llama-7b-finnish-instruct-v0.2_Fi__components_size_252_epochs_10_2024-06-21_09-35-06_3556544

This model is a fine-tuned version of Finnish-NLP/llama-7b-finnish-instruct-v0.2 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4970
• Accuracy: 0.77
• Chrf: 0.504
• Bleu: 0.417
• Sacrebleu: 0.4
• Rouge1: 0.566
• Rouge2: 0.386
• Rougel: 0.552
• Rougelsum: 0.554
• Meteor: 0.59

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: 0.001
• train_batch_size: 1
• eval_batch_size: 1
• seed: 3407
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 4
• total_eval_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-06
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 252
• training_steps: 2520

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Chrf	Bleu	Sacrebleu	Rouge1	Rouge2	Rougel	Rougelsum	Meteor
0.0297	4.0	252	1.3305	0.775	0.207	0.139	0.1	0.407	0.272	0.4	0.402	0.476
0.0414	8.0	504	0.9221	0.772	0.319	0.23	0.2	0.484	0.333	0.477	0.48	0.563
0.0472	12.0	756	0.8856	0.775	0.364	0.261	0.3	0.496	0.322	0.482	0.487	0.569
1.2941	16.0	1008	0.9528	0.772	0.349	0.265	0.3	0.479	0.327	0.471	0.474	0.552
0.1395	20.0	1260	0.8777	0.771	0.384	0.284	0.3	0.508	0.338	0.495	0.495	0.567
0.3282	24.0	1512	0.7412	0.771	0.403	0.312	0.3	0.514	0.336	0.504	0.508	0.568
0.0135	28.0	1764	0.7096	0.77	0.409	0.309	0.3	0.532	0.367	0.524	0.522	0.573
0.1001	32.0	2016	0.6087	0.77	0.451	0.362	0.4	0.544	0.373	0.529	0.53	0.578
0.0189	36.0	2268	0.5685	0.77	0.458	0.363	0.4	0.535	0.357	0.523	0.525	0.604
0.0168	40.0	2520	0.4970	0.77	0.504	0.417	0.4	0.566	0.386	0.552	0.554	0.59

Framework versions

• Transformers 4.37.0
• Pytorch 2.2.1+cu121
• Datasets 2.20.0
• Tokenizers 0.15.2