423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2026-07-31 13:16:12 +08:00
Code Issues Packages Projects Releases Wiki Activity

[v1] Add FlashAttention selection and implement normal / padding-free / dynamic batching (#10469)

Browse Source
This commit is contained in:
jiaqiw09
2026-05-21 17:14:19 +08:00
committed by GitHub
parent 7e20db5735
commit bdcb92d035
23 changed files with 507 additions and 105 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/llamafactory/v1/config/training_args.py
View File

@@ -116,3 +116,9 @@ class TrainingArguments:
self.dist_config = get_plugin_config(self.dist_config)
self.optim_config = get_plugin_config(self.optim_config)
self.lr_scheduler_config = get_plugin_config(self.lr_scheduler_config)
if str(self.batching_strategy) == str(BatchingStrategy.DYNAMIC_BATCHING):
if self.max_steps is None or self.max_steps <= 0:
raise ValueError("`dynamic_batching` requires `max_steps` because it is step-driven.")
if self.save_epochs is not None:
raise ValueError("`save_epochs` is not supported with `dynamic_batching`; use `save_steps` instead.")