for head size %d.",
                         head_size)
                     target_backend = _Backend.XFORMERS
+                fp8_kv_cache = (kv_cache_dtype is not None
+                                and kv_cache_dtype.startswith("fp8"))
+                if (fp8_kv_cache and get_flash_attn_version() != 3):
+                    logger.info(
+                        "Cannot use FlashAttention-2 backend for FP8 KV cache."
+                    )
+                    logger.warning(
+                        "Please use FlashInfer backend with FP8 KV Cache for "
+                        "better performance by setting environment variable "
+                        "VLLM_ATTENTION_BACKEND=FLASHINFER")
+                    target_backend = _Backend.XFORMERS
             except ImportError:
                 logger.info(
                     "Cannot use FlashAttention-2 backend because the "

vllm/v1/attention/backends/flash_attn.py

@@ -6,11 +6,12 @@ from typing import TYPE_CHECKING, Any, Optional
 import numpy as np
 import torch
 
+from vllm import _custom_ops as ops
 from vllm.attention.backends.abstract import (AttentionBackend, AttentionImpl,
                                               AttentionMetadata, AttentionType,
                                               is_quantized_kv_cache)
-from vllm.attention.backends.utils import get_flash_attn_version
 from vllm.attention.ops.triton_merge_attn_states import merge_attn_states
+from vllm.fa_utils import get_flash_attn_version
 from vllm.logger import init_logger
 from vllm.platforms import current_platform
 from vllm.utils import cdiv
@@ -226,6 +227,9 @@ class FlashAttentionImpl(AttentionImpl):
             attn_metadata: Metadata for attention.
         Returns:
             shape = [num_tokens, num_heads * head_size]
+        NOTE: FP8 quantization, flash-attn expect the size of
+              {q,k,v}_descale to be (num_sequences, num_kv_heads).
+              We use torch's .expand() to avoid duplicating values
         """
         assert output is not None, "Output tensor must be provided."
 
@@ -259,6 +263,17 @@ class FlashAttentionImpl(AttentionImpl):
             layer._k_scale,
             layer._v_scale,
         )
+        descale_shape = (attn_metadata.query_start_loc.shape[0] - 1,
+                         key.shape[1])
+        if self.kv_cache_dtype.startswith("fp8"):
+            key_cache = key_cache.view(torch.float8_e4m3fn)
+            value_cache = value_cache.view(torch.float8_e4m3fn)
+            num_tokens, num_heads, head_size = query.shape
+            query, _ = ops.scaled_fp8_quant(
+                query.reshape(
+                    (num_tokens, num_heads * head_size)).contiguous(),
+                layer._q_scale)
+            query = query.reshape((num_tokens, num_heads, head_size))
 
         # Compute attention and update output up to `num_actual_tokens`.
         if not attn_metadata.use_cascade:
@@ -279,6 +294,9 @@ class FlashAttentionImpl(AttentionImpl):
                 block_table=attn_metadata.block_table,
                 softcap=self.logits_soft_cap,
                 fa_version=self.vllm_flash_attn_version,
+                q_descale=layer._q_scale.expand(descale_shape),
+                k_descale=layer._k_scale.expand(descale_shape),
+                v_descale=layer._v_scale.expand(descale_shape),
             )
             return output
 
@@ -301,6 +319,9 @@ class FlashAttentionImpl(AttentionImpl):
             block_table=attn_metadata.block_table,
             common_prefix_len=attn_metadata.common_prefix_len,
             fa_version=self.vllm_flash_attn_version,
+            q_descale=layer._q_scale,
+            k_descale=layer._k_scale,
+            v_descale=layer._v_scale,
         )
         return output
 
@@ -391,6 +412,9 @@ def cascade_attention(
     block_table: torch.Tensor,
     common_prefix_len: int,
     fa_version: int,
+    q_descale: Optional[torch.Tensor] = None,
+    k_descale: Optional[torch.Tensor] = None,
+    v_descale: Optional[torch.Tensor] = None,
 ) -> torch.Tensor:
     assert alibi_slopes is None, ("Cascade attention does not support ALiBi.")
     # TODO: Support sliding window.
@@ -402,6 +426,7 @@ def cascade_attention(
     assert common_prefix_len % block_size == 0
     num_common_kv_blocks = common_prefix_len // block_size
     assert num_common_kv_blocks > 0
+    descale_shape = (cu_prefix_query_lens.shape[0] - 1, key_cache.shape[-2])
 
     # Process shared prefix.
     prefix_output, prefix_lse = flash_attn_varlen_func(
@@ -419,8 +444,16 @@ def cascade_attention(
         softcap=logits_soft_cap,
         return_softmax_lse=True,
         fa_version=fa_version,
+        q_descale=q_descale.expand(descale_shape)
+        if q_descale is not None else None,
+        k_descale=k_descale.expand(descale_shape)
+        if k_descale is not None else None,
+        v_descale=v_descale.expand(descale_shape)
+        if v_descale is not None else None,
     )
 
+    descale_shape = (cu_query_lens.shape[0] - 1, key_cache.shape[-2])
+
     # Process suffix per query.
     suffix_output, suffix_lse = flash_attn_varlen_func(
         q=query,
@@ -437,6 +470,12 @@ def cascade_attention(
         softcap=logits_soft_cap,
         return_softmax_lse=True,
         fa_version=fa_version,
+        q_descale=q_descale.expand(descale_shape)
+        if q_descale is not None else None,
+        k_descale=k_descale.expand(descale_shape)
+        if k_descale is not None else None,
+        v_descale=v_descale.expand(descale_shape)
+        if v_descale is not None else None,
     )
 
     # Merge prefix and suffix outputs, and store the result in output.

vllm/v1/executor/multiproc_executor.py

@@ -5,6 +5,7 @@ import pickle
 import signal
 import sys
 import time
+import traceback
 import weakref
 from dataclasses import dataclass
 from enum import Enum, auto
@@ -370,6 +371,9 @@ class WorkerProc:
                     func = partial(cloudpickle.loads(method), self.worker)
                 output = func(*args, **kwargs)
             except Exception as e:
+                # Notes have been introduced in python 3.11
+                if hasattr(e, "add_note"):
+                    e.add_note(traceback.format_exc())
                 self.worker_response_mq.enqueue(
                     (WorkerProc.ResponseStatus.FAILURE, e))
                 logger.exception("WorkerProc hit an exception: %s", exc_info=e)

vllm/v1/worker/gpu_model_runner.py

@@ -1558,7 +1558,7 @@ class GPUModelRunner(LoRAModelRunnerMixin):
                     block_size=block_size,
                     num_kv_heads=attn_module.num_kv_heads,
                     head_size=attn_module.head_size,
-                    dtype=attn_module.dtype,
+                    dtype=self.kv_cache_dtype,
                     use_mla=use_mla)
             elif attn_module.attn_type in (AttentionType.ENCODER,
                                            AttentionType.ENCODER_ONLY):