zephyr-7b-spin-iter1-v0

This model is a fine-tuned version of argilla/zephyr-7b-spin-iter0-v0 on the argilla/10k_prompts_SPIN_iter1_zephyr_top and argilla/10k_prompts_SPIN_iter0_zephyr_top dataset.

It achieves the following results on the evaluation set:

• Loss: 0.0831
• Rewards/real: 1.3037
• Rewards/generated: -5.4434
• Rewards/accuracies: 0.9792
• Rewards/margins: 6.7471
• Logps/generated: -545.0309
• Logps/real: -272.3726
• Logits/generated: -2.6844
• Logits/real: -2.7197

MT-Bench results

Model	1st Turn Score	2nd Turn Score	Average Score
zephyr-7b-sft-full	6.6625	6.0250	6.34375
zephyr-7b-spin-iter0-v0	6.64375	6.1750	6.409375
zephyr-7b-spin-iter1-v0	6.90625	6.3000	6.603125
zephyr-7b-spin-iter2-v0	7.1375	6.3125	6.725000
zephyr-7b-spin-iter3-v0	7.09375	6.4500	6.771875

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
• 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: 2.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/real	Rewards/generated	Rewards/accuracies	Rewards/margins	Logps/generated	Logps/real	Logits/generated	Logits/real
0.1827	0.49	25	0.1651	0.1714	-3.3650	0.9688	3.5364	-524.2469	-283.6962	-2.7482	-2.7944
0.0462	0.97	50	0.0835	1.4823	-4.4998	1.0	5.9821	-535.5947	-270.5871	-2.6963	-2.7356
0.0047	1.46	75	0.0837	1.3725	-5.2500	0.9896	6.6225	-543.0965	-271.6846	-2.6847	-2.7211
0.0034	1.94	100	0.0831	1.3037	-5.4434	0.9792	6.7471	-545.0309	-272.3726	-2.6844	-2.7197

Framework versions

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