phi-2-sft-openhermes-128k-v2

This model is a fine-tuned version of microsoft/phi-2 on the generator dataset. It achieves the following results on the evaluation set:

• Loss: 1.0663
• Model Preparation Time: 0.0259

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: 1.5e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 1
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time
1.0842	0.0809	200	1.0744	0.0259
1.0767	0.1617	400	1.0724	0.0259
1.0997	0.2426	600	1.0716	0.0259
1.0758	0.3234	800	1.0707	0.0259
1.0949	0.4043	1000	1.0697	0.0259
1.0892	0.4851	1200	1.0687	0.0259
1.0876	0.5660	1400	1.0682	0.0259
1.0873	0.6469	1600	1.0676	0.0259
1.0679	0.7277	1800	1.0671	0.0259
1.0802	0.8086	2000	1.0669	0.0259
1.0823	0.8894	2200	1.0667	0.0259
1.0812	0.9703	2400	1.0663	0.0259

Framework versions

• PEFT 0.12.0
• Transformers 4.44.0
• Pytorch 2.4.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1