G-ML-Hyly's picture
End of training
b62bc5c
metadata
license: llama2
library_name: peft
tags:
  - generated_from_trainer
base_model: codellama/CodeLlama-13b-Instruct-hf
model-index:
  - name: stg-cli13b-t7-cdp-ca.dt.hlms.cln.inter-b4s1e1-20240102-0727
    results: []

stg-cli13b-t7-cdp-ca.dt.hlms.cln.inter-b4s1e1-20240102-0727

This model is a fine-tuned version of codellama/CodeLlama-13b-Instruct-hf on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0656

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: 0.0002
  • train_batch_size: 4
  • eval_batch_size: 8
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • num_epochs: 1
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss
0.4007 0.02 100 0.0985
0.0888 0.04 200 0.0823
0.0834 0.05 300 0.0800
0.0783 0.07 400 0.0799
0.0782 0.09 500 0.0755
0.0798 0.11 600 0.0771
0.077 0.13 700 0.0734
0.0747 0.14 800 0.0745
0.076 0.16 900 0.0727
0.0791 0.18 1000 0.0775
0.0752 0.2 1100 0.0717
0.0721 0.22 1200 0.0729
0.0731 0.23 1300 0.0710
0.0832 0.25 1400 0.0727
0.0722 0.27 1500 0.0715
0.0738 0.29 1600 0.0715
0.071 0.31 1700 0.0705
0.0738 0.32 1800 0.0713
0.075 0.34 1900 0.0710
0.0732 0.36 2000 0.0703
0.0712 0.38 2100 0.0701
0.0702 0.4 2200 0.0699
0.0733 0.41 2300 0.0697
0.0739 0.43 2400 0.0691
0.0688 0.45 2500 0.0684
0.0692 0.47 2600 0.0689
0.0727 0.49 2700 0.0690
0.073 0.5 2800 0.0685
0.0752 0.52 2900 0.0691
0.0696 0.54 3000 0.0681
0.0708 0.56 3100 0.0684
0.072 0.58 3200 0.0681
0.0716 0.59 3300 0.0689
0.0723 0.61 3400 0.0678
0.0678 0.63 3500 0.0676
0.0695 0.65 3600 0.0672
0.0689 0.67 3700 0.0676
0.0716 0.68 3800 0.0671
0.07 0.7 3900 0.0667
0.0683 0.72 4000 0.0665
0.0704 0.74 4100 0.0664
0.0702 0.76 4200 0.0665
0.0678 0.77 4300 0.0662
0.0679 0.79 4400 0.0661
0.069 0.81 4500 0.0660
0.0675 0.83 4600 0.0661
0.0682 0.85 4700 0.0660
0.0697 0.86 4800 0.0659
0.0689 0.88 4900 0.0658
0.0665 0.9 5000 0.0658
0.067 0.92 5100 0.0657
0.0666 0.94 5200 0.0657
0.0704 0.95 5300 0.0656
0.0682 0.97 5400 0.0656
0.0663 0.99 5500 0.0656

Framework versions

  • Transformers 4.37.0.dev0
  • Pytorch 2.1.2+cu121
  • Datasets 2.15.0
  • Tokenizers 0.15.0

Training procedure

The following bitsandbytes quantization config was used during training:

  • quant_method: QuantizationMethod.BITS_AND_BYTES
  • load_in_8bit: False
  • load_in_4bit: True
  • 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: nf4
  • bnb_4bit_use_double_quant: True
  • bnb_4bit_compute_dtype: bfloat16

Framework versions

  • PEFT 0.6.2