meditron-7b-dpo-full-wo-kqa_golden-ep3

This model is a fine-tuned version of epfl-llm/meditron-7b on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.4459
• Rewards/chosen: -0.4566
• Rewards/rejected: -1.4012
• Rewards/accuracies: 0.8068
• Rewards/margins: 0.9447
• Logps/rejected: -1444.6896
• Logps/chosen: -859.0582
• Logits/rejected: -0.9203
• Logits/chosen: -0.8310

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: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Logits/chosen	Logits/rejected	Logps/chosen	Logps/rejected	Validation Loss	Rewards/accuracies	Rewards/chosen	Rewards/margins	Rewards/rejected
0.5643	0.5	100	-0.6995	-0.8645	-818.2397	-1334.0771	0.5890	0.7727	-0.0484	0.2467	-0.2951
0.3959	1.0	200	-0.8310	-0.9203	-859.0582	-1444.6896	0.4459	0.8068	-0.4566	0.9447	-1.4012

Framework versions

• Transformers 4.39.0.dev0
• Pytorch 2.1.2
• Datasets 2.14.6
• Tokenizers 0.15.2