FactAlign-Phi-3-Mini

<!-- 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. -->

FactAlign-Phi-3-Mini

This model is aligned with our FactAlign framework for improved long-form factuality, from microsoft/Phi-3-mini-4k-instruct.

For more information, please refer to our paper: FactAlign: Long-form Factuality Alignment of Large Language Models.

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

This model is a fine-tuned version of microsoft/Phi-3-mini-4k-instruct on the trl-lib/kto-mix-14k and the chaoweihuang/lf-response-phi3-f1_100_0.7-fg0.5 datasets. It achieves the following results on the evaluation set:

• Loss: 0.4815
• Rewards/chosen: -0.6601
• Logps/chosen: -299.7121
• Rewards/rejected: -2.6435
• Logps/rejected: -364.3744
• Rewards/margins: 1.9834
• Kl: 0.0081
• Fg Kl: nan
• Fg Rewards/chosen Sum: 0.0694
• Fg Logps/policy Chosen: -15.2781
• Fg Logps/reference Chosen: -14.9295
• Count/fg Chosen: 16.0137
• Fg Rewards/rejected Sum: -0.3623
• Fg Logps/policy Rejected: -19.6552
• Fg Logps/reference Rejected: -18.7868
• Count/fg Rejected: 4.0824
• Fg Logps/policy Kl: -21.1260
• Fg Logps/reference Kl: -20.2070
• Fg Loss: 0.7365

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-07
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• total_eval_batch_size: 2
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1.0
• mixed_precision_training: Native AMP

Training results

Framework versions

• Transformers 4.41.1
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1