Skip to content

update triton 3.2.0 h200 fused moe triton config and add warning about triton fused_moe_kernel performance degradation due to different Triton versions. #5740

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Apr 25, 2025
Merged

update triton 3.2.0 h200 fused moe triton config and add warning about triton fused_moe_kernel performance degradation due to different Triton versions. #5740

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
d977734
update triton 3.2.0 E=264,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a…
BBuf Apr 25, 2025
fa532ca
logging
BBuf Apr 25, 2025
4183cf8
Merge branch 'main' into update_triton_3.2.0_fused_moe_config
merrymercy Apr 25, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
82 changes: 41 additions & 41 deletions ...on/configs/E=264,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,102 +1,102 @@
{
"1": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16,
"GROUP_SIZE_M": 1,
"num_warps": 4,
"num_stages": 4
},
"2": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"GROUP_SIZE_M": 16,
"num_warps": 4,
"num_stages": 3
"num_stages": 4
},
"4": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"GROUP_SIZE_M": 16,
"num_warps": 4,
"num_stages": 4
},
"8": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32,
"num_warps": 4,
"num_stages": 3
"num_stages": 4
},
"16": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16,
"GROUP_SIZE_M": 1,
"num_warps": 4,
"num_stages": 3
},
"24": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16,
"GROUP_SIZE_M": 1,
"num_warps": 4,
"num_stages": 3
"num_stages": 4
},
"32": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32,
"GROUP_SIZE_M": 16,
"num_warps": 4,
"num_stages": 3
"num_stages": 5
},
"48": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32,
"GROUP_SIZE_M": 64,
"num_warps": 4,
"num_stages": 3
"num_stages": 4
},
"64": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 64,
"GROUP_SIZE_M": 32,
"num_warps": 4,
"num_stages": 3
},
"96": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 64,
"GROUP_SIZE_M": 32,
"num_warps": 4,
"num_stages": 3
},
"128": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16,
"GROUP_SIZE_M": 64,
"num_warps": 4,
"num_stages": 3
},
"256": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16,
"GROUP_SIZE_M": 64,
"num_warps": 4,
"num_stages": 3
},
"512": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16,
Expand All @@ -107,40 +107,40 @@
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32,
"GROUP_SIZE_M": 16,
"num_warps": 4,
"num_stages": 3
"num_stages": 4
},
"1536": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32,
"num_warps": 4,
"num_stages": 3
"num_stages": 4
},
"2048": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16,
"GROUP_SIZE_M": 32,
"num_warps": 4,
"num_stages": 3
"num_stages": 4
},
"3072": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32,
"GROUP_SIZE_M": 16,
"num_warps": 4,
"num_stages": 3
},
"4096": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 64,
"GROUP_SIZE_M": 16,
"num_warps": 4,
"num_stages": 3
"num_stages": 4
}
}
5 changes: 4 additions & 1 deletion python/sglang/srt/layers/moe/fused_moe_triton/fused_moe.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -940,7 +940,10 @@ def get_moe_configs(
)
if os.path.exists(config_file_path):
with open(config_file_path) as f:
logger.info("Using configuration from %s for MoE layer.", config_file_path)
logger.info(
"Using configuration from %s for MoE layer. Please note that due to the large number of configs under fused_moe_triton/configs potentially not being tuned with the corresponding Triton version in your current environment, using the current configs may result in performance degradation. To achieve best performance, you can consider re-tuning the Triton fused MOE kernel in your current environment. For the tuning method, please refer to: https://github.com/sgl-project/sglang/blob/main/benchmark/kernels/fused_moe_triton/tuning_fused_moe_triton.py. ",
config_file_path,
)
# If a configuration has been found, return it
return {int(key): val for key, val in json.load(f).items()}

Expand Down
Loading