9c34570ffc
The mtlgemm Metal extension's metal_context.mm calls at::mps::dispatch_sync_with_rethrow, which was added to that namespace in PyTorch 2.11.0 (pytorch/pytorch#167445, merged 2025-11-11). Earlier stable releases (2.6 – 2.10) only expose it under at::native::mps::, so installing mtlgemm against them fails the C++ extension build. torchvision pinned to >=0.26.0 (matches torch 2.11) to keep the torch/torchvision wheel pair ABI-compatible on resolution.
46 lines
1.1 KiB
Plaintext
46 lines
1.1 KiB
Plaintext
# Trellis2 MLX requirements (Apple Silicon)
|
||
# NO: flash_attn, spconv, torchsparse (CUDA-only)
|
||
|
||
# Core
|
||
# torch >= 2.11.0 is required because mtlgemm's metal_context.mm calls
|
||
# at::mps::dispatch_sync_with_rethrow, which was promoted from
|
||
# at::native::mps:: to at::mps:: in PyTorch 2.11 (PR #167445, merged
|
||
# 2025-11-11). Earlier stable versions (2.6 – 2.10) only expose it under
|
||
# at::native::mps:: and won't link.
|
||
torch>=2.11.0
|
||
torchvision>=0.26.0
|
||
transformers>=4.40.0,<5
|
||
safetensors
|
||
huggingface_hub
|
||
pillow
|
||
numpy
|
||
|
||
# Mesh processing
|
||
trimesh
|
||
xatlas
|
||
fast-simplification
|
||
pygltflib
|
||
|
||
# Image processing
|
||
opencv-python-headless
|
||
|
||
# Server
|
||
fastapi
|
||
uvicorn[standard]
|
||
pydantic>=2.0
|
||
|
||
# MLX (Apple Silicon native)
|
||
mlx>=0.31.0
|
||
|
||
# Metal GPU packages (Apple Silicon — install with: pip install --no-build-isolation <url>)
|
||
mtldiffrast @ https://github.com/pedronaugusto/mtldiffrast/archive/main.tar.gz
|
||
cumesh @ https://github.com/pedronaugusto/mtlmesh/archive/main.tar.gz
|
||
flex_gemm @ https://github.com/pedronaugusto/mtlgemm/archive/main.tar.gz
|
||
|
||
# Misc
|
||
tqdm
|
||
imageio
|
||
imageio-ffmpeg
|
||
scipy
|
||
einops
|