saiteja33/gemma2-2b-hdpo-zh

<!-- 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. -->

polyalign

This model is a fine-tuned version of /vast/users/muhammad.haris/Sathira/PolyAlign/saves/gemma2_2b_dist-sft-zh on the polyalignhdpotrain dataset. It achieves the following results on the evaluation set:

• —Loss: 0.3364
• —Rewards/chosen: -1.2067
• —Rewards/rejected: -6.7700
• —Rewards/accuracies: 0.8957
• —Rewards/margins: 5.5633
• —Logps/chosen: -206.9487
• —Logps/rejected: -130.6130
• —Logits/chosen: -1.6500
• —Logits/rejected: -1.8634
• —Hdpo/weight Mean: 1.1321
• —Hdpo/chosen Dist Score: 0.0442
• —Hdpo/rejected Dist Score: 0.1547
• —Hdpo/dist Gap: 0.1105
• —Hdpo/dist Reg: 0.0870
• —Hdpo/sft Anchor: 1.5021

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: 5e-07
• —trainbatchsize: 1
• —evalbatchsize: 1
• —seed: 42
• —distributed_type: multi-GPU
• —num_devices: 8
• —gradientaccumulationsteps: 8
• —totaltrainbatch_size: 64
• —totalevalbatch_size: 8
• —optimizer: Use OptimizerNames.ADAMWTORCHFUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• —lrschedulertype: cosine
• —lrschedulerwarmup_ratio: 0.1
• —num_epochs: 1.0

Training results

Framework versions

• —Transformers 4.56.2
• —Pytorch 2.9.1+rocm6.3
• —Datasets 4.0.0
• —Tokenizers 0.22.2