metadata

license: apache-2.0
base_model: Intel/neural-chat-7b-v3-1
tags:
  - alignment-handbook
  - generated_from_trainer
datasets:
  - HuggingFaceH4/ultrafeedback_binarized
model-index:
  - name: juanako-7b-v2-UNA-v3
    results: []

juanako-7b-v2-UNA-v3

This model is a fine-tuned version of Intel/neural-chat-7b-v3-1 on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

Loss: 0.4658
Rewards/chosen: -0.6178
Rewards/rejected: -1.7640
Rewards/accuracies: 0.7622
Rewards/margins: 1.1462
Logps/rejected: -239.1435
Logps/chosen: -227.6597
Logits/rejected: -2.2452
Logits/chosen: -2.4812

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: 0.0001
train_batch_size: 1
eval_batch_size: 1
seed: 42
distributed_type: multi-GPU
num_devices: 14
gradient_accumulation_steps: 16
total_train_batch_size: 224
total_eval_batch_size: 14
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.01
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
0.4855	0.25	69	0.4897	-0.7296	-1.8563	0.7413	1.1266	-240.0657	-228.7776	-2.2474	-2.4843
0.4835	0.5	138	0.4734	-0.5055	-1.5652	0.7483	1.0597	-237.1553	-226.5366	-2.2448	-2.4811
0.5193	0.75	207	0.4717	-0.6888	-1.7561	0.7343	1.0673	-239.0642	-228.3696	-2.2426	-2.4783
0.4514	1.0	276	0.4658	-0.6178	-1.7640	0.7622	1.1462	-239.1435	-227.6597	-2.2452	-2.4812

Framework versions

Transformers 4.35.0
Pytorch 2.1.0
Datasets 2.14.6
Tokenizers 0.14.1