Llama2-jp-123M

This model was built by referring to the config in TinyLlama/TinyLlama-1.1B-Chat-v1.0, 123M The model is a pre-trained Llama2 of Parameters with only 1 epoch on a Japanese dataset. The dataset used is range3/wiki40b-ja.

• Loss: 2.6296

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: 0.0005
• train_batch_size: 156
• eval_batch_size: 156
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
4.4744	0.04	1000	3.5915
3.4036	0.08	2000	3.2836
3.2088	0.12	3000	3.1575
3.1151	0.16	4000	3.0811
3.0471	0.2	5000	3.0247
3.0006	0.24	6000	2.9798
2.9581	0.28	7000	2.9409
2.9187	0.32	8000	2.9103
2.8899	0.36	9000	2.8805
2.8603	0.4	10000	2.8523
2.8329	0.44	11000	2.8259
2.8068	0.48	12000	2.8011
2.7825	0.52	13000	2.7782
2.7601	0.56	14000	2.7567
2.7384	0.6	15000	2.7362
2.7194	0.64	16000	2.7164
2.6986	0.68	17000	2.6973
2.683	0.72	18000	2.6810
2.667	0.76	19000	2.6664
2.6511	0.8	20000	2.6540
2.6421	0.84	21000	2.6441
2.6312	0.88	22000	2.6369
2.6286	0.92	23000	2.6322
2.6234	0.96	24000	2.6301
2.6216	1.0	25000	2.6296

Framework versions

• Transformers 4.38.2
• Pytorch 2.1.0+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2