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
zhaoyu6
sglang
Commits
9e2f7252
"vscode:/vscode.git/clone" did not exist on "dc907c334cf5c9996321656d1a99b8b6379abb87"
Unverified
Commit
9e2f7252
authored
Sep 11, 2025
by
Yi Zhang
Committed by
GitHub
Sep 10, 2025
Browse files
add dual stream for qwen2_moe (#10252)
parent
21176b00
Changes
1
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
with
53 additions
and
11 deletions
+53
-11
python/sglang/srt/models/qwen2_moe.py
python/sglang/srt/models/qwen2_moe.py
+53
-11
No files found.
python/sglang/srt/models/qwen2_moe.py
View file @
9e2f7252
...
@@ -65,10 +65,12 @@ from sglang.srt.managers.schedule_batch import global_server_args_dict
...
@@ -65,10 +65,12 @@ from sglang.srt.managers.schedule_batch import global_server_args_dict
from
sglang.srt.model_executor.forward_batch_info
import
ForwardBatch
,
PPProxyTensors
from
sglang.srt.model_executor.forward_batch_info
import
ForwardBatch
,
PPProxyTensors
from
sglang.srt.model_loader.weight_utils
import
default_weight_loader
from
sglang.srt.model_loader.weight_utils
import
default_weight_loader
from
sglang.srt.two_batch_overlap
import
model_forward_maybe_tbo
from
sglang.srt.two_batch_overlap
import
model_forward_maybe_tbo
from
sglang.srt.utils
import
add_prefix
,
make_layers
from
sglang.srt.utils
import
add_prefix
,
is_cuda
,
make_layers
logger
=
logging
.
getLogger
(
__name__
)
logger
=
logging
.
getLogger
(
__name__
)
_is_cuda
=
is_cuda
()
class
Qwen2MoeMLP
(
nn
.
Module
):
class
Qwen2MoeMLP
(
nn
.
Module
):
def
__init__
(
def
__init__
(
...
@@ -122,11 +124,13 @@ class Qwen2MoeSparseMoeBlock(nn.Module):
...
@@ -122,11 +124,13 @@ class Qwen2MoeSparseMoeBlock(nn.Module):
layer_id
:
int
,
layer_id
:
int
,
config
:
PretrainedConfig
,
config
:
PretrainedConfig
,
quant_config
:
Optional
[
QuantizationConfig
]
=
None
,
quant_config
:
Optional
[
QuantizationConfig
]
=
None
,
alt_stream
:
Optional
[
torch
.
cuda
.
Stream
]
=
None
,
prefix
:
str
=
""
,
prefix
:
str
=
""
,
):
):
super
().
__init__
()
super
().
__init__
()
self
.
tp_size
=
get_tensor_model_parallel_world_size
()
self
.
tp_size
=
get_tensor_model_parallel_world_size
()
self
.
layer_id
=
layer_id
self
.
layer_id
=
layer_id
self
.
alt_stream
=
alt_stream
if
self
.
tp_size
>
config
.
num_experts
:
if
self
.
tp_size
>
config
.
num_experts
:
raise
ValueError
(
raise
ValueError
(
f
"Tensor parallel size
{
self
.
tp_size
}
is greater than "
f
"Tensor parallel size
{
self
.
tp_size
}
is greater than "
...
@@ -168,14 +172,7 @@ class Qwen2MoeSparseMoeBlock(nn.Module):
...
@@ -168,14 +172,7 @@ class Qwen2MoeSparseMoeBlock(nn.Module):
self
.
shared_expert
=
None
self
.
shared_expert
=
None
self
.
shared_expert_gate
=
torch
.
nn
.
Linear
(
config
.
hidden_size
,
1
,
bias
=
False
)
self
.
shared_expert_gate
=
torch
.
nn
.
Linear
(
config
.
hidden_size
,
1
,
bias
=
False
)
def
forward
(
def
_forward_shared_experts
(
self
,
hidden_states
:
torch
.
Tensor
):
self
,
hidden_states
:
torch
.
Tensor
,
forward_batch
:
Optional
[
ForwardBatch
]
=
None
,
use_reduce_scatter
:
bool
=
False
,
)
->
torch
.
Tensor
:
num_tokens
,
hidden_dim
=
hidden_states
.
shape
hidden_states
=
hidden_states
.
view
(
-
1
,
hidden_dim
)
shared_output
=
None
shared_output
=
None
if
self
.
shared_expert
is
not
None
:
if
self
.
shared_expert
is
not
None
:
shared_output
=
self
.
shared_expert
(
hidden_states
)
shared_output
=
self
.
shared_expert
(
hidden_states
)
...
@@ -183,11 +180,51 @@ class Qwen2MoeSparseMoeBlock(nn.Module):
...
@@ -183,11 +180,51 @@ class Qwen2MoeSparseMoeBlock(nn.Module):
shared_output
=
(
shared_output
=
(
F
.
sigmoid
(
self
.
shared_expert_gate
(
hidden_states
))
*
shared_output
F
.
sigmoid
(
self
.
shared_expert_gate
(
hidden_states
))
*
shared_output
)
)
return
shared_output
def
_forward_router_experts
(
self
,
hidden_states
:
torch
.
Tensor
):
# router_logits: (num_tokens, n_experts)
# router_logits: (num_tokens, n_experts)
router_logits
,
_
=
self
.
gate
(
hidden_states
)
router_logits
,
_
=
self
.
gate
(
hidden_states
)
topk_output
=
self
.
topk
(
hidden_states
,
router_logits
)
topk_output
=
self
.
topk
(
hidden_states
,
router_logits
)
final_hidden_states
=
self
.
experts
(
hidden_states
,
topk_output
)
return
self
.
experts
(
hidden_states
,
topk_output
)
def
forward_normal_dual_stream
(
self
,
hidden_states
:
torch
.
Tensor
,
)
->
torch
.
Tensor
:
current_stream
=
torch
.
cuda
.
current_stream
()
self
.
alt_stream
.
wait_stream
(
current_stream
)
shared_output
=
self
.
_forward_shared_experts
(
hidden_states
)
with
torch
.
cuda
.
stream
(
self
.
alt_stream
):
router_output
=
self
.
_forward_router_experts
(
hidden_states
)
current_stream
.
wait_stream
(
self
.
alt_stream
)
return
router_output
,
shared_output
def
forward
(
self
,
hidden_states
:
torch
.
Tensor
,
forward_batch
:
Optional
[
ForwardBatch
]
=
None
,
use_reduce_scatter
:
bool
=
False
,
)
->
torch
.
Tensor
:
num_tokens
,
hidden_dim
=
hidden_states
.
shape
hidden_states
=
hidden_states
.
view
(
-
1
,
hidden_dim
)
DUAL_STREAM_TOKEN_THRESHOLD
=
1024
if
(
self
.
alt_stream
is
not
None
and
hidden_states
.
shape
[
0
]
>
0
and
hidden_states
.
shape
[
0
]
<=
DUAL_STREAM_TOKEN_THRESHOLD
):
final_hidden_states
,
shared_output
=
self
.
forward_normal_dual_stream
(
hidden_states
)
else
:
shared_output
=
self
.
_forward_shared_experts
(
hidden_states
)
final_hidden_states
=
self
.
_forward_router_experts
(
hidden_states
)
if
shared_output
is
not
None
:
if
shared_output
is
not
None
:
final_hidden_states
=
final_hidden_states
+
shared_output
final_hidden_states
=
final_hidden_states
+
shared_output
if
self
.
tp_size
>
1
and
not
use_reduce_scatter
:
if
self
.
tp_size
>
1
and
not
use_reduce_scatter
:
...
@@ -346,6 +383,7 @@ class Qwen2MoeDecoderLayer(nn.Module):
...
@@ -346,6 +383,7 @@ class Qwen2MoeDecoderLayer(nn.Module):
layer_id
=
layer_id
,
layer_id
=
layer_id
,
config
=
config
,
config
=
config
,
quant_config
=
quant_config
,
quant_config
=
quant_config
,
alt_stream
=
alt_stream
,
prefix
=
add_prefix
(
"mlp"
,
prefix
),
prefix
=
add_prefix
(
"mlp"
,
prefix
),
)
)
else
:
else
:
...
@@ -528,8 +566,12 @@ class Qwen2MoeForCausalLM(nn.Module):
...
@@ -528,8 +566,12 @@ class Qwen2MoeForCausalLM(nn.Module):
self
.
pp_group
=
get_pp_group
()
self
.
pp_group
=
get_pp_group
()
self
.
config
=
config
self
.
config
=
config
self
.
quant_config
=
quant_config
self
.
quant_config
=
quant_config
alt_stream
=
torch
.
cuda
.
Stream
()
if
_is_cuda
else
None
self
.
model
=
Qwen2MoeModel
(
self
.
model
=
Qwen2MoeModel
(
config
,
quant_config
,
prefix
=
add_prefix
(
"model"
,
prefix
)
config
,
quant_config
,
prefix
=
add_prefix
(
"model"
,
prefix
),
alt_stream
=
alt_stream
,
)
)
self
.
lm_head
=
ParallelLMHead
(
self
.
lm_head
=
ParallelLMHead
(
config
.
vocab_size
,
config
.
vocab_size
,
...
...
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