stablelm-2-1.6-dpo-disticoder-v0.1

This model is a fine-tuned version of plaguss/stablelm-2-1_6-sft-disticoder-v01 on the argilla/DistiCoder-dpo-binarized-train dataset. It achieves the following results on the evaluation set:

• Loss: 0.7398
• Rewards/chosen: -0.0026
• Rewards/rejected: -0.0002
• Rewards/accuracies: 0.4902
• Rewards/margins: -0.0024
• Logps/rejected: -359.7791
• Logps/chosen: -297.9016
• Logits/rejected: -0.9458
• Logits/chosen: -0.9673

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: 1e-07
• 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: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.7397	1.0	288	0.7419	-0.0098	-0.0101	0.5	0.0003	-359.7990	-297.9159	-0.9478	-0.9696
0.718	2.0	576	0.7291	0.0095	-0.0100	0.5117	0.0194	-359.7986	-297.8773	-0.9464	-0.9679
0.6923	3.0	864	0.7398	-0.0026	-0.0002	0.4902	-0.0024	-359.7791	-297.9016	-0.9458	-0.9673

Framework versions

• PEFT 0.8.2
• Transformers 4.37.2
• Pytorch 2.1.1+cu121
• Datasets 2.16.1
• Tokenizers 0.15.2