[feature] added translation, rallio instruct tuning dataset, prosocial for safety, new summary dataset

This commit is contained in:
theblackcat102
2023-01-20 03:02:07 +00:00
parent da6a3b687e
commit 74cb9aaa5a
11 changed files with 399 additions and 18 deletions
@@ -2,7 +2,7 @@ model_name: microsoft/deberta-v3-base
learning_rate: 1e-5
scheduler: cosine
gradient_checkpointing: false
gradient_accumulation_steps: 32
gradient_accumulation_steps: 16
per_device_train_batch_size: 2
warmup_steps: 600
eval_steps: 200