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_0.1_beta_DPO
    results: []

mistralit2_250_STEPS_1e7_rate_0.1_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: 0.6226
Rewards/chosen: -1.1689
Rewards/rejected: -1.4764
Rewards/accuracies: 0.5912
Rewards/margins: 0.3075
Logps/rejected: -43.3368
Logps/chosen: -35.0753
Logits/rejected: -2.8107
Logits/chosen: -2.8112

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
0.6877	0.1	50	0.6853	-0.0549	-0.0714	0.5846	0.0165	-29.2866	-23.9352	-2.8632	-2.8635
0.6422	0.2	100	0.6445	-0.6597	-0.8190	0.5758	0.1593	-36.7627	-29.9831	-2.8277	-2.8281
0.5398	0.29	150	0.6295	-1.2391	-1.5382	0.5978	0.2990	-43.9541	-35.7771	-2.8089	-2.8094
0.5742	0.39	200	0.6233	-1.1703	-1.4764	0.5934	0.3061	-43.3366	-35.0892	-2.8107	-2.8112
0.6425	0.49	250	0.6226	-1.1689	-1.4764	0.5912	0.3075	-43.3368	-35.0753	-2.8107	-2.8112

Framework versions

Transformers 4.38.2
Pytorch 2.0.0+cu117
Datasets 2.18.0
Tokenizers 0.15.2