distilgpt2-dpo

This model is a fine-tuned version of mNLP-project/distilgpt2-finetuned on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.1792
• Rewards/chosen: 6.1190
• Rewards/rejected: 5.0796
• Rewards/accuracies: 0.6061
• Rewards/margins: 1.0394
• Logps/rejected: -703.7405
• Logps/chosen: -844.3468
• Logits/rejected: -11.5397
• Logits/chosen: -8.7315

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-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
1.2146	1.0	1337	1.1792	6.1190	5.0796	0.6061	1.0394	-703.7405	-844.3468	-11.5397	-8.7315
0.8026	2.0	2674	1.1980	6.3028	5.0877	0.6142	1.2151	-703.6594	-842.5087	-9.1682	-7.1950
0.3605	3.0	4011	1.3136	5.3889	4.2456	0.5960	1.1433	-712.0801	-851.6475	-8.0251	-5.8074
0.117	4.0	5348	1.4214	6.6526	5.0410	0.6134	1.6116	-704.1267	-839.0112	-6.1296	-4.2746
0.0663	5.0	6685	1.5485	5.0321	3.6157	0.5947	1.4164	-718.3795	-855.2162	-2.6173	-0.7400
0.0078	6.0	8022	1.7565	5.1090	3.1954	0.6059	1.9136	-722.5821	-854.4468	-4.4487	-2.6082
0.0095	7.0	9359	1.7638	4.7802	2.8888	0.6043	1.8913	-725.6480	-857.7352	-3.9409	-2.1229
0.0178	8.0	10696	1.9119	3.9489	1.9819	0.5990	1.9669	-734.7172	-866.0483	-4.2940	-2.5345
0.0089	9.0	12033	1.9710	3.7315	1.6704	0.5966	2.0611	-737.8326	-868.2217	-5.5045	-3.8933
0.0046	10.0	13370	2.0149	3.5136	1.4530	0.5940	2.0606	-740.0063	-870.4007	-5.9962	-4.4521

Framework versions

• Transformers 4.40.2
• Pytorch 2.1.0+cu118
• Datasets 2.19.1
• Tokenizers 0.19.1