suzume-llama-3-8B-financial

This model is a fine-tuned version of lightblue/suzume-llama-3-8B-japanese on the generator dataset. It achieves the following results on the evaluation set:

• Loss: 0.9951

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: 0.0002
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant
• lr_scheduler_warmup_ratio: 0.03
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
2.4249	0.2222	1	1.8524
1.9685	0.4444	2	1.5389
1.72	0.6667	3	1.4369
1.5908	0.8889	4	1.3468
1.4306	1.1111	5	1.2865
1.368	1.3333	6	1.2355
1.2995	1.5556	7	1.1926
1.2591	1.7778	8	1.1429
1.1882	2.0	9	1.1208
1.0992	2.2222	10	1.0954
1.0199	2.4444	11	1.0754
1.0231	2.6667	12	1.0629
0.9314	2.8889	13	1.0317
0.8405	3.1111	14	1.0161
0.7706	3.3333	15	1.0216
0.782	3.5556	16	1.0055
0.8002	3.7778	17	0.9985
0.7849	4.0	18	0.9908
0.6481	4.2222	19	1.0070
0.5567	4.4444	20	0.9951

Framework versions

• PEFT 0.11.0
• Transformers 4.40.2
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1