zephyr-7b-sft-full

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the generator dataset. It achieves the following results on the evaluation set:

• Loss: 1.0243

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: 8
• eval_batch_size: 16
• seed: 42
• distributed_type: multi-GPU
• num_devices: 16
• total_train_batch_size: 128
• total_eval_batch_size: 256
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
0.9838	0.1845	200	0.9937
1.0162	0.3690	400	1.0329
1.0095	0.5535	600	1.0302
0.9857	0.7380	800	1.0204
0.9803	0.9225	1000	1.0051
0.736	1.1070	1200	1.0061
0.7249	1.2915	1400	1.0004
0.7355	1.4760	1600	0.9855
0.7151	1.6605	1800	0.9713
0.7023	1.8450	2000	0.9557
0.3925	2.0295	2200	1.0150
0.3871	2.2140	2400	1.0319
0.3927	2.3985	2600	1.0269
0.3872	2.5830	2800	1.0267
0.3918	2.7675	3000	1.0242
0.3764	2.9520	3200	1.0243

Framework versions

• Transformers 4.51.3
• Pytorch 2.5.1+rocm6.2
• Datasets 3.5.0
• Tokenizers 0.21.1