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.
29 lines
775 B
Python
29 lines
775 B
Python
from dataclasses import dataclass
|
|
from typing import Literal, Tuple
|
|
from urllib.parse import urljoin
|
|
|
|
import librosa
|
|
import numpy as np
|
|
|
|
from vllm.assets.base import get_vllm_public_assets, vLLM_S3_BUCKET_URL
|
|
|
|
ASSET_DIR = "multimodal_asset"
|
|
|
|
|
|
@dataclass(frozen=True)
|
|
class AudioAsset:
|
|
name: Literal["winning_call", "mary_had_lamb"]
|
|
|
|
@property
|
|
def audio_and_sample_rate(self) -> Tuple[np.ndarray, int]:
|
|
|
|
audio_path = get_vllm_public_assets(filename=f"{self.name}.ogg",
|
|
s3_prefix=ASSET_DIR)
|
|
y, sr = librosa.load(audio_path, sr=None)
|
|
assert isinstance(sr, int)
|
|
return y, sr
|
|
|
|
@property
|
|
def url(self) -> str:
|
|
return urljoin(vLLM_S3_BUCKET_URL, f"{ASSET_DIR}/{self.name}.ogg")
|