Skip to content

Commit 945a640

Browse files
committed
fix issues
1 parent a894fc2 commit 945a640

2 files changed

Lines changed: 8 additions & 7 deletions

File tree

examples/ascend/megatron/train_sft_full.sh

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -21,7 +21,7 @@ megatron sft \
2121
--lr 1e-5 \
2222
--lr_warmup_fraction 0.05 \
2323
--min_lr 1e-6 \
24-
--output_dir megatron_output/Qwen3-32B \
24+
--output_dir megatron_output/Qwen3-4B \
2525
--eval_steps 500 \
2626
--save_steps 500 \
2727
--max_length 8192 \
@@ -30,4 +30,4 @@ megatron sft \
3030
--no_save_optim true \
3131
--no_save_rng true \
3232
--sequence_parallel true \
33-
--attention_backend flash
33+
--attention_backend flash

swift/megatron/model/model_config.py

Lines changed: 6 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -189,10 +189,10 @@ def _augment_mindspeed_defaults(self):
189189
if MegatronModelConfig._mindspeed_defaults_cache is None:
190190
defaults = {}
191191
try:
192-
from mindspeed.arguments import process_args
193-
from argparse import ArgumentParser
194192
import mindspeed.features_manager as mfm
195193
import sys
194+
from argparse import ArgumentParser
195+
from mindspeed.arguments import process_args
196196

197197
original_features = list(mfm.FEATURES_LIST)
198198
full_features = mfm.create_features_list()
@@ -202,20 +202,21 @@ def _augment_mindspeed_defaults(self):
202202
parser = ArgumentParser()
203203
process_args(parser)
204204
# Parse args from sys.argv
205-
args, _ = parser.parse_known_args()
205+
args, _ = parser.parse_known_args([])
206206
defaults = vars(args)
207207
finally:
208208
mfm.FEATURES_LIST.clear()
209209
mfm.FEATURES_LIST.extend(original_features)
210-
except Exception:
210+
except Exception as e:
211+
logger.warning(f'Failed to get MindSpeed defaults, which may cause issues on NPU: {e}')
211212
defaults = {}
212213
MegatronModelConfig._mindspeed_defaults_cache = defaults
213214

214215
for name, value in MegatronModelConfig._mindspeed_defaults_cache.items():
215216
if not hasattr(self, name):
216217
setattr(self, name, value)
217218
elif hasattr(self, name) and getattr(self, name) is None and value is not None:
218-
setattr(self, name, value)
219+
setattr(self, name, value)
219220

220221
def __post_init__(self):
221222
self._augment_mindspeed_defaults()

0 commit comments

Comments
 (0)