This model is a fine-tuned version of google/gemma-7b on the HuggingFaceH4/OpenHermes-2.5-1k-longest dataset. It achieves the following results on the evaluation set:

Loss: 1.7445

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-06
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
num_epochs: 15

Training results

Training Loss	Epoch	Step	Validation Loss
5.6993	1.0	6	2.8191
3.3379	2.0	12	2.2503
2.8978	3.0	18	2.0730
2.7495	4.0	24	1.9771
2.5265	5.0	30	1.9129
2.4727	6.0	36	1.8681
2.443	7.0	42	1.8344
2.3432	8.0	48	1.8083
2.3291	9.0	54	1.7878
2.2843	10.0	60	1.7719
2.2529	11.0	66	1.7595
2.2723	12.0	72	1.7509
2.2302	13.0	78	1.7465
2.2224	14.0	84	1.7448
2.2309	15.0	90	1.7445

Framework versions

Transformers 4.39.0.dev0
Pytorch 2.1.2+cu121
Datasets 2.14.6
Tokenizers 0.15.1

lewtun/gemma-7b-sft-full-longest-1k-v0

gemma-7b-sft-full-longest-1k-v0

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

lewtun/
gemma-7b-sft-full-longest-1k-v0