UTI_L3_175steps_1e7rate_03beta_CSFTDPO

This model is a fine-tuned version of tsavage68/UTI_L3_1000steps_1e5rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3086
• Rewards/chosen: 0.2238
• Rewards/rejected: -0.9996
• Rewards/accuracies: 0.9900
• Rewards/margins: 1.2234
• Logps/rejected: -66.5267
• Logps/chosen: -31.7329
• Logits/rejected: -1.3243
• Logits/chosen: -1.3093

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: 1e-07
• train_batch_size: 2
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• training_steps: 175

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6945	0.3333	25	0.6909	-0.0005	-0.0056	0.4900	0.0051	-63.2133	-32.4808	-1.3230	-1.3079
0.6631	0.6667	50	0.6538	0.0144	-0.0676	0.8500	0.0820	-63.4201	-32.4310	-1.3232	-1.3082
0.6008	1.0	75	0.5691	0.0482	-0.2288	0.9600	0.2770	-63.9573	-32.3183	-1.3233	-1.3082
0.4499	1.3333	100	0.4399	0.1150	-0.5411	0.9600	0.6561	-64.9983	-32.0957	-1.3238	-1.3088
0.3285	1.6667	125	0.3355	0.1971	-0.8933	0.9900	1.0904	-66.1723	-31.8220	-1.3241	-1.3092
0.3053	2.0	150	0.3088	0.2246	-1.0014	0.9900	1.2260	-66.5327	-31.7303	-1.3242	-1.3092
0.2456	2.3333	175	0.3086	0.2238	-0.9996	0.9900	1.2234	-66.5267	-31.7329	-1.3243	-1.3093

Framework versions

• Transformers 4.41.2
• Pytorch 2.0.0+cu117
• Datasets 2.19.2
• Tokenizers 0.19.1