accelerate>=1.13.0
torch>=2.8.0
safetensors>=0.7.0
transformers>=5.4.0
threadpoolctl>=3.6.0
packaging>=24.2
device-smi>=0.5.5
protobuf>=7.34.0
pillow>=11.3.0
pypcre>=0.3.2
tokenicer>=0.0.13
logbar>=0.4.3
jinja2>=3.1.0
ninja>=1.13.0
maturin>=1.9.4
datasets>=3.6.0
pyarrow>=21.0
dill>=0.3.8
torchao>=0.16.0
defuser>=0.0.21

[:python_version < "3.14"]
numpy==2.2.6

[:python_version >= "3.14"]
numpy>=2.3.0

[bitblas]
bitblas==0.1.0.post1

[bitsandbytes]
bitsandbytes>=0.49.3

[eval]
Evalution

[hf]
optimum>=1.21.2

[marlin-cuda]
nvidia-cuda-runtime>=13.0.96
nvidia-cublas>=13.1.0.3
nvidia-cusparse>=12.6.3.3
nvidia-cusolver>=12.0.4.66

[marlin-cuda12]
nvidia-cuda-runtime-cu12==12.9.79
nvidia-cublas-cu12==12.9.1.4
nvidia-cusparse-cu12==12.5.10.65
nvidia-cusolver-cu12==11.7.5.82

[mlx]
mlx_lm>=0.24.0

[openai]
uvicorn
fastapi
pydantic

[quality]
ruff==0.14.2

[sglang]
sglang[srt]>=0.4.6
flashinfer-python>=0.3.1

[test]
pytest>=8.3.5
pytest-timeout>=2.3.1
parameterized

[triton]
triton>=3.4.0

[vllm]
vllm>=0.10.2
flashinfer-python>=0.3.1
