apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-03 07:46:50 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[MTP] support expert-parellel in mtp (#2835)

Browse Source
This commit is contained in:
freeliuzc
2025-07-14 14:28:50 +08:00
committed by GitHub
parent ece88596ed
commit 7f64d408a9
2 changed files with 5 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
fastdeploy/model_executor/pre_and_post_process.py
View File

@@ -182,7 +182,7 @@ def post_process_normal(sampler_output: SamplerOutput,
)
def post_process_specualate(model_output, skip_save_output: bool = False):
def post_process_specualate(model_output, save_each_rank: bool = False, skip_save_output: bool = False):
""""""
speculate_update_v3(
model_output.seq_lens_encoder,
@@ -204,7 +204,7 @@ def post_process_specualate(model_output, skip_save_output: bool = False):
model_output.accept_num,
model_output.not_need_stop,
model_output.mp_rank,
False,
save_each_rank,
)
speculate_clear_accept_nums(model_output.accept_num,
@@ -231,7 +231,7 @@ def post_process(sampler_output: SamplerOutput,
skip_save_output: bool = False) -> None:
""" Post-processing steps after completing a single token generation. """
if speculative_decoding:
post_process_specualate(model_output, skip_save_output)
post_process_specualate(model_output, save_each_rank, skip_save_output)
else:
post_process_normal(sampler_output, model_output, save_each_rank,
skip_save_output)