Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
Menu
Open sidebar
change
sglang
Commits
afa0341e
Unverified
Commit
afa0341e
authored
Dec 29, 2024
by
HandH1998
Committed by
GitHub
Dec 29, 2024
Browse files
Update Triton configs for block fp8 kernels (#2641)
parent
30828e71
Changes
43
Hide whitespace changes
Inline
Side-by-side
Showing
3 changed files
with
187 additions
and
42 deletions
+187
-42
python/sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
...me=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
+39
-39
python/sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
...me=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
+146
-0
python/sglang/srt/layers/quantization/fp8_kernel.py
python/sglang/srt/layers/quantization/fp8_kernel.py
+2
-3
No files found.
python/sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
View file @
afa0341e
{
"1"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
3
},
"2"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
4
},
"4"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
4
},
"8"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
4
},
"16"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
6
,
"GROUP_SIZE_M"
:
6
4
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
4
},
"24"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
64
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
4
},
"32"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
4
},
"48"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
4
},
"64"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
4
},
"96"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
4
},
"128"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
64
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
4
},
"256"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
6
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
3
},
"512"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
64
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
3
},
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
6
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
3
},
"1536"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
3
},
"2048"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
64
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
3
},
"3072"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
64
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
3
},
"4096"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
"num_stages"
:
3
}
}
python/sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json
0 → 100644
View file @
afa0341e
{
"1"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
3
},
"2"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
3
},
"4"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
3
},
"8"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
3
},
"16"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
3
},
"24"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
3
},
"32"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
3
},
"48"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
3
},
"64"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
4
},
"96"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
5
},
"128"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
5
},
"256"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
64
,
"num_warps"
:
4
,
"num_stages"
:
4
},
"512"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
3
},
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
3
},
"1536"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
},
"2048"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
},
"3072"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
},
"4096"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
}
}
python/sglang/srt/layers/quantization/fp8_kernel.py
View file @
afa0341e
...
...
@@ -309,15 +309,14 @@ def w8a8_block_fp8_matmul(
config
=
configs
[
min
(
configs
.
keys
(),
key
=
lambda
x
:
abs
(
x
-
M
))]
else
:
# Default config
# Block-wise quant: BLOCK_SIZE_N must be divisable by block_size[0]
# BLOCK_SIZE_K must be divisable by block_size[1]
# Block-wise quant: BLOCK_SIZE_K must be divisable by block_size[1]
config
=
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
block_size
[
0
],
"BLOCK_SIZE_K"
:
block_size
[
1
],
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_stages"
:
3
,
}
def
grid
(
META
):
...
...
Prev
1
2
3
Next
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment
