apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 08:37:06 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

【FIX】Change the name of sparse attn from moba to plas (#3845)

Browse Source 
* 更新文档

* 更新文档

* 更新文档

* 更新文档

* 修改moba为plas

* code style

* update ci

* code style

* update ci
This commit is contained in:
yangjianfengo1
2025-09-09 10:56:50 +08:00
committed by GitHub
parent de34222842
commit e31c8f7336
13 changed files with 150 additions and 150 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
fastdeploy/engine/args_utils.py
View File

@@ -28,9 +28,9 @@ from fastdeploy.config import (
FDConfig,
GraphOptimizationConfig,
LoadConfig,
MobaAttentionConfig,
ModelConfig,
ParallelConfig,
PlasAttentionConfig,
SpeculativeConfig,
TaskOption,
)
@@ -342,9 +342,9 @@ class EngineArgs:
"""
Configuration for graph optimization backend execution.
"""
moba_attention_config: Optional[Dict[str, Any]] = None
plas_attention_config: Optional[Dict[str, Any]] = None
"""
Configuration for moba attention.
Configuration for plas attention.
"""
enable_logprob: bool = False
@@ -559,9 +559,9 @@ class EngineArgs:
help="",
)
model_group.add_argument(
"--moba-attention-config",
"--plas-attention-config",
type=json.loads,
default=EngineArgs.moba_attention_config,
default=EngineArgs.plas_attention_config,
help="",
)
model_group.add_argument(
@@ -959,17 +959,17 @@ class EngineArgs:
graph_optimization_args[k] = v
return GraphOptimizationConfig(graph_optimization_args)
def create_moba_attention_config(self) -> MobaAttentionConfig:
def create_plas_attention_config(self) -> PlasAttentionConfig:
"""
Create and retuan a MobaAttentionConfig object based on the current settings.
Create and retuan a PlasAttentionConfig object based on the current settings.
"""
attention_args = asdict(self)
if self.moba_attention_config is not None:
for k, v in self.moba_attention_config.items():
if self.plas_attention_config is not None:
for k, v in self.plas_attention_config.items():
attention_args[k] = v
return MobaAttentionConfig(attention_args)
return PlasAttentionConfig(attention_args)
else:
return MobaAttentionConfig(None)
return PlasAttentionConfig(None)
def create_early_stop_config(self) -> EarlyStopConfig:
"""
@@ -1025,7 +1025,7 @@ class EngineArgs:
scheduler_cfg = self.create_scheduler_config()
graph_opt_cfg = self.create_graph_optimization_config()
graph_opt_cfg.update_use_cudagraph(self.use_cudagraph)
moba_attention_config = self.create_moba_attention_config()
plas_attention_config = self.create_plas_attention_config()
early_stop_cfg = self.create_early_stop_config()
early_stop_cfg.update_enable_early_stop(self.enable_early_stop)
@@ -1063,7 +1063,7 @@ class EngineArgs:
max_long_partial_prefills=self.max_long_partial_prefills,
long_prefill_token_threshold=self.long_prefill_token_threshold,
graph_opt_config=graph_opt_cfg,
moba_attention_config=moba_attention_config,
plas_attention_config=plas_attention_config,
guided_decoding_backend=self.guided_decoding_backend,
disable_any_whitespace=self.guided_decoding_disable_any_whitespace,
early_stop_config=early_stop_cfg,