squall/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[SpecDecode] Support FlashInfer in DraftModelRunner (#6926)

Browse Source
This commit is contained in:
Bongwon Jang 2024-08-06 00:05:05 +09:00 committed by GitHub
parent 82a1b1a82b
commit e9630458c7
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 47 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

47
vllm/spec_decode/draft_model_runner.py
View File

@ -11,6 +11,17 @@ except ModuleNotFoundError:
from vllm.attention.backends.rocm_flash_attn import ( from vllm.attention.backends.rocm_flash_attn import (
ROCmFlashAttentionMetadata as FlashAttentionMetadata) ROCmFlashAttentionMetadata as FlashAttentionMetadata)
try:
from flashinfer import BatchDecodeWithPagedKVCacheWrapper
from flashinfer.decode import CUDAGraphBatchDecodeWithPagedKVCacheWrapper
from flashinfer.prefill import BatchPrefillWithPagedKVCacheWrapper
FLASHINFER_WORKSPACE_BUFFER_SIZE = 256 * 1024 * 1024
except ImportError:
BatchDecodeWithPagedKVCacheWrapper = None
CUDAGraphBatchDecodeWithPagedKVCacheWrapper = None
BatchPrefillWithPagedKVCacheWrapper = None
FLASHINFER_WORKSPACE_BUFFER_SIZE = 0
from vllm.config import (CacheConfig, DeviceConfig, LoadConfig, LoRAConfig, from vllm.config import (CacheConfig, DeviceConfig, LoadConfig, LoRAConfig,
ModelConfig, MultiModalConfig, ParallelConfig, ModelConfig, MultiModalConfig, ParallelConfig,
PromptAdapterConfig, SchedulerConfig) PromptAdapterConfig, SchedulerConfig)
@ -79,6 +90,11 @@ class TP1DraftModelRunner(ModelRunner):
return_hidden_states=return_hidden_states, return_hidden_states=return_hidden_states,
) )
self.flashinfer_decode_workspace_buffer = None
self.flashinfer_decode_wrapper = None
self.flashinfer_prefill_workspace_buffer = None
self.flashinfer_prefill_wrapper = None
def _update_flash_attn_metadata(self, attn_metadata, num_seqs, def _update_flash_attn_metadata(self, attn_metadata, num_seqs,
num_queries): num_queries):
assert isinstance(attn_metadata, FlashAttentionMetadata) assert isinstance(attn_metadata, FlashAttentionMetadata)
@ -286,6 +302,37 @@ class TP1DraftModelRunner(ModelRunner):
model_input.prompt_adapter_requests, model_input.prompt_adapter_requests,
model_input.prompt_adapter_mapping) model_input.prompt_adapter_mapping)
if self.attn_backend.get_name() == "flashinfer":
assert model_input.attn_metadata is not None
assert model_input.input_tokens is not None
if self.flashinfer_decode_workspace_buffer is None:
self.flashinfer_decode_workspace_buffer = torch.empty(
FLASHINFER_WORKSPACE_BUFFER_SIZE,
dtype=torch.uint8,
device=self.device)
self.flashinfer_decode_wrapper = \
BatchDecodeWithPagedKVCacheWrapper(
self.flashinfer_decode_workspace_buffer, "NHD")
self.flashinfer_prefill_workspace_buffer = torch.empty(
FLASHINFER_WORKSPACE_BUFFER_SIZE,
dtype=torch.uint8,
device=self.device)
self.flashinfer_prefill_wrapper = \
BatchPrefillWithPagedKVCacheWrapper(
self.flashinfer_prefill_workspace_buffer, "NHD")
model_input.attn_metadata.prefill_wrapper = \
self.flashinfer_prefill_wrapper
if model_input.attn_metadata.use_cuda_graph:
batch_size = model_input.input_tokens.shape[0]
model_input.attn_metadata.decode_wrapper = \
self.graph_runners[model_input.
virtual_engine][batch_size].flashinfer_decode_wrapper
else:
model_input.attn_metadata.decode_wrapper = \
self.flashinfer_decode_wrapper
model_input.attn_metadata.begin_forward()
# Detect exec mode # Detect exec mode
assert model_input.attn_metadata is not None assert model_input.attn_metadata is not None
use_cuda_graph = False use_cuda_graph = False