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

Allow disabling flashinfer sampling kernel (#778)

Browse Source
This commit is contained in:
Lianmin Zheng
2024-07-27 20:18:56 -07:00
committed by GitHub
parent 30db99b3d9
commit 752e643007
6 changed files with 41 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
python/sglang/srt/server_args.py
View File

@@ -52,13 +52,14 @@ class ServerArgs:
# Optimization/debug options
disable_flashinfer: bool = False
disable_flashinfer_sampling: bool = False
disable_radix_cache: bool = False
disable_regex_jump_forward: bool = False
disable_cuda_graph: bool = False
disable_disk_cache: bool = False
enable_torch_compile: bool = False
attention_reduce_in_fp32: bool = False
enable_p2p_check: bool = False
attention_reduce_in_fp32: bool = False
efficient_weight_load: bool = False
# Distributed args
@@ -303,7 +304,12 @@ class ServerArgs:
parser.add_argument(
"--disable-flashinfer",
action="store_true",
help="Disable flashinfer inference kernels.",
help="Disable flashinfer attention kernels.",
)
parser.add_argument(
"--disable-flashinfer-sampling",
action="store_true",
help="Disable flashinfer sampling kernels.",
)
parser.add_argument(
"--disable-radix-cache",
@@ -330,17 +336,17 @@ class ServerArgs:
action="store_true",
help="Optimize the model with torch.compile, experimental feature.",
)
parser.add_argument(
"--enable-p2p-check",
action="store_true",
help="Enable P2P check for GPU access, otherwise the p2p access is allowed by default.",
)
parser.add_argument(
"--attention-reduce-in-fp32",
action="store_true",
help="Cast the intermidiate attention results to fp32 to avoid possible crashes related to fp16."
"This only affects Triton attention kernels",
)
parser.add_argument(
"--enable-p2p-check",
action="store_true",
help="Enable P2P check for GPU access, otherwise the p2p access is allowed by default.",
)
parser.add_argument(
"--efficient-weight-load",
action="store_true",