wandb
test_rm_labeling:
is_reward_model: true
pooling: last
sort_by_length: false
use_custom_sampler: true
model_name: microsoft/deberta-v3-base
learning_rate: 3e-5
residual_dropout: 0.0
weight_decay: 0.0
max_length: 2048
use_flash_attention: true
gradient_checkpointing: true
warmup_steps: 50
dtype: float16
gradient_accumulation_steps: 5
per_device_train_batch_size: 4
per_device_eval_batch_size: 4
num_train_epochs: 3
eval_steps: 251
save_steps: 500
loss_fn: HybridRMLoss
datasets:
- oasst_export_w_label:
lang: "bg,ca,cs,da,de,en,es,fr,hr,hu,it,nl,pl,pt,ro,ru,sl,sr,sv,uk,zh,ja,th,vi"
input_file_path: 2023-04-12_oasst_release_ready_synth.jsonl.gz
input_label_path: 2023-04-12_oasst_all.messages.jsonl.gz
val_split: 0.1
