EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Fix nodeepgemm init (#6417)

Browse Source
This commit is contained in:
lukec
2025-05-19 15:44:03 +08:00
committed by GitHub
parent 4f39bcf7ab
commit 844e2f227a
1 changed files with 5 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
python/sglang/srt/layers/quantization/deep_gemm.py
View File

@@ -11,8 +11,10 @@ from tqdm.contrib.concurrent import thread_map
from sglang.srt.server_args import ServerArgs
from sglang.srt.utils import get_bool_env_var, get_device_sm, get_int_env_var, is_cuda
logger = logging.getLogger(__name__)
_ENABLE_JIT_DEEPGEMM = False
if is_cuda():
try:
import deep_gemm
from deep_gemm import get_num_sms
from deep_gemm.jit.compiler import get_nvcc_compiler
@@ -24,14 +26,14 @@ if is_cuda():
if sm_version == 90:
if get_bool_env_var("SGL_ENABLE_JIT_DEEPGEMM", default="true"):
_ENABLE_JIT_DEEPGEMM = True
except ImportError:
logger.warning("Failed to import deepgemm, disable _ENABLE_JIT_DEEPGEMM.")
def get_enable_jit_deepgemm():
return _ENABLE_JIT_DEEPGEMM
logger = logging.getLogger(__name__)
_BUILTIN_M_LIST = list(range(1, 1024 * 16 + 1))
_ENABLE_JIT_DEEPGEMM_PRECOMPILE = get_bool_env_var(
"SGL_JIT_DEEPGEMM_PRECOMPILE", "true"