IE_M2_350steps_1e8rate_01beta_cSFTDPO

This model is a fine-tuned version of tsavage68/IE_M2_1000steps_1e7rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.6859
• Rewards/chosen: -0.0025
• Rewards/rejected: -0.0173
• Rewards/accuracies: 0.3900
• Rewards/margins: 0.0148
• Logps/rejected: -41.1945
• Logps/chosen: -42.2302
• Logits/rejected: -2.9157
• Logits/chosen: -2.8544

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-08
• 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: 350

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6934	0.4	50	0.6935	0.0012	0.0018	0.2150	-0.0007	-41.0035	-42.1940	-2.9159	-2.8546
0.6929	0.8	100	0.6915	-0.0004	-0.0037	0.2700	0.0033	-41.0589	-42.2096	-2.9161	-2.8547
0.6886	1.2	150	0.6887	-0.0011	-0.0101	0.3450	0.0091	-41.1233	-42.2161	-2.9157	-2.8544
0.6887	1.6	200	0.6863	0.0011	-0.0128	0.3800	0.0139	-41.1500	-42.1946	-2.9154	-2.8540
0.6886	2.0	250	0.6863	-0.0008	-0.0146	0.3850	0.0138	-41.1676	-42.2131	-2.9157	-2.8544
0.6893	2.4	300	0.6857	-0.0018	-0.0170	0.3900	0.0152	-41.1921	-42.2240	-2.9157	-2.8544
0.6851	2.8	350	0.6859	-0.0025	-0.0173	0.3900	0.0148	-41.1945	-42.2302	-2.9157	-2.8544

Framework versions

• Transformers 4.44.2
• Pytorch 2.0.0+cu117
• Datasets 3.0.0
• Tokenizers 0.19.1