tsavage68's picture
End of training
d5357ee verified
metadata
library_name: transformers
license: llama3
base_model: tsavage68/IE_L3_1000steps_1e6rate_SFT
tags:
  - trl
  - dpo
  - generated_from_trainer
model-index:
  - name: IE_L3_1000steps_1e8rate_01beta_cSFTDPO
    results: []

IE_L3_1000steps_1e8rate_01beta_cSFTDPO

This model is a fine-tuned version of tsavage68/IE_L3_1000steps_1e6rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.6897
  • Rewards/chosen: -0.0098
  • Rewards/rejected: -0.0175
  • Rewards/accuracies: 0.4200
  • Rewards/margins: 0.0078
  • Logps/rejected: -75.8027
  • Logps/chosen: -82.8953
  • Logits/rejected: -0.7964
  • Logits/chosen: -0.7394

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-08
  • train_batch_size: 2
  • eval_batch_size: 1
  • seed: 42
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 4
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 100
  • training_steps: 1000

Training results

Training Loss Epoch Step Validation Loss Rewards/chosen Rewards/rejected Rewards/accuracies Rewards/margins Logps/rejected Logps/chosen Logits/rejected Logits/chosen
0.6965 0.4 50 0.6929 -0.0030 -0.0041 0.3700 0.0011 -75.6681 -82.8275 -0.7963 -0.7392
0.6948 0.8 100 0.6908 -0.0022 -0.0074 0.4250 0.0052 -75.7008 -82.8198 -0.7961 -0.7393
0.6921 1.2 150 0.6946 -0.0077 -0.0055 0.375 -0.0022 -75.6824 -82.8750 -0.7972 -0.7399
0.6892 1.6 200 0.6941 -0.0042 -0.0030 0.3950 -0.0012 -75.6573 -82.8394 -0.7973 -0.7404
0.6937 2.0 250 0.6911 -0.0037 -0.0083 0.4000 0.0046 -75.7098 -82.8345 -0.7973 -0.7405
0.6933 2.4 300 0.6899 -0.0039 -0.0110 0.4300 0.0071 -75.7376 -82.8367 -0.7965 -0.7395
0.6915 2.8 350 0.6870 -0.0023 -0.0151 0.4700 0.0128 -75.7783 -82.8204 -0.7964 -0.7393
0.6933 3.2 400 0.6894 -0.0069 -0.0151 0.4100 0.0082 -75.7783 -82.8666 -0.7958 -0.7387
0.6981 3.6 450 0.6882 0.0006 -0.0100 0.4350 0.0106 -75.7275 -82.7918 -0.7968 -0.7398
0.6904 4.0 500 0.6896 -0.0001 -0.0078 0.4050 0.0077 -75.7054 -82.7989 -0.7958 -0.7391
0.6964 4.4 550 0.6867 -0.0021 -0.0157 0.4400 0.0136 -75.7838 -82.8187 -0.7965 -0.7396
0.6939 4.8 600 0.6902 0.0015 -0.0050 0.4000 0.0065 -75.6771 -82.7829 -0.7968 -0.7398
0.6963 5.2 650 0.6892 -0.0069 -0.0155 0.4200 0.0085 -75.7818 -82.8672 -0.7964 -0.7394
0.6951 5.6 700 0.6873 -0.0025 -0.0149 0.4650 0.0124 -75.7766 -82.8228 -0.7963 -0.7389
0.6855 6.0 750 0.6876 -0.0066 -0.0183 0.4550 0.0118 -75.8105 -82.8633 -0.7965 -0.7394
0.6873 6.4 800 0.6877 -0.0072 -0.0189 0.4550 0.0117 -75.8165 -82.8698 -0.7964 -0.7394
0.6848 6.8 850 0.6898 -0.0098 -0.0173 0.4100 0.0075 -75.8003 -82.8958 -0.7964 -0.7394
0.6983 7.2 900 0.6897 -0.0098 -0.0175 0.4200 0.0078 -75.8027 -82.8953 -0.7964 -0.7394
0.6859 7.6 950 0.6897 -0.0098 -0.0175 0.4200 0.0078 -75.8027 -82.8953 -0.7964 -0.7394
0.6888 8.0 1000 0.6897 -0.0098 -0.0175 0.4200 0.0078 -75.8027 -82.8953 -0.7964 -0.7394

Framework versions

  • Transformers 4.44.2
  • Pytorch 2.0.0+cu117
  • Datasets 3.0.0
  • Tokenizers 0.19.1