Llama-3.1-8B-Instruct-KTO-700

This model is a fine-tuned version of meta-llama/Meta-Llama-3.1-8B-Instruct on the bct_non_cot_kto_700 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2141
• Rewards/chosen: -0.1237
• Logps/chosen: -17.0960
• Logits/chosen: -4337002.6667
• Rewards/rejected: -6.8517
• Logps/rejected: -87.8667
• Logits/rejected: -6325528.9351
• Rewards/margins: 6.7280
• Kl: 0.0

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
0.4998	0.6349	50	0.4994	0.0114	-15.7443	-6144470.3492	0.0050	-19.2993	-7140612.1558	0.0064	4.0692
0.4897	1.2730	100	0.4845	0.1738	-14.1208	-6027973.0794	0.0297	-19.0521	-7100376.1039	0.1440	8.7304
0.3937	1.9079	150	0.3745	0.3253	-12.6062	-5056723.3016	-0.7501	-26.8503	-6579766.0260	1.0753	0.8562
0.2821	2.5460	200	0.2748	0.3148	-12.7111	-4519878.6032	-2.1742	-41.0916	-6449476.5714	2.4890	0.0
0.2081	3.1841	250	0.2370	0.3186	-12.6723	-4612124.9524	-3.4610	-53.9597	-6554266.5974	3.7797	0.0
0.2651	3.8190	300	0.2267	0.0307	-15.5522	-4312815.7460	-5.2909	-72.2581	-6286123.2208	5.3215	0.0
0.194	4.4571	350	0.2218	0.0573	-15.2853	-4256054.3492	-5.4598	-73.9474	-6266110.3377	5.5171	0.0
0.168	5.0952	400	0.2218	-0.0280	-16.1385	-4093047.3651	-6.0281	-79.6308	-6134765.7143	6.0002	0.0
0.2268	5.7302	450	0.2163	-0.1488	-17.3463	-4156519.1111	-6.7700	-87.0499	-6205292.8831	6.6213	0.0
0.1915	6.3683	500	0.2194	-0.0833	-16.6922	-4369097.6508	-6.4897	-84.2469	-6383271.0649	6.4064	0.0
0.201	7.0063	550	0.2189	-0.1198	-17.0568	-4304407.3651	-6.7293	-86.6421	-6343543.6883	6.6095	0.0
0.1961	7.6413	600	0.2157	-0.1073	-16.9320	-4324965.0794	-6.7349	-86.6987	-6328156.6753	6.6276	0.0
0.1721	8.2794	650	0.2157	-0.1282	-17.1405	-4329249.5238	-6.8581	-87.9301	-6320159.5844	6.7299	0.0
0.1879	8.9143	700	0.2141	-0.1237	-17.0960	-4337002.6667	-6.8517	-87.8667	-6325528.9351	6.7280	0.0
0.2335	9.5524	750	0.2158	-0.1282	-17.1407	-4329628.4444	-6.8618	-87.9676	-6335641.3506	6.7336	0.0

Framework versions

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