[BUG] Failing to build from source

[chengyupku]

Describe the bug
I tried to build from source as follows:

conda create -n flux python=3.11
conda activate flux
pip3 install packaging
pip3 install ninja
pip3 install torch==2.6.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124

./build.sh --clean-all
./build.sh --arch "80;89;90" --nvshmem --package

But I encountered a compilation error when building_flux_cuda:

/home/dlisuser/cy/flux/include/flux/flux.h:1047:34: error: no match for ‘operator<’ (operand types are ‘const cute::tuple<long int, long int, long int>’ and ‘const cute::tuple<long int, long int, long int>’)
 1047 |       return bool(cute::get<I>(t) < cute::get<I>(u)) ||

Environment
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2024 NVIDIA Corporation
Built on Tue_Feb_27_16:19:38_PST_2024
Cuda compilation tools, release 12.4, V12.4.99
Build cuda_12.4.r12.4/compiler.33961263_0

H100 NVL

[houqi]

check the READMe.md and run the install_deps.sh. there is a CUTLASS patch which helps.

git clone --recursive https://github.com/bytedance/flux.git && cd flux

# Install dependencies
bash ./install_deps.sh

# For Ampere(sm80) GPU
./build.sh --arch 80 --nvshmem
# For Ada Lovelace(sm89) GPU
./build.sh --arch 89 --nvshmem
# For Hopper(sm90) GPU
./build.sh --arch 90 --nvshmem

[chengyupku]

I have already run bash ./install_deps.sh, but the bug still exists. 😕

[houqi]

@ZSL98 please help?

[jiurizz]

same error

environment: nvcr.io/nvidia/pytorch:24.07-py3 container with:

• CUDA: 12.5.82
• GCC: 11.4.0
• Python: 3.10.12
• PyTorch: 2.4.0

[ZSL98]

environment: nvcr.io/nvidia/pytorch:24.07-py3 container with:

I'll try this image today.