Training log: https://wandb.ai/toanbku/reward-model/runs/2ekuy6lg/overview

```
deepspeed --include=localhost:0 --master_port 61000 trainer_rm.py --config \
defaults_rm oasst-rm-2.1-pythia-1.4b \
--cache_dir /home/ubuntu/OA/model/model_training/.cache \
--per_device_eval_batch_size 1 --per_device_train_batch_size 1 \
--wandb_entity toanbku --deepspeed
```


```
oasst-rm-2.1-pythia-1.4b:
  is_reward_model: true
  pooling: last
  datasets:
    - oasst_export:
      lang: "en"
      hf_dataset_name: toanbku/oa-df
      val_split: 0.1
  use_custom_sampler: true
  sort_by_length: false
  model_name: OpenAssistant/oasst-rm-2.1-pythia-1.4b-epoch-2.5
  learning_rate: 8e-6
  residual_dropout: 0.01
  weight_decay: 0.0
  dtype: float32
  max_length: 2048
  use_flash_attention: true
  warmup_steps: 2
  gradient_accumulation_steps: 2
  per_device_train_batch_size: 1
  per_device_eval_batch_size: 2
  num_train_epochs: 2
  eval_steps: 50
  save_steps: 100
  use_system_tag: false
  system_property_dropout: 0.5
  system_add_length: false
```