metadata

license: llama2
base_model: epfl-llm/meditron-7b
tags:
  - trl
  - dpo
  - generated_from_trainer
model-index:
  - name: 500STEPS_01beta_DPO_Meditron7B
    results: []

500STEPS_01beta_DPO_Meditron7B

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

eval_loss: 0.6910
eval_runtime: 328.325
eval_samples_per_second: 1.386
eval_steps_per_second: 1.386
eval_rewards/chosen: 0.0001
eval_rewards/rejected: -0.0042
eval_rewards/accuracies: 0.5121
eval_rewards/margins: 0.0044
eval_logps/rejected: -27.8358
eval_logps/chosen: -26.4800
eval_logits/rejected: -0.6115
eval_logits/chosen: -0.6114
epoch: 0.78
step: 400

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: 4
eval_batch_size: 1
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 8
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 100
training_steps: 500

Framework versions

Transformers 4.37.2
Pytorch 2.0.0+cu117
Datasets 2.17.0
Tokenizers 0.15.1