Edit model card

tinymistral-248-DPO

This model is a fine-tuned version of Locutusque/TinyMistral-248M on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.3205
  • Accuracy: 0.0
  • Rewards/chosen: 0.7722
  • Rewards/rejected: -0.2727
  • Rewards/accuracies: 1.0
  • Rewards/margins: 1.0449
  • Logps/rejected: -286.5494
  • Logps/chosen: -398.5646
  • Logits/rejected: -2.3562
  • Logits/chosen: -1.8620

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: 12
  • eval_batch_size: 1
  • seed: 42
  • gradient_accumulation_steps: 12
  • total_train_batch_size: 144
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: constant
  • lr_scheduler_warmup_ratio: 0.05
  • num_epochs: 6
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Accuracy Rewards/chosen Rewards/rejected Rewards/accuracies Rewards/margins Logps/rejected Logps/chosen Logits/rejected Logits/chosen
0.5815 0.48 10 0.3205 0.0 0.7722 -0.2727 1.0 1.0449 -286.5494 -398.5646 -2.3562 -1.8620
0.3287 0.95 20 0.0970 0.0 1.0191 -1.8694 1.0 2.8886 -302.5168 -396.0956 -2.0547 -1.5790
0.2126 1.43 30 0.0414 0.0 0.3685 -4.5314 1.0 4.8999 -329.1370 -402.6024 -1.8100 -1.4099
0.1844 1.9 40 0.0260 0.0 0.9879 -4.8275 1.0 5.8153 -332.0973 -396.4084 -1.8704 -1.4976
0.1546 2.38 50 0.0190 0.0 1.1813 -5.2560 1.0 6.4373 -336.3821 -394.4740 -1.9098 -1.5582
0.1532 2.86 60 0.0140 0.0 1.0583 -6.0198 1.0 7.0780 -344.0201 -395.7045 -1.8920 -1.5654
0.1402 3.33 70 0.0112 0.0 1.0134 -6.5382 1.0 7.5517 -349.2049 -396.1526 -1.8823 -1.5706
0.1544 3.81 80 0.0089 0.0 0.8836 -7.1726 1.0 8.0562 -355.5490 -397.4513 -1.8518 -1.5535
0.1357 4.29 90 0.0072 0.0 0.7532 -7.7663 1.0 8.5195 -361.4852 -398.7546 -1.8193 -1.5345
0.1418 4.76 100 0.0061 0.0 0.6041 -8.3133 1.0 8.9174 -366.9556 -400.2459 -1.7889 -1.5150
0.1482 5.24 110 0.0051 0.0 0.4867 -8.7961 1.0 9.2828 -371.7837 -401.4203 -1.7611 -1.4971
0.141 5.71 120 0.0045 0.0 0.4212 -9.1494 1.0 9.5706 -375.3166 -402.0751 -1.7409 -1.4842

Framework versions

  • Transformers 4.35.2
  • Pytorch 2.1.0+cu121
  • Datasets 2.15.0
  • Tokenizers 0.15.0
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for jtatman/tinymistral-248-DPO-lora

Adapter
(2)
this model