Hugging Face's logo

VitaliiVrublevskyi
/
Llama-2-7b-hf-finetuned-mrpc-v0.4

TensorBoard
Generated from Trainer
Model card Files Metrics Community
1
Llama-2-7b-hf-finetuned-mrpc-v0.4 / README.md
VitaliiVrublevskyi's picture
VitaliiVrublevskyi
End of training
ca43712
preview code
|
raw
history blame
3.91 kB
metadata
base_model: meta-llama/Llama-2-7b-hf
tags:
  - generated_from_trainer
datasets:
  - glue
metrics:
  - accuracy
  - f1
model-index:
  - name: Llama-2-7b-hf-finetuned-mrpc-v0.4
    results: []
library_name: peft

Llama-2-7b-hf-finetuned-mrpc-v0.4

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

  • Loss: 0.4717
  • Accuracy: 0.8676
  • F1: 0.9046

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

The following bitsandbytes quantization config was used during training:

  • load_in_8bit: True
  • load_in_4bit: False
  • llm_int8_threshold: 6.0
  • llm_int8_skip_modules: None
  • llm_int8_enable_fp32_cpu_offload: False
  • llm_int8_has_fp16_weight: False
  • bnb_4bit_quant_type: fp4
  • bnb_4bit_use_double_quant: False
  • bnb_4bit_compute_dtype: float32

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 1.6000000000000003e-05
  • train_batch_size: 16
  • eval_batch_size: 16
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 30

Training results

Training Loss Epoch Step Accuracy F1 Validation Loss
No log 1.0 230 0.6446 0.7695 0.6542
No log 2.0 460 0.6912 0.7968 0.5938
0.6489 3.0 690 0.7230 0.8151 0.5694
0.6489 4.0 920 0.7230 0.8138 0.5503
0.5299 5.0 1150 0.7402 0.8251 0.5492
0.5299 6.0 1380 0.7794 0.8432 0.4880
0.4687 7.0 1610 0.8064 0.8663 0.4559
0.4687 8.0 1840 0.8186 0.875 0.4298
0.374 9.0 2070 0.8284 0.8818 0.4210
0.374 10.0 2300 0.8456 0.8916 0.3953
0.3096 11.0 2530 0.8431 0.8897 0.4074
0.3096 12.0 2760 0.8407 0.8862 0.4030
0.3096 13.0 2990 0.8456 0.8904 0.3982
0.2799 14.0 3220 0.8456 0.8881 0.3873
0.2799 15.0 3450 0.8529 0.8940 0.3939
0.2511 16.0 3680 0.8431 0.8877 0.4018
0.2511 17.0 3910 0.8529 0.8947 0.3969
0.2371 18.0 4140 0.8456 0.8912 0.3963
0.2371 19.0 4370 0.8578 0.8964 0.3865
0.2211 20.0 4600 0.8505 0.8928 0.4165
0.2211 21.0 4830 0.4070 0.8456 0.8901
0.2136 22.0 5060 0.4090 0.8578 0.8972
0.2136 23.0 5290 0.4328 0.8578 0.8961
0.1774 24.0 5520 0.4602 0.8382 0.8791
0.1774 25.0 5750 0.4551 0.8627 0.9018
0.1774 26.0 5980 0.4677 0.8505 0.8920
0.1521 27.0 6210 0.4854 0.8578 0.8953
0.1521 28.0 6440 0.5064 0.8505 0.8932
0.134 29.0 6670 0.4971 0.8603 0.8988
0.134 30.0 6900 0.4717 0.8676 0.9046

Framework versions

  • PEFT 0.4.0
  • Transformers 4.31.0
  • Pytorch 2.0.1+cu118
  • Datasets 2.14.5
  • Tokenizers 0.13.3