metadata

license: apache-2.0
library_name: peft
tags:
  - trl
  - dpo
  - generated_from_trainer
base_model: TheBloke/OpenHermes-2-Mistral-7B-GPTQ
model-index:
  - name: openhermes-mistral-dpo-gptq
    results: []

openhermes-mistral-dpo-gptq

This model is a fine-tuned version of TheBloke/OpenHermes-2-Mistral-7B-GPTQ on the None dataset. It achieves the following results on the evaluation set:

Loss: 0.5794
Rewards/chosen: 0.3240
Rewards/rejected: -0.1304
Rewards/accuracies: 0.8125
Rewards/margins: 0.4544
Logps/rejected: -257.9009
Logps/chosen: -299.4479
Logits/rejected: -2.5816
Logits/chosen: -2.4558

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: 0.0002
train_batch_size: 1
eval_batch_size: 8
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 2
training_steps: 50
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6803	0.005	10	0.6477	0.1422	-0.0051	0.9375	0.1472	-256.6473	-301.2661	-2.5813	-2.4540
0.7051	0.01	20	0.6019	0.1764	-0.1259	0.9375	0.3023	-257.8556	-300.9241	-2.5817	-2.4542
0.7343	0.015	30	0.5848	0.1973	-0.1493	0.875	0.3465	-258.0895	-300.7151	-2.5833	-2.4551
0.6977	0.02	40	0.5773	0.2710	-0.1295	0.8125	0.4005	-257.8920	-299.9776	-2.5815	-2.4542
0.6483	0.025	50	0.5794	0.3240	-0.1304	0.8125	0.4544	-257.9009	-299.4479	-2.5816	-2.4558

Framework versions

PEFT 0.11.1
Transformers 4.41.1
Pytorch 2.0.1+cu117
Datasets 2.19.1
Tokenizers 0.19.1