Update Wall-X to 1.1.0 (#104)

This commit is contained in:
Starrick Liu
2026-06-15 11:40:00 +08:00
committed by GitHub
parent e23a586846
commit 72834e7de5
200 changed files with 33916 additions and 16771 deletions
+61 -46
View File
@@ -1,63 +1,78 @@
import os
import torch
from pathlib import Path
from setuptools import setup, find_packages
from setuptools import find_packages, setup
from torch.utils.cpp_extension import BuildExtension, CUDAExtension
cwd = Path(os.path.dirname(os.path.abspath(__file__)))
ROOT = Path(__file__).resolve().parent
OPS_DIR = Path("wall_x") / "model" / "core" / "ops"
CSRC_DIR = OPS_DIR / "csrc"
nvcc_flags = [
"-std=c++17", # NOTE: CUTLASS requires c++17
"-DENABLE_BF16", # Enable BF16 for cuda_version >= 11
PUBLIC_SCRIPTS = [
"scripts/compute_norm_stats.py",
"scripts/draw_openloop_plot.py",
"scripts/fake_inference.py",
"scripts/infer_libero.py",
"scripts/merge_sharded_weights.py",
"scripts/merge_tokenizer.py",
"scripts/run_libero.sh",
"scripts/run_serving.sh",
]
env_arch_list = os.environ.get("TORCH_CUDA_ARCH_LIST", None)
if env_arch_list:
# Let PyTorch builder to choose device to target for.
device_capability = ""
else:
device_capability = torch.cuda.get_device_capability()
device_capability = f"{device_capability[0]}{device_capability[1]}"
def read_readme() -> str:
readme = ROOT / "README.md"
return readme.read_text(encoding="utf-8") if readme.exists() else ""
if device_capability:
nvcc_flags.extend(
[
f"--generate-code=arch=compute_{device_capability},code=sm_{device_capability}",
f"-DGROUPED_GEMM_DEVICE_CAPABILITY={device_capability}",
]
def build_ext_modules():
binding = CSRC_DIR / "binding.cu"
if not (ROOT / binding).exists():
raise RuntimeError(
"CUDA operator sources are missing. The OSS export must include "
"wall_x/model/core/ops/csrc/binding.cu."
)
cuda_sources = [binding] + sorted(
path for path in (ROOT / CSRC_DIR).rglob("*.cu") if path.name != "binding.cu"
)
cuda_sources = [
path if path == binding else path.relative_to(ROOT) for path in cuda_sources
]
return [
CUDAExtension(
name="wall_x.model.core.ops._cuda_ext_bin",
sources=[str(path) for path in cuda_sources],
include_dirs=[str(CSRC_DIR), str(CSRC_DIR / "common")],
extra_compile_args={
"cxx": ["-O3", "-std=c++17"],
"nvcc": ["-O3", "--use_fast_math", "-std=c++17"],
},
)
]
ext_modules = [
CUDAExtension(
"wallx_csrc",
[
"csrc/ops.cu",
"csrc/dual_asym_grouped_gemm.cu",
"csrc/permute.cu",
"csrc/rope.cu",
"csrc/rope_index.cu",
"csrc/rot_pos.cu",
"csrc/window_index.cu",
],
include_dirs=[f"{cwd}/3rdparty/cutlass/include/", f"{cwd}/csrc"],
extra_compile_args={
"cxx": ["-fopenmp", "-fPIC", "-Wno-strict-aliasing"],
"nvcc": nvcc_flags,
},
)
]
setup(
name="wall_x",
version="1.0.1",
author="X2Robot Team",
version="1.1.0",
description="Training and inference code for WALL open-source embodied models.",
long_description=read_readme(),
long_description_content_type="text/markdown",
author="X-Square Robot",
url="https://github.com/X-Square-Robot/wall-x",
python_requires=">=3.10",
packages=find_packages(exclude=("tests", "tests.*")),
scripts=PUBLIC_SCRIPTS,
ext_modules=build_ext_modules(),
cmdclass={"build_ext": BuildExtension.with_options(use_ninja=True)},
classifiers=[
"Development Status :: 4 - Beta",
"Intended Audience :: Developers",
"Intended Audience :: Science/Research",
"License :: OSI Approved :: Apache Software License",
"Programming Language :: Python :: 3",
"License :: OSI Approved :: BSD License",
"Operating System :: Unix",
"Programming Language :: Python :: 3.10",
"Programming Language :: Python :: 3.11",
"Topic :: Scientific/Engineering :: Artificial Intelligence",
],
packages=find_packages(),
ext_modules=ext_modules,
cmdclass={"build_ext": BuildExtension},
)