ros/LLaMA-Factory
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

support val set in streaming mode

Browse Source 
Former-commit-id: faed15b58ed00b1e09bb091e7eee48f5ef7c508b
This commit is contained in:
hiyouga
2023-08-09 23:00:26 +08:00
parent 972bfa700a
commit 467d571206
10 changed files with 58 additions and 50 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
src/llmtuner/tuner/ppo/workflow.py
View File

@@ -41,14 +41,15 @@ def run_ppo(
max_grad_norm=training_args.max_grad_norm
)
optimizer = AdamW(filter(lambda p: p.requires_grad, model.parameters()), lr=ppo_config.learning_rate)
optimizer = AdamW(filter(lambda p: p.requires_grad, model.parameters()), lr=training_args.learning_rate)
total_train_batch_size = \
training_args.per_device_train_batch_size * training_args.gradient_accumulation_steps * training_args.world_size
num_training_steps = training_args.num_train_epochs * math.ceil(len(dataset) / total_train_batch_size)
lr_scheduler = get_scheduler(
training_args.lr_scheduler_type,
optimizer=optimizer,
num_warmup_steps=training_args.warmup_steps,
num_training_steps=(training_args.num_train_epochs * math.ceil(len(dataset) / total_train_batch_size))
num_warmup_steps=training_args.get_warmup_steps(num_training_steps),
num_training_steps=num_training_steps
)
# Initialize our Trainer