AutoRE / relation /README.md
dante123's picture
Upload 42 files
ab5df65 verified
metadata
license: other
library_name: peft
tags:
  - llama-factory
  - lora
  - generated_from_trainer
base_model: /workspace/xll/checkpoints/lmsys/vicuna-7b-v1.5
model-index:
  - name: relation
    results: []

relation

This model is a fine-tuned version of /workspace/xll/checkpoints/lmsys/vicuna-7b-v1.5 on the vicuna_relation_test dataset. It achieves the following results on the evaluation set:

  • Loss: 0.4715

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:

  • quant_method: bitsandbytes
  • 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: float16

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 5e-05
  • train_batch_size: 8
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 32
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • training_steps: 1200
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss
3.0929 0.02 10 2.1679
1.8074 0.04 20 1.4093
1.2188 0.05 30 1.1259
1.0841 0.07 40 0.9606
0.9594 0.09 50 0.8682
0.8765 0.11 60 0.8166
0.852 0.12 70 0.7803
0.8404 0.14 80 0.7602
0.8183 0.16 90 0.7223
0.7816 0.18 100 0.7134
0.7792 0.19 110 0.7234
0.7648 0.21 120 0.6883
0.8132 0.23 130 0.7020
0.7599 0.25 140 0.6684
0.7518 0.26 150 0.6716
0.7452 0.28 160 0.6634
0.7215 0.3 170 0.6610
0.7088 0.32 180 0.6598
0.7237 0.33 190 0.6470
0.7353 0.35 200 0.6315
0.7111 0.37 210 0.6466
0.7136 0.39 220 0.6329
0.7044 0.4 230 0.6357
0.7369 0.42 240 0.6215
0.6995 0.44 250 0.6103
0.7027 0.46 260 0.5964
0.6872 0.47 270 0.6044
0.7182 0.49 280 0.6127
0.6897 0.51 290 0.6016
0.6824 0.53 300 0.5880
0.673 0.54 310 0.5902
0.6956 0.56 320 0.5811
0.6889 0.58 330 0.5850
0.6773 0.6 340 0.5934
0.6782 0.61 350 0.5942
0.719 0.63 360 0.5849
0.6809 0.65 370 0.5799
0.6412 0.67 380 0.5710
0.6411 0.69 390 0.5628
0.6519 0.7 400 0.5612
0.6446 0.72 410 0.5562
0.6574 0.74 420 0.5645
0.6633 0.76 430 0.5664
0.6673 0.77 440 0.5483
0.6466 0.79 450 0.5542
0.653 0.81 460 0.5411
0.6384 0.83 470 0.5362
0.6287 0.84 480 0.5453
0.661 0.86 490 0.5645
0.608 0.88 500 0.5245
0.6584 0.9 510 0.5376
0.6416 0.91 520 0.5471
0.6527 0.93 530 0.5426
0.6164 0.95 540 0.5284
0.611 0.97 550 0.5313
0.614 0.98 560 0.5263
0.6382 1.0 570 0.5317
0.5804 1.02 580 0.5207
0.6291 1.04 590 0.5238
0.5911 1.05 600 0.5174
0.6111 1.07 610 0.5281
0.5578 1.09 620 0.5255
0.6055 1.11 630 0.5177
0.6015 1.12 640 0.5131
0.6072 1.14 650 0.5168
0.5956 1.16 660 0.5169
0.6099 1.18 670 0.5170
0.6038 1.19 680 0.5056
0.583 1.21 690 0.5121
0.5885 1.23 700 0.5234
0.5784 1.25 710 0.5028
0.5744 1.26 720 0.5100
0.6014 1.28 730 0.5038
0.6185 1.3 740 0.5146
0.6184 1.32 750 0.5317
0.6141 1.34 760 0.5080
0.6146 1.35 770 0.5165
0.5721 1.37 780 0.5040
0.5931 1.39 790 0.4934
0.5944 1.41 800 0.4876
0.6002 1.42 810 0.4930
0.5557 1.44 820 0.4913
0.58 1.46 830 0.4910
0.5459 1.48 840 0.4884
0.5871 1.49 850 0.4860
0.5554 1.51 860 0.4857
0.5819 1.53 870 0.4649
0.5649 1.55 880 0.4790
0.5779 1.56 890 0.4807
0.5756 1.58 900 0.4834
0.5563 1.6 910 0.4946
0.5393 1.62 920 0.4848
0.5551 1.63 930 0.4845
0.5687 1.65 940 0.4807
0.5469 1.67 950 0.4749
0.5771 1.69 960 0.4859
0.5689 1.7 970 0.4734
0.5741 1.72 980 0.4882
0.5643 1.74 990 0.4816
0.5603 1.76 1000 0.4676
0.5925 1.77 1010 0.4686
0.5834 1.79 1020 0.4743
0.5902 1.81 1030 0.4916
0.5777 1.83 1040 0.4748
0.5921 1.84 1050 0.4843
0.5877 1.86 1060 0.4742
0.5453 1.88 1070 0.4705
0.5445 1.9 1080 0.4663
0.5686 1.92 1090 0.4745
0.5712 1.93 1100 0.4888
0.6032 1.95 1110 0.4861
0.5491 1.97 1120 0.4721
0.5452 1.99 1130 0.4645
0.5526 2.0 1140 0.4877
0.5443 2.02 1150 0.4716
0.5103 2.04 1160 0.4632
0.5202 2.06 1170 0.4802
0.5436 2.07 1180 0.4681
0.5454 2.09 1190 0.4709
0.5183 2.11 1200 0.4742

Framework versions

  • PEFT 0.7.0
  • Transformers 4.37.1
  • Pytorch 2.1.2+cu121
  • Datasets 2.16.1
  • Tokenizers 0.15.1