apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 00:17:25 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity

[BugFix] Fix _disable_sequence_parallel_moe_if_needed (#5740)

Browse Source
This commit is contained in:
Yuanle Liu
2025-12-25 12:02:22 +08:00
committed by GitHub
parent e48e306134
commit 75b3180280
1 changed files with 10 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastdeploy/config.py
+10 -9
View File
@@ -1779,15 +1779,6 @@ class FDConfig:
if not current_platform.is_cuda() and not current_platform.is_maca():
self.graph_opt_config.use_cudagraph = False
logger.info("CUDAGraph currently only support on GPU!")
if self.parallel_config.use_sequence_parallel_moe and self.graph_opt_config.use_cudagraph:
if self.scheduler_config.max_num_seqs < self.parallel_config.tensor_parallel_size:
self.parallel_config.use_sequence_parallel_moe = False
logger.info(
"Warning: sequence parallel moe do not support max_num_seqs < tensor_parallel_size when cudagraph enabled. We set use_sequence_parallel_moe to False."
)
else:
# It will hang when real batch_size < tp_size
self.graph_opt_config.filter_capture_size(tp_size=self.parallel_config.tensor_parallel_size)
# adjust speculative config
if self.speculative_config is not None and self.speculative_config.method == "mtp":
@@ -1806,6 +1797,16 @@ class FDConfig:
else:
raise NotImplementedError
if self.parallel_config.use_sequence_parallel_moe and self.graph_opt_config.use_cudagraph:
if self.scheduler_config.max_num_seqs < self.parallel_config.tensor_parallel_size:
self.parallel_config.use_sequence_parallel_moe = False
logger.info(
"Warning: sequence parallel moe do not support max_num_seqs < tensor_parallel_size when cudagraph enabled. We set use_sequence_parallel_moe to False."
)
else:
# It will hang when real batch_size < tp_size
self.graph_opt_config.filter_capture_size(tp_size=self.parallel_config.tensor_parallel_size)
self.postprocess_devices_and_ports()
def postprocess_devices_and_ports(self):