metadata

license: apache-2.0
base_model: mistralai/Mistral-7B-Instruct-v0.2
tags:
  - trl
  - dpo
  - generated_from_trainer
model-index:
  - name: mistralit2_250_STEPS_1e7_rate_05_beta_DPO
    results: []

mistralit2_250_STEPS_1e7_rate_05_beta_DPO

This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.2 on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 0.6043
Rewards/chosen: -1.1749
Rewards/rejected: -1.6856
Rewards/accuracies: 0.6330
Rewards/margins: 0.5107
Logps/rejected: -31.9435
Logps/chosen: -25.7355
Logits/rejected: -2.8536
Logits/chosen: -2.8539

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: 250

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6684	0.1	50	0.6660	-0.2264	-0.2957	0.5934	0.0693	-29.1637	-23.8386	-2.8636	-2.8639
0.5945	0.2	100	0.6396	-1.5064	-1.9635	0.6044	0.4572	-32.4994	-26.3985	-2.8444	-2.8447
0.4857	0.29	150	0.6440	-2.0003	-2.6364	0.6198	0.6362	-33.8453	-27.3863	-2.8460	-2.8463
0.5631	0.39	200	0.6018	-1.1675	-1.6769	0.6330	0.5093	-31.9261	-25.7209	-2.8536	-2.8539
0.6109	0.49	250	0.6043	-1.1749	-1.6856	0.6330	0.5107	-31.9435	-25.7355	-2.8536	-2.8539

Framework versions

Transformers 4.38.2
Pytorch 2.0.0+cu117
Datasets 2.18.0
Tokenizers 0.15.2