metadata

license: other
base_model: stabilityai/stablelm-2-1_6b
tags:
  - alignment-handbook
  - trl
  - orpo
  - generated_from_trainer
  - trl
  - orpo
  - generated_from_trainer
datasets:
  - alvarobartt/dpo-mix-7k-simplified
model-index:
  - name: stablelm-2-1_6b-orpo-full-v1
    results: []

stablelm-2-1_6b-orpo-full-v1

This model is a fine-tuned version of stabilityai/stablelm-2-1_6b on the alvarobartt/dpo-mix-7k-simplified dataset. It achieves the following results on the evaluation set:

Loss: 1.3790
Rewards/chosen: -0.1279
Rewards/rejected: -0.1387
Rewards/accuracies: 0.5625
Rewards/margins: 0.0114
Logps/rejected: -1.3906
Logps/chosen: -1.2734
Logits/rejected: -2.4688
Logits/chosen: -2.4844
Nll Loss: 1.3359
Log Odds Ratio: -0.6799
Log Odds Chosen: 0.1751

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-07
train_batch_size: 8
eval_batch_size: 8
seed: 42
distributed_type: multi-GPU
num_devices: 3
gradient_accumulation_steps: 2
total_train_batch_size: 48
total_eval_batch_size: 24
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.1
num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen	Nll Loss	Log Odds Ratio	Log Odds Chosen
1.4572	0.71	100	1.4374	-0.1406	-0.1514	0.5469	0.0109	-1.5078	-1.4062	-2.3594	-2.375	1.3906	-0.6874	0.1528
1.3846	1.42	200	1.3829	-0.1279	-0.1396	0.5625	0.0115	-1.3984	-1.2812	-2.4688	-2.4844	1.3359	-0.6793	0.1746

Framework versions

Transformers 4.39.0.dev0
Pytorch 2.2.1
Datasets 2.18.0
Tokenizers 0.15.2