43 lines
1.3 KiB
Python
43 lines
1.3 KiB
Python
import os, subprocess
|
|
from pathlib import Path
|
|
|
|
def _pick_free_gpu():
|
|
if "CUDA_VISIBLE_DEVICES" in os.environ:
|
|
return os.environ["CUDA_VISIBLE_DEVICES"]
|
|
try:
|
|
out = subprocess.check_output(
|
|
["nvidia-smi", "--query-gpu=index,memory.free", "--format=csv,noheader"],
|
|
text=True,
|
|
)
|
|
lines = [l.strip() for l in out.strip().split("\n") if l.strip()]
|
|
best = max(lines, key=lambda x: int(x.split(",")[1].strip().split()[0]))
|
|
gpu = best.split(",")[0].strip()
|
|
except Exception:
|
|
gpu = "0"
|
|
os.environ["CUDA_VISIBLE_DEVICES"] = gpu
|
|
return gpu
|
|
|
|
if __name__ == "__main__":
|
|
_pick_free_gpu()
|
|
|
|
model = str(Path(__file__).resolve().parent / "models/Qwen2.5-7B-Instruct")
|
|
|
|
os.environ["FLASHINFER_DISABLE_VERSION_CHECK"] = "1"
|
|
|
|
os.execvp("vllm", [
|
|
"vllm", "serve",
|
|
model,
|
|
"--host", "0.0.0.0",
|
|
"--port", "8000",
|
|
"--served-model-name", "Qwen2.5-7B-Instruct",
|
|
"--trust-remote-code",
|
|
"--max-model-len", "32768",
|
|
"--dtype", "bfloat16",
|
|
"--gpu-memory-utilization", "0.85",
|
|
"--enforce-eager",
|
|
"--enable-auto-tool-choice",
|
|
"--tool-call-parser", "openai",
|
|
"--generation-config", "vllm",
|
|
"--override-generation-config", '{"temperature": 0.8, "top_k": 50, "top_p": 0.9}',
|
|
])
|