# PyTorch with CUDA 12.8 / 13.0
safetensors==0.7.0
--extra-index-url https://download.pytorch.org/whl/cu128
--extra-index-url https://download.pytorch.org/whl/cu130
torch==2.7.1+cu128; sys_platform == 'win32'
torchvision==0.22.1+cu128; sys_platform == 'win32'
torchaudio==2.7.1+cu128; sys_platform == 'win32'
# macOS arm64 (Apple Silicon): latest CPU/MPS wheels
torch>=2.9.1; sys_platform == 'darwin' and platform_machine == 'arm64'
torchaudio>=2.9.1; sys_platform == 'darwin' and platform_machine == 'arm64'
torchvision; sys_platform == 'darwin' and platform_machine == 'arm64'
# Linux x86_64 with CUDA 12.8
torch==2.10.0+cu128; sys_platform == 'linux' and platform_machine == 'x86_64'
torchvision==0.25.0+cu128; sys_platform == 'linux' and platform_machine == 'x86_64'
torchaudio==2.10.0+cu128; sys_platform == 'linux' and platform_machine == 'x86_64'
# Linux aarch64 with CUDA 13.0 (e.g. NVIDIA DGX Spark)
torch==2.10.0+cu130; sys_platform == 'linux' and platform_machine == 'aarch64'
torchvision==0.25.0+cu130; sys_platform == 'linux' and platform_machine == 'aarch64'
torchaudio==2.10.0+cu130; sys_platform == 'linux' and platform_machine == 'aarch64'

# Core dependencies
transformers>=4.51.0,<4.58.0
diffusers
gradio==6.2.0
matplotlib>=3.7.5
scipy>=1.10.1
soundfile>=0.13.1
loguru>=0.7.3
einops>=0.8.1
accelerate>=1.12.0
fastapi>=0.110.0
uvicorn[standard]>=0.27.0
numba>=0.63.1
vector-quantize-pytorch>=1.27.15
torchcodec>=0.9.1; platform_machine != 'aarch64'
torchao
toml
modelscope

# Training dependencies (required for LoRA/LoKr training)
peft>=0.18.0
lycoris-lora
lightning>=2.0.0
tensorboard>=2.0.0

# MLX dependencies (Apple Silicon native acceleration - macOS only)
# Provides significant speedup for LLM inference on Apple Silicon GPUs
mlx>=0.25.2; sys_platform == 'darwin' and platform_machine == 'arm64'
mlx-lm>=0.20.0; sys_platform == 'darwin' and platform_machine == 'arm64'

# nano-vllm dependencies
triton-windows>=3.0.0,<3.4; sys_platform == 'win32'
triton>=3.0.0; sys_platform == 'linux'
flash-attn @ https://github.com/sdbds/flash-attention-for-windows/releases/download/2.8.2/flash_attn-2.8.2+cu128torch2.7.1cxx11abiFALSEfullbackward-cp311-cp311-win_amd64.whl ; sys_platform == 'win32' and python_version == '3.11' and platform_machine == 'AMD64'
flash-attn; sys_platform == 'linux' and platform_machine == 'x86_64'
xxhash

# Local package - install with: pip install -e acestep/third_parts/nano-vllm
# nano-vllm