[DSV4] Add knob to enable pre-attn gemm (#41443)

Signed-off-by: Yongye Zhu <zyy1102000@gmail.com>
This commit is contained in:
Yongye Zhu
2026-05-01 12:23:17 -07:00
committed by GitHub
parent a9484dac7b
commit bcf5cac9fb
3 changed files with 24 additions and 3 deletions
+12
View File
@@ -245,6 +245,7 @@ if TYPE_CHECKING:
VLLM_DEBUG_WORKSPACE: bool = False
VLLM_DISABLE_SHARED_EXPERTS_STREAM: bool = False
VLLM_SHARED_EXPERTS_STREAM_TOKEN_THRESHOLD: int = 256
VLLM_MULTI_STREAM_GEMM_TOKEN_THRESHOLD: int = 4096
VLLM_COMPILE_CACHE_SAVE_FORMAT: Literal["binary", "unpacked"] = "binary"
VLLM_USE_V2_MODEL_RUNNER: bool = False
VLLM_LOG_MODEL_INSPECTION: bool = False
@@ -1678,6 +1679,17 @@ environment_variables: dict[str, Callable[[], Any]] = {
"VLLM_SHARED_EXPERTS_STREAM_TOKEN_THRESHOLD": lambda: int(
int(os.getenv("VLLM_SHARED_EXPERTS_STREAM_TOKEN_THRESHOLD", 256))
),
# Token-count cutoff for multi-stream overlap of the attention input
# GEMM with auxiliary GEMMs (e.g. fused_wqa_wkv overlapped with indexer
# weights / kv-score projections in DeepSeek-V4). At or below this many
# tokens the FP8 main GEMM has idle SMs to share with the bf16 aux GEMMs
# and overlap is a 5-45% win; above it the FP8 GEMM saturates the device
# and the cross-stream sync becomes pure overhead. Set to 0 to disable
# the multi-stream path entirely. Empirical crossover on B300 (148 SMs)
# is ~4096; B200 (132 SMs) is expected ~3072.
"VLLM_MULTI_STREAM_GEMM_TOKEN_THRESHOLD": lambda: int(
os.getenv("VLLM_MULTI_STREAM_GEMM_TOKEN_THRESHOLD", "4096")
),
# Format for saving torch.compile cache artifacts
# - "binary": saves as binary file
# Safe for multiple vllm serve processes accessing the same torch compile cache.
@@ -13,6 +13,7 @@ import torch.nn as nn
import torch.nn.functional as F
from transformers import DeepseekV2Config, DeepseekV3Config
import vllm.envs as envs
from vllm.model_executor.layers.linear import (
ReplicatedLinear,
)
@@ -385,6 +386,8 @@ class DeepseekV4MultiHeadLatentAttentionWrapper(PluggableLayer):
self.ln_events[0],
self.ln_events[1:4],
self.aux_stream_list[:3],
enable=hidden_states.shape[0]
<= envs.VLLM_MULTI_STREAM_GEMM_TOKEN_THRESHOLD,
)
return qr_kv, kv_score, indexer_kv_score, indexer_weights
+9 -3
View File
@@ -64,6 +64,7 @@ def execute_in_parallel(
start_event: torch.cuda.Event,
done_events: list[torch.cuda.Event],
aux_streams: list[torch.cuda.Stream] | None = None,
enable: bool = False,
) -> tuple[Any, list[Any]]:
"""Run default_fn on the current stream and aux_fns concurrently on
aux_streams.
@@ -74,8 +75,9 @@ def execute_in_parallel(
start_event fans out from the current stream to every launched aux stream;
done_events[i] is recorded after aux_fns[i] so the current stream joins
before returning. When aux_streams is None, all aux_fns run sequentially
on the current stream.
before returning. Falls back to sequential execution on the current stream
when aux_streams is None or enable is False; in that case default_fn runs
first, then aux_fns in order.
Args:
default_fn: Callable for the default (current) stream.
@@ -86,13 +88,17 @@ def execute_in_parallel(
corresponding aux_fn. Length must match aux_fns.
aux_streams: Per-aux CUDA streams. Length must match aux_fns.
Multi-stream is disabled when None.
enable: Opt-in switch for the multi-stream path. Defaults to False,
so callers that pass aux_streams must also pass enable=True
(typically gated by an env var) to actually overlap. When False,
execution falls back to sequential on the current stream.
Returns:
Tuple of (default_result, aux_results) where aux_results[i] is the
result of aux_fns[i] (or None when skipped).
"""
aux_results: list[Any]
if aux_streams is None:
if aux_streams is None or not enable:
default_result = default_fn()
aux_results = [fn() if fn is not None else None for fn in aux_fns]
return default_result, aux_results