SiliangZ/IRL_Iter0_Policy_Epoch5_RM_Data_SPIN_Iter0

IRL_iter0_best_of_16_spin_iter0_epoch_5_saving

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the d, the a, the t, the a, the _, the g, the e, the n, the e, the r, the a, the t, the e, the d, the /, the s, the p, the i, the n, the _, the i, the t, the e, the r, the 0, the _, the b, the e, the s, the t, the _, the o, the f, the _, the 1, the 6, the /, the t, the o, the p, the 1, the _, the s, the e, the l, the e, the c, the t, the e, the d, the _, the I, the R, the L, the _, the r, the e, the w, the a, the r, the d, the _, the s, the e, the l, the e, the c, the t, the e and the d datasets. It achieves the following results on the evaluation set:

• Loss: 0.0396

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: 2e-05
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 5.0

Training results

Training Loss	Epoch	Step	Validation Loss
1.0895	1.0	79	0.7223
0.6454	2.0	158	0.3317
0.2926	3.0	237	0.1293
0.1048	4.0	316	0.0542
0.0465	5.0	395	0.0396

Framework versions

• Transformers 4.44.2
• Pytorch 2.1.2+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1