zephyr-7b-align-scan-9e-07-0.86-linear-2.0

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 1.2207
• Rewards/chosen: 0.7526
• Rewards/rejected: -1.3224
• Rewards/accuracies: 0.3294
• Rewards/margins: 2.0750
• Logps/rejected: -82.6661
• Logps/chosen: -73.6161
• Logits/rejected: -2.6026
• Logits/chosen: -2.6190

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: 9e-07
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.9734	0.3484	100	0.9203	2.0652	1.3351	0.3294	0.7302	-79.5760	-72.0898	-2.5695	-2.5853
0.9883	0.6969	200	1.0967	2.4271	1.1885	0.3373	1.2386	-79.7464	-71.6691	-2.5708	-2.5875
0.4215	1.0453	300	1.1234	3.0876	1.7905	0.3313	1.2970	-79.0463	-70.9010	-2.6403	-2.6560
0.393	1.3937	400	1.2234	-0.1343	-1.8934	0.3234	1.7591	-83.3299	-74.6474	-2.6093	-2.6250
0.3986	1.7422	500	1.2247	0.1937	-1.8484	0.3214	2.0420	-83.2776	-74.2660	-2.5909	-2.6070

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.0
• Datasets 2.21.0
• Tokenizers 0.19.1