zephyr-spin-mix-data-zephyr-llama2

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: 0.0032
Rewards/real: -1.1301
Rewards/generated: -24.4098
Rewards/accuracies: 1.0
Rewards/margins: 23.2798
Logps/generated: -343.8705
Logps/real: -147.1187
Logits/generated: -2.4706
Logits/real: -2.1426

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: 3
total_train_batch_size: 24
total_eval_batch_size: 24
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
0.004	0.23	500	0.0090	-0.6333	-17.9967	0.9978	17.3635	-279.7397	-142.1509	-2.7055	-2.3748
0.0009	0.46	1000	0.0052	0.1708	-19.1382	1.0	19.3091	-291.1548	-134.1098	-2.6678	-2.3707
0.0004	0.69	1500	0.0035	-1.8569	-23.8467	1.0	21.9898	-338.2391	-154.3867	-2.5554	-2.1748
0.0004	0.91	2000	0.0032	-1.1301	-24.4098	1.0	23.2798	-343.8705	-147.1187	-2.4706	-2.1426