Skip to content

GitLab

"git@developer.sourcefind.cn:OpenDAS/colossalai.git" did not exist on "19d153057efebf65259fdc21f582ff2e55dbe2ec"
Unverified Commit bbbcac26 authored by Tong Li's avatar Tong Li Committed by GitHub
Browse files

fix format (#4815)

parent fb46d05c
Hide whitespace changes
Inline Side-by-side
Showing with 26 additions and 22 deletions
op_builder/gptq.py
View file @ bbbcac26
import os
import torch
import re import re
import torch
from .builder import Builder from .builder import Builder
from .utils import append_nvcc_threads, get_cuda_cc_flag from .utils import append_nvcc_threads
class GPTQBuilder(Builder):
class GPTQBuilder(Builder):
NAME = "cu_gptq" NAME = "cu_gptq"
PREBUILT_IMPORT_PATH = "colossalai._C.cu_gptq" PREBUILT_IMPORT_PATH = "colossalai._C.cu_gptq"
def __init__(self): def __init__(self):
super().__init__(name=GPTQBuilder.NAME, super().__init__(name=GPTQBuilder.NAME, prebuilt_import_path=GPTQBuilder.PREBUILT_IMPORT_PATH)
prebuilt_import_path=GPTQBuilder.PREBUILT_IMPORT_PATH)
def include_dirs(self): def include_dirs(self):
ret = [self.csrc_abs_path("gptq"), self.get_cuda_home_include()] ret = [self.csrc_abs_path("gptq"), self.get_cuda_home_include()]
...@@ -21,32 +19,38 @@ class GPTQBuilder(Builder): ...@@ -21,32 +19,38 @@ class GPTQBuilder(Builder):
def sources_files(self): def sources_files(self):
ret = [ ret = [
self.csrc_abs_path(fname) for fname in [ self.csrc_abs_path(fname)
'gptq/linear_gptq.cpp', for fname in [
'gptq/column_remap.cu', "gptq/linear_gptq.cpp",
'gptq/cuda_buffers.cu', "gptq/column_remap.cu",
'gptq/q4_matmul.cu', "gptq/cuda_buffers.cu",
'gptq/q4_matrix.cu' "gptq/q4_matmul.cu",
"gptq/q4_matrix.cu",
] ]
] ]
return ret return ret
def cxx_flags(self): def cxx_flags(self):
return ['-O3'] + self.version_dependent_macros return ["-O3"] + self.version_dependent_macros
def nvcc_flags(self): def nvcc_flags(self):
extra_cuda_flags = ['-v', extra_cuda_flags = [
'-std=c++14', '-U__CUDA_NO_HALF_OPERATORS__', '-U__CUDA_NO_HALF_CONVERSIONS__', "-v",
'-U__CUDA_NO_HALF2_OPERATORS__', '-DTHRUST_IGNORE_CUB_VERSION_CHECK', "-lcublas", "-std=c++17" "-std=c++14",
"-U__CUDA_NO_HALF_OPERATORS__",
"-U__CUDA_NO_HALF_CONVERSIONS__",
"-U__CUDA_NO_HALF2_OPERATORS__",
"-DTHRUST_IGNORE_CUB_VERSION_CHECK",
"-lcublas",
"-std=c++17",
] ]
for arch in torch.cuda.get_arch_list(): for arch in torch.cuda.get_arch_list():
res = re.search(r'sm_(\d+)', arch) res = re.search(r"sm_(\d+)", arch)
if res: if res:
arch_cap = res[1] arch_cap = res[1]
if int(arch_cap) >= 80: if int(arch_cap) >= 80:
extra_cuda_flags.extend(['-gencode', f'arch=compute_{arch_cap},code={arch}']) extra_cuda_flags.extend(["-gencode", f"arch=compute_{arch_cap},code={arch}"])
ret = ['-O3', '--use_fast_math'] + self.version_dependent_macros + extra_cuda_flags ret = ["-O3", "--use_fast_math"] + self.version_dependent_macros + extra_cuda_flags
return append_nvcc_threads(ret) return append_nvcc_threads(ret)
\ No newline at end of file
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment