meditron-7b-dpo-full-wo-live_qa-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: 0.5356
• Rewards/chosen: -0.2871
• Rewards/rejected: -0.7760
• Rewards/accuracies: 0.6923
• Rewards/margins: 0.4889
• Logps/rejected: -1205.3544
• Logps/chosen: -986.1032
• Logits/rejected: -0.8878
• Logits/chosen: -0.8900

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	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.583	0.49	100	0.6358	-0.0392	-0.1400	0.6442	0.1009	-1141.7539	-961.3083	-0.8346	-0.8420
0.3768	0.98	200	0.5356	-0.2875	-0.7751	0.7019	0.4876	-1205.2603	-986.1399	-0.8883	-0.8904

Framework versions

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