sade-adrien's picture
Model save
02d1431
metadata
license: apache-2.0
base_model: mistralai/Mistral-7B-Instruct-v0.1
tags:
  - generated_from_trainer
model-index:
  - name: Mistral-7B-Instruct-v0.1-LC16k-PI
    results: []

Mistral-7B-Instruct-v0.1-LC16k-PI

This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.1 on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 1.6455

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
  • gradient_accumulation_steps: 32
  • total_train_batch_size: 32
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 30
  • training_steps: 1000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss
1.6742 0.12 100 1.6880
1.6683 0.24 200 1.6711
1.7301 0.36 300 1.6636
1.6867 0.47 400 1.6588
1.4718 0.59 500 1.6557
1.6843 0.71 600 1.6519
1.5966 0.83 700 1.6492
1.9016 0.95 800 1.6472
1.7488 1.07 900 1.6461
1.5596 1.19 1000 1.6455

Framework versions

  • Transformers 4.35.2
  • Pytorch 2.0.0+cu117
  • Datasets 2.14.6
  • Tokenizers 0.14.1