spin-v-mix

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3123
• Rewards/real: -1.0020
• Rewards/generated: -6.2891
• Rewards/accuracies: 0.8255
• Rewards/margins: 5.2870
• Logps/generated: -171.3206
• Logps/real: -146.1215
• Logits/generated: -2.8818
• Logits/real: -2.7241

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-07
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 32
• 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: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/real	Rewards/generated	Rewards/accuracies	Rewards/margins	Logps/generated	Logps/real	Logits/generated	Logits/real
0.342	0.32	500	0.3170	-0.0900	-5.0675	0.8199	4.9775	-159.1051	-137.0015	-2.8380	-2.6008
0.2836	0.64	1000	0.3160	-0.4491	-5.3698	0.8255	4.9207	-162.1281	-140.5925	-2.8683	-2.6428
0.232	0.96	1500	0.3123	-1.0020	-6.2891	0.8255	5.2870	-171.3206	-146.1215	-2.8818	-2.7241

Framework versions

• Transformers 4.37.0
• Pytorch 2.1.2+cu121
• Datasets 2.14.6
• Tokenizers 0.15.2