mirror of
https://github.com/hiyouga/LLaMA-Factory.git
synced 2025-08-03 04:02:49 +08:00
Merge pull request #1553 from hannlp/hans
Change the default argument settings for PPO training Former-commit-id: 48211e3799a16de946360930d3d92f5a40e9d12d
This commit is contained in:
commit
d72f123851
@ -313,6 +313,8 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
|
||||
--per_device_train_batch_size 2 \
|
||||
--gradient_accumulation_steps 4 \
|
||||
--lr_scheduler_type cosine \
|
||||
--top_k 0 \
|
||||
--top_p 0.9 \
|
||||
--logging_steps 10 \
|
||||
--save_steps 1000 \
|
||||
--learning_rate 1e-5 \
|
||||
|
@ -313,6 +313,8 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
|
||||
--per_device_train_batch_size 2 \
|
||||
--gradient_accumulation_steps 4 \
|
||||
--lr_scheduler_type cosine \
|
||||
--top_k 0 \
|
||||
--top_p 0.9 \
|
||||
--logging_steps 10 \
|
||||
--save_steps 1000 \
|
||||
--learning_rate 1e-5 \
|
||||
|
@ -74,6 +74,10 @@ class RLHFArguments:
|
||||
default=None,
|
||||
metadata={"help": "Log with either 'wandb' or 'tensorboard' in PPO training."}
|
||||
)
|
||||
ppo_epochs: Optional[int] = field(
|
||||
default=4,
|
||||
metadata={"help": "Number of optimisation epochs per batch of samples"},
|
||||
)
|
||||
ppo_score_norm: Optional[bool] = field(
|
||||
default=False,
|
||||
metadata={"help": "Use score normalization in PPO training."}
|
||||
|
@ -45,7 +45,7 @@ def run_ppo(
|
||||
mini_batch_size=training_args.per_device_train_batch_size,
|
||||
batch_size=training_args.per_device_train_batch_size * training_args.gradient_accumulation_steps,
|
||||
gradient_accumulation_steps=training_args.gradient_accumulation_steps,
|
||||
ppo_epochs=1,
|
||||
ppo_epochs=finetuning_args.ppo_epochs,
|
||||
max_grad_norm=training_args.max_grad_norm,
|
||||
seed=training_args.seed,
|
||||
optimize_device_cache=True,
|
||||
|
Loading…
x
Reference in New Issue
Block a user