Add support for parallel quantization on Mac
Browse files- quantization.py +7 -2
quantization.py
CHANGED
@@ -7,6 +7,7 @@ import bz2
|
|
7 |
import torch
|
8 |
import base64
|
9 |
import ctypes
|
|
|
10 |
from transformers.utils import logging
|
11 |
|
12 |
from typing import List
|
@@ -142,8 +143,12 @@ class CPUKernel:
|
|
142 |
kernel_file = source_code[:-2] + ".so"
|
143 |
|
144 |
if compile_parallel_kernel:
|
145 |
-
|
146 |
-
|
|
|
|
|
|
|
|
|
147 |
print("Compiling", compile_command)
|
148 |
exit_state = os.system(compile_command)
|
149 |
if not exit_state:
|
|
|
7 |
import torch
|
8 |
import base64
|
9 |
import ctypes
|
10 |
+
import sys
|
11 |
from transformers.utils import logging
|
12 |
|
13 |
from typing import List
|
|
|
143 |
kernel_file = source_code[:-2] + ".so"
|
144 |
|
145 |
if compile_parallel_kernel:
|
146 |
+
if sys.platform != 'darwin':
|
147 |
+
compile_command = "gcc -O3 -fPIC -pthread -fopenmp -std=c99 {} -shared -o {}".format(
|
148 |
+
source_code, kernel_file)
|
149 |
+
else:
|
150 |
+
compile_command = "clang -O3 -fPIC -pthread -Xclang -fopenmp -lomp -std=c99 {} -shared -o {}".format(
|
151 |
+
source_code, kernel_file)
|
152 |
print("Compiling", compile_command)
|
153 |
exit_state = os.system(compile_command)
|
154 |
if not exit_state:
|