models / arr2-10w /README.md
yijiexu's picture
Upload folder using huggingface_hub
1c96952 verified
|
Raw History Blame Contribute Delete
1.44 kB
metadata
license: other
base_model: /root/models/Llama-2-7b-hf
tags:
  - llama-factory
  - full
  - generated_from_trainer
model-index:
  - name: en
    results: []

en

This model is a fine-tuned version of /root/models/Llama-2-7b-hf on the aar_k2_en dataset. It achieves the following results on the evaluation set:

  • Loss: 1.8003

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: 5e-05
  • train_batch_size: 1
  • eval_batch_size: 1
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 8
  • total_train_batch_size: 8
  • total_eval_batch_size: 8
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 0.1
  • num_epochs: 1.0

Training results

Training Loss Epoch Step Validation Loss
1.9563 0.52 10000 1.9556

Framework versions

  • Transformers 4.37.2
  • Pytorch 2.3.0
  • Datasets 2.19.1
  • Tokenizers 0.15.1