zephyr-7b-align-scan-0.0-0.9-polynomial-3

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: 0.9815
• Rewards/chosen: 0.2790
• Rewards/rejected: -1.2918
• Rewards/accuracies: 0.3452
• Rewards/margins: 1.5708
• Logps/rejected: -82.5074
• Logps/chosen: -74.1934
• Logits/rejected: -2.5411
• Logits/chosen: -2.5570

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: 4.135505422578228e-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: polynomial
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.5697	1.0417	100	0.7314	0.9086	0.2746	0.3433	0.6340	-80.8352	-73.5212	-2.5080	-2.5235
0.3568	2.0833	200	0.9404	1.5621	0.3036	0.3294	1.2585	-80.8042	-72.8236	-2.5182	-2.5344

Framework versions

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