Llama-3.1-8B-Instruct-KTO-200

This model is a fine-tuned version of meta-llama/Meta-Llama-3.1-8B-Instruct on the bct_non_cot_kto_200 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4178
• Rewards/chosen: 0.0409
• Logps/chosen: -18.0110
• Logits/chosen: -5352758.4
• Rewards/rejected: -0.9102
• Logps/rejected: -31.7024
• Logits/rejected: -8100233.0667
• Rewards/margins: 0.9511
• Kl: 0.9340

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-06
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 10.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Logps/chosen	Logits/chosen	Rewards/rejected	Logps/rejected	Logits/rejected	Rewards/margins
0.4954	2.2222	50	0.4960	0.0262	-18.1577	-6157739.52	-0.0022	-22.6226	-8250512.0	0.0285	0.8099
0.4457	4.4444	100	0.4627	0.0628	-17.7918	-5773483.52	-0.3038	-25.6385	-8211033.6	0.3666	1.5104
0.3734	6.6667	150	0.4264	0.0610	-17.8096	-5435994.88	-0.7496	-30.0960	-8134988.8	0.8106	1.2779
0.3749	8.8889	200	0.4178	0.0409	-18.0110	-5352758.4	-0.9102	-31.7024	-8100233.0667	0.9511	0.9340

Framework versions

• PEFT 0.12.0
• Transformers 4.46.1
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.20.3