phi-2-dpo-test-iter-0

This model is a fine-tuned version of lole25/phi-2-sft-ultrachat-lora on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.0002
• Rewards/chosen: -0.0029
• Rewards/rejected: -0.0032
• Rewards/accuracies: 0.5130
• Rewards/margins: 0.0003
• Logps/rejected: -233.8547
• Logps/chosen: -256.9005
• Logits/rejected: 0.8721
• Logits/chosen: 0.8145

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: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.0001	0.32	100	0.0002	-0.0012	-0.0015	0.5200	0.0003	-233.6874	-256.7341	0.8840	0.8263
0.0001	0.64	200	0.0002	-0.0021	-0.0023	0.5005	0.0002	-233.7691	-256.8278	0.8778	0.8201
0.0001	0.96	300	0.0002	-0.0021	-0.0024	0.4985	0.0003	-233.7780	-256.8272	0.8783	0.8206
0.0001	1.28	400	0.0002	-0.0026	-0.0029	0.5195	0.0003	-233.8277	-256.8757	0.8769	0.8192
0.0001	1.6	500	0.0002	-0.0027	-0.0030	0.5170	0.0003	-233.8388	-256.8869	0.8729	0.8151
0.0001	1.92	600	0.0002	-0.0027	-0.0030	0.5070	0.0003	-233.8414	-256.8860	0.8757	0.8180
0.0001	2.24	700	0.0002	-0.0030	-0.0032	0.5065	0.0002	-233.8592	-256.9123	0.8719	0.8142
0.0001	2.56	800	0.0002	-0.0028	-0.0030	0.5190	0.0003	-233.8422	-256.8898	0.8713	0.8135
0.0001	2.88	900	0.0002	-0.0030	-0.0031	0.5015	0.0002	-233.8529	-256.9111	0.8714	0.8136
0.0001	3.2	1000	0.0002	-0.0029	-0.0033	0.5180	0.0004	-233.8666	-256.9036	0.8733	0.8156
0.0001	3.52	1100	0.0002	-0.0029	-0.0034	0.5265	0.0005	-233.8779	-256.9080	0.8724	0.8145
0.0001	3.84	1200	0.0002	-0.0031	-0.0033	0.5045	0.0003	-233.8733	-256.9227	0.8705	0.8127

Framework versions

• PEFT 0.7.1
• Transformers 4.36.2
• Pytorch 2.2.1+cu121
• Datasets 2.14.6
• Tokenizers 0.15.2