tsavage68's picture
End of training
ccb05b5 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_450steps_1e8rate_01beta_cSFTDPO
    results: []

IE_L3_450steps_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: 0.6907
  • Rewards/chosen: -0.0040
  • Rewards/rejected: -0.0095
  • Rewards/accuracies: 0.4050
  • Rewards/margins: 0.0055
  • Logps/rejected: -75.7223
  • Logps/chosen: -82.8379
  • Logits/rejected: -0.7979
  • Logits/chosen: -0.7409

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: 450

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
0.6904 1.2 150 0.6912 -0.0066 -0.0112 0.4200 0.0046 -75.7390 -82.8636 -0.7971 -0.7401
0.6902 1.6 200 0.6897 -0.0027 -0.0101 0.4250 0.0074 -75.7282 -82.8243 -0.7964 -0.7397
0.6858 2.0 250 0.6904 -0.0049 -0.0110 0.3950 0.0061 -75.7372 -82.8472 -0.7971 -0.7403
0.6903 2.4 300 0.6887 -0.0076 -0.0170 0.4500 0.0094 -75.7977 -82.8741 -0.7971 -0.7401
0.6859 2.8 350 0.6898 -0.0058 -0.0130 0.4150 0.0072 -75.7575 -82.8558 -0.7979 -0.7409
0.6978 3.2 400 0.6907 -0.0040 -0.0095 0.4050 0.0055 -75.7223 -82.8379 -0.7979 -0.7409
0.6889 3.6 450 0.6907 -0.0040 -0.0095 0.4050 0.0055 -75.7223 -82.8379 -0.7979 -0.7409

Framework versions

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