biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Perf] Add Triton config for DeepSeek V3 FP8 EP32 H200 (#23504)

Browse Source 
Signed-off-by: Ming Yang <minos.future@gmail.com>
This commit is contained in:
Ming Yang
2025-08-24 18:06:35 -07:00
committed by GitHub
parent 47455c424f
commit 504d914314
4 changed files with 184 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
vllm/model_executor/layers/quantization/utils/configs/N=2112,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json Normal file
View File

@@ -0,0 +1,26 @@
{
"2048": {
"BLOCK_SIZE_M": 256,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32,
"num_warps": 8,
"num_stages": 3
},
"3072": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 4,
"num_stages": 3
},
"4096": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32,
"num_warps": 4,
"num_stages": 4
}
}

3
vllm/model_executor/layers/quantization/utils/configs/README.md Normal file
View File

@@ -0,0 +1,3 @@
# Quantization Kernel Config
Use scripts under `benchmarks/kernels/` to generate these config files.