Mirrors/modelscope
1
0
Fork 0
mirror of https://github.com/modelscope/modelscope.git synced 2025-12-25 04:29:22 +01:00
Code Issues Packages Projects Releases Wiki Activity

[to #47630265] fix modelscope gpt moe finetune checkpoint path error

Browse Source 
Link: https://code.alibaba-inc.com/Ali-MaaS/MaaS-lib/codereview/11716118
This commit is contained in:
jerry.lp
2023-02-20 14:59:50 +08:00
committed by wenmeng.zwm
parent 1ae4782c83
commit ada2fcf48e
2 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
modelscope/models/nlp/gpt_moe/checkpointing.py
View File

@@ -38,7 +38,7 @@ def get_checkpoint_names(checkpoints_path,
f'mp_rank_{tensor_rank:02d}')
if num_experts[0] > 0:
model_name = common_path + '_model_states.pt'
model_name = os.path.join(common_path, 'model_rng.pt')
optim_name = os.path.join(
checkpoints_path, path_load_tag,
f'expp_rank_{expp_rank}_mp_rank_{tensor_rank:02d}_optim_states.pt')

3
modelscope/models/nlp/gpt_moe/distributed_gpt_moe.py
View File

@@ -1151,7 +1151,8 @@ class DistributedGPTMoE(TorchModel):
attention_mask=None,
position_ids=None,
labels=None,
prompt_length=None):
prompt_length=None,
is_pair=(False, )):
outputs, *other_losses = self.dist_model(
tokens,