lukeleeai's picture
End of training
759985a verified
metadata
license: apache-2.0
base_model: mistralai/Mistral-7B-Instruct-v0.1
tags:
  - generated_from_trainer
datasets:
  - openwebtext
model-index:
  - name: sparse_sparse_80_percent_pretraining_warmup_20K_steps_5k
    results: []

sparse_sparse_80_percent_pretraining_warmup_20K_steps_5k

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

  • Loss: 0.7590

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-05
  • train_batch_size: 8
  • eval_batch_size: 16
  • seed: 0
  • distributed_type: multi-GPU
  • num_devices: 6
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 96
  • total_eval_batch_size: 96
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • training_steps: 5000

Training results

Training Loss Epoch Step Validation Loss
1.2712 0.05 50 1.2374
1.0533 0.1 100 1.0529
0.9603 0.15 150 0.9668
0.9102 0.19 200 0.9145
0.8754 0.24 250 0.8775
0.8514 0.29 300 0.8503
0.8417 0.34 350 0.8298
0.8217 0.39 400 0.8146
0.8019 0.44 450 0.8026
0.7902 0.48 500 0.7914
0.7856 0.53 550 0.7819
0.7599 0.58 600 0.7734
0.7646 0.63 650 0.7689
0.7542 0.68 700 0.7635
0.7529 0.73 750 0.7581
0.7594 0.78 800 0.7533
0.7489 0.82 850 0.7493
0.7494 0.87 900 0.7452
0.7441 0.92 950 0.7472
0.7467 0.97 1000 0.7442
0.728 1.02 1050 0.7413
0.7263 1.07 1100 0.7384
0.7206 1.11 1150 0.7362
0.7223 1.16 1200 0.7343
0.7362 1.21 1250 0.7421
0.7374 1.26 1300 0.7401
0.7284 1.31 1350 0.7378
0.7309 1.36 1400 0.7356
0.724 1.41 1450 0.7339
0.72 1.45 1500 0.7317
0.73 1.5 1550 0.7509
0.7464 1.55 1600 0.7489
0.742 1.6 1650 0.7461
0.7378 1.65 1700 0.7447
0.7328 1.7 1750 0.7433
0.7433 1.75 1800 0.7411

Framework versions

  • Transformers 4.35.2
  • Pytorch 2.1.1+cu121
  • Datasets 2.15.0
  • Tokenizers 0.15.0