icml2025-submission409/VersaPRM

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

VersaPRM

This model is a fine-tuned version of icml2025-submission409/LlamaPRM800K on an unknown dataset. It achieves the following results on the evaluation set:

• —Loss: 0.2555
• —Prm accuracy: 0.8968
• —Prm precision: 0.8983
• —Prm recall: 0.9907
• —Prm specificty: 0.3684
• —Prm npv: 0.875
• —Prm f1: 0.9422
• —Prm f1 neg: 0.5185
• —Prm f1 auc: 0.6795
• —Prm f1 auc (fixed): 0.9223

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: 0.0001
• —trainbatchsize: 2
• —evalbatchsize: 4
• —seed: 908932403
• —distributed_type: multi-GPU
• —num_devices: 8
• —gradientaccumulationsteps: 2
• —totaltrainbatch_size: 32
• —totalevalbatch_size: 32
• —optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
• —lrschedulertype: cosine
• —lrschedulerwarmup_ratio: 0.1
• —num_epochs: 1

Training results

Framework versions

• —PEFT 0.12.0
• —Transformers 4.46.0
• —Pytorch 2.4.0+cu118
• —Datasets 3.0.0
• —Tokenizers 0.20.1