ef6abf3dc7
Adds ALL files needed for Dockerfile build:
- qwen3_6_scripts/ (baseline patches + our optimizations)
- vllm/ (full vllm package)
- paged_attention_v2_pytorch.py (V2 with single-bmm optimization)
- Dockerfile + computility-run.yaml
Our optimizations vs baseline:
1. paged_attn.py: pre-gathered context KV (eliminates 194 gather calls),
Triton try/fallback, V2 heuristic, threshold 32K→64K
2. paged_attention_v2_pytorch.py: fills NotImplementedError,
single-bmm Phase 1 (195 launches → 3)
3. patch_enable_triton.py: HAS_TRITON=True with safety fallback
4. patch_triton_tuning.py: BLOCK=64, NUM_WARPS=4 for BI-V100
5. computility-run.yaml: gpu-memory-utilization 0.9→0.95,
max-num-batched-tokens 8192→16384
This repo can now be submitted to dev.modelhub.org.cn as-is.
104 lines
2.9 KiB
Python
104 lines
2.9 KiB
Python
from typing import (AsyncGenerator, List, Mapping, Optional, Protocol,
|
|
runtime_checkable)
|
|
|
|
from vllm.config import DecodingConfig, ModelConfig
|
|
from vllm.core.scheduler import SchedulerOutputs
|
|
from vllm.inputs.data import PromptType
|
|
from vllm.lora.request import LoRARequest
|
|
from vllm.model_executor.layers.sampler import SamplerOutput
|
|
from vllm.outputs import EmbeddingRequestOutput, RequestOutput
|
|
from vllm.pooling_params import PoolingParams
|
|
from vllm.prompt_adapter.request import PromptAdapterRequest
|
|
from vllm.sampling_params import SamplingParams
|
|
from vllm.transformers_utils.tokenizer import AnyTokenizer
|
|
|
|
|
|
@runtime_checkable
|
|
class EngineClient(Protocol):
|
|
"""Protocol class for Clients to Engine"""
|
|
|
|
@property
|
|
def is_running(self) -> bool:
|
|
...
|
|
|
|
@property
|
|
def is_stopped(self) -> bool:
|
|
...
|
|
|
|
@property
|
|
def errored(self) -> bool:
|
|
...
|
|
|
|
@property
|
|
def dead_error(self) -> BaseException:
|
|
...
|
|
|
|
def generate(
|
|
self,
|
|
prompt: PromptType,
|
|
sampling_params: SamplingParams,
|
|
request_id: str,
|
|
lora_request: Optional[LoRARequest] = None,
|
|
trace_headers: Optional[Mapping[str, str]] = None,
|
|
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
|
priority: int = 0,
|
|
) -> AsyncGenerator[RequestOutput, None]:
|
|
"""Generate outputs for a request."""
|
|
...
|
|
|
|
def encode(
|
|
self,
|
|
prompt: PromptType,
|
|
pooling_params: PoolingParams,
|
|
request_id: str,
|
|
lora_request: Optional[LoRARequest] = None,
|
|
trace_headers: Optional[Mapping[str, str]] = None,
|
|
priority: int = 0,
|
|
) -> AsyncGenerator[EmbeddingRequestOutput, None]:
|
|
"""Generate outputs for a request from an embedding model."""
|
|
...
|
|
|
|
async def abort(self, request_id: str) -> None:
|
|
"""Abort a request.
|
|
|
|
Args:
|
|
request_id: The unique id of the request.
|
|
"""
|
|
|
|
async def get_model_config(self) -> ModelConfig:
|
|
"""Get the model configuration of the vLLM engine."""
|
|
...
|
|
|
|
async def get_decoding_config(self) -> DecodingConfig:
|
|
...
|
|
"""Get the decoding configuration of the vLLM engine."""
|
|
|
|
async def get_tokenizer(
|
|
self,
|
|
lora_request: Optional[LoRARequest] = None,
|
|
) -> AnyTokenizer:
|
|
"""Get the appropriate tokenizer for the request"""
|
|
...
|
|
|
|
async def is_tracing_enabled(self) -> bool:
|
|
...
|
|
|
|
async def do_log_stats(
|
|
self,
|
|
scheduler_outputs: Optional[SchedulerOutputs] = None,
|
|
model_output: Optional[List[SamplerOutput]] = None,
|
|
) -> None:
|
|
...
|
|
|
|
async def check_health(self) -> None:
|
|
"""Raise if unhealthy"""
|
|
...
|
|
|
|
async def start_profile(self) -> None:
|
|
"""Start profiling the engine"""
|
|
...
|
|
|
|
async def stop_profile(self) -> None:
|
|
"""Start profiling the engine"""
|
|
...
|