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
351a72d4
"git@developer.sourcefind.cn:OpenDAS/torchaudio.git" did not exist on "47eb1e6a8fabda4b784367be9babfd30345ee590"
Unverified
Commit
351a72d4
authored
Jan 27, 2025
by
yigex
Committed by
GitHub
Jan 27, 2025
Browse files
add dsv3 mi300 triton config for block scale (#3146)
parent
514f37c3
Changes
11
Hide whitespace changes
Inline
Side-by-side
Showing
11 changed files
with
1683 additions
and
11 deletions
+1683
-11
benchmark/kernels/fused_moe_triton/tuning_fused_moe_triton.py
...hmark/kernels/fused_moe_triton/tuning_fused_moe_triton.py
+43
-11
python/sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=256,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
...nstinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
+164
-0
python/sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
...nstinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
+164
-0
python/sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
...nstinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
+164
-0
python/sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
...nstinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
+164
-0
python/sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
...nstinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
+164
-0
python/sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
...nstinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
+164
-0
python/sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
...nstinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
+164
-0
python/sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
...nstinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
+164
-0
python/sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
...nstinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
+164
-0
python/sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
...nstinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
+164
-0
No files found.
benchmark/kernels/fused_moe_triton/tuning_fused_moe_triton.py
View file @
351a72d4
...
@@ -18,6 +18,9 @@ from sglang.srt.layers.moe.fused_moe_triton.fused_moe import (
...
@@ -18,6 +18,9 @@ from sglang.srt.layers.moe.fused_moe_triton.fused_moe import (
get_default_config
,
get_default_config
,
get_moe_configs
,
get_moe_configs
,
)
)
from
sglang.srt.utils
import
is_hip
_is_hip_
=
is_hip
()
class
BenchmarkConfig
(
TypedDict
):
class
BenchmarkConfig
(
TypedDict
):
...
@@ -102,8 +105,8 @@ def benchmark_config(
...
@@ -102,8 +105,8 @@ def benchmark_config(
(
num_experts
,
n_tiles_w2
,
k_tiles_w2
),
dtype
=
torch
.
float32
(
num_experts
,
n_tiles_w2
,
k_tiles_w2
),
dtype
=
torch
.
float32
)
)
w1
=
w1
.
to
(
torch
.
float8_e4m3fn
)
w1
=
w1
.
to
(
torch
.
float8_e4m3fnuz
if
_is_hip_
else
torch
.
float8_e4m3fn
)
w2
=
w2
.
to
(
torch
.
float8_e4m3fn
)
w2
=
w2
.
to
(
torch
.
float8_e4m3fnuz
if
_is_hip_
else
torch
.
float8_e4m3fn
)
input_gating
=
torch
.
empty
(
num_tokens
,
num_experts
,
dtype
=
torch
.
float32
)
input_gating
=
torch
.
empty
(
num_tokens
,
num_experts
,
dtype
=
torch
.
float32
)
...
@@ -165,17 +168,15 @@ def benchmark_config(
...
@@ -165,17 +168,15 @@ def benchmark_config(
return
avg
return
avg
def
get_configs_compute_bound
()
->
List
[
Dict
[
str
,
int
]]:
def
get_rocm_configs_compute_bound
()
->
List
[
Dict
[
str
,
int
]]:
# Reduced search space for faster tuning.
# TODO(woosuk): Increase the search space and use a performance model to
# prune the search space.
configs
:
List
[
BenchmarkConfig
]
=
[]
configs
:
List
[
BenchmarkConfig
]
=
[]
for
num_stages
in
[
2
,
3
,
4
,
5
]:
waves_per_eu_range
=
0
for
block_m
in
[
16
,
32
,
64
,
128
,
256
]:
for
num_stages
in
[
2
]:
for
block_k
in
[
64
,
128
,
256
]:
for
block_m
in
[
32
,
64
,
128
,
256
]:
for
block_n
in
[
32
,
64
,
128
,
256
]:
for
block_k
in
[
32
,
64
,
128
,
256
]:
for
block_n
in
[
16
,
32
,
64
,
128
,
256
]:
for
num_warps
in
[
4
,
8
]:
for
num_warps
in
[
4
,
8
]:
for
group_size
in
[
1
,
16
,
32
,
64
]:
for
group_size
in
[
1
,
4
,
8
,
16
,
32
]:
configs
.
append
(
configs
.
append
(
{
{
"BLOCK_SIZE_M"
:
block_m
,
"BLOCK_SIZE_M"
:
block_m
,
...
@@ -184,11 +185,39 @@ def get_configs_compute_bound() -> List[Dict[str, int]]:
...
@@ -184,11 +185,39 @@ def get_configs_compute_bound() -> List[Dict[str, int]]:
"GROUP_SIZE_M"
:
group_size
,
"GROUP_SIZE_M"
:
group_size
,
"num_warps"
:
num_warps
,
"num_warps"
:
num_warps
,
"num_stages"
:
num_stages
,
"num_stages"
:
num_stages
,
"waves_per_eu"
:
waves_per_eu_range
,
}
}
)
)
return
configs
return
configs
def
get_configs_compute_bound
()
->
List
[
Dict
[
str
,
int
]]:
# Reduced search space for faster tuning.
# TODO(woosuk): Increase the search space and use a performance model to
# prune the search space.
configs
:
List
[
BenchmarkConfig
]
=
[]
if
_is_hip_
:
configs
=
get_rocm_configs_compute_bound
()
else
:
for
num_stages
in
[
2
,
3
,
4
,
5
]:
for
block_m
in
[
16
,
32
,
64
,
128
,
256
]:
for
block_k
in
[
64
,
128
,
256
]:
for
block_n
in
[
32
,
64
,
128
,
256
]:
for
num_warps
in
[
4
,
8
]:
for
group_size
in
[
1
,
16
,
32
,
64
]:
configs
.
append
(
{
"BLOCK_SIZE_M"
:
block_m
,
"BLOCK_SIZE_N"
:
block_n
,
"BLOCK_SIZE_K"
:
block_k
,
"GROUP_SIZE_M"
:
group_size
,
"num_warps"
:
num_warps
,
"num_stages"
:
num_stages
,
}
)
return
configs
@
ray
.
remote
(
num_gpus
=
1
)
@
ray
.
remote
(
num_gpus
=
1
)
class
BenchmarkWorker
:
class
BenchmarkWorker
:
...
@@ -297,6 +326,9 @@ def sort_config(config: BenchmarkConfig) -> BenchmarkConfig:
...
@@ -297,6 +326,9 @@ def sort_config(config: BenchmarkConfig) -> BenchmarkConfig:
"GROUP_SIZE_M"
:
config
[
"GROUP_SIZE_M"
],
"GROUP_SIZE_M"
:
config
[
"GROUP_SIZE_M"
],
"num_warps"
:
config
[
"num_warps"
],
"num_warps"
:
config
[
"num_warps"
],
"num_stages"
:
config
[
"num_stages"
],
"num_stages"
:
config
[
"num_stages"
],
**
(
{
"waves_per_eu"
:
config
[
"waves_per_eu"
]}
if
"waves_per_eu"
in
config
else
{}
),
}
}
...
...
python/sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=256,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
0 → 100644
View file @
351a72d4
{
"1"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"8"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"16"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"24"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"32"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"48"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"64"
:
{
"BLOCK_SIZE_M"
:
256
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"96"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"128"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"256"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"512"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
8
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1536"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2048"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"3072"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4096"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
}
}
python/sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
0 → 100644
View file @
351a72d4
{
"1"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"8"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"16"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"24"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"32"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"48"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"64"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"96"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"128"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"256"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"512"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1536"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2048"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"3072"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4096"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
}
}
python/sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
0 → 100644
View file @
351a72d4
{
"1"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"8"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"16"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"24"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"32"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"48"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"64"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"96"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"128"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"256"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"512"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1536"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2048"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"3072"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4096"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
}
}
python/sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
0 → 100644
View file @
351a72d4
{
"1"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"8"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"16"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"24"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"32"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"48"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"64"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"96"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"128"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"256"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"512"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1536"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2048"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"3072"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4096"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
}
}
python/sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
0 → 100644
View file @
351a72d4
{
"1"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"8"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"16"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"24"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"32"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"48"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"64"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"96"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"128"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"256"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"512"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1536"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2048"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"3072"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4096"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
}
}
python/sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
0 → 100644
View file @
351a72d4
{
"1"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"8"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"16"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"24"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"32"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"48"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"64"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"96"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"128"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"256"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"512"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1536"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2048"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"3072"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4096"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
}
}
python/sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
0 → 100644
View file @
351a72d4
{
"1"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"8"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"16"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"24"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"32"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"48"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"64"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"96"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"128"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"256"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"512"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1536"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2048"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"3072"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4096"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
}
}
python/sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
0 → 100644
View file @
351a72d4
{
"1"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"8"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"16"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"24"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"32"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"48"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"64"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"96"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"128"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"256"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"512"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1536"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2048"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"3072"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4096"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
}
}
python/sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
0 → 100644
View file @
351a72d4
{
"1"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"8"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"16"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"24"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"32"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"48"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"64"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"96"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"128"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"256"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"512"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1536"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2048"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"3072"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4096"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
}
}
python/sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json
0 → 100644
View file @
351a72d4
{
"1"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"8"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"16"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"24"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"32"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"48"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"64"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"96"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"128"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"256"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"512"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"1536"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"2048"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"3072"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
},
"4096"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"waves_per_eu"
:
0
}
}
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
