llama-7b-finnish-instruct-v0.2_En__translations_size_104_epochs_10_2024-06-22_03-40-15_3557998

This model is a fine-tuned version of Finnish-NLP/llama-7b-finnish-instruct-v0.2 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5080
• Accuracy: 0.574
• Chrf: 0.692
• Bleu: 0.544
• Sacrebleu: 0.5
• Rouge1: 0.598
• Rouge2: 0.374
• Rougel: 0.581
• Rougelsum: 0.585
• Meteor: 0.465

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: 104
• training_steps: 1040

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Chrf	Bleu	Sacrebleu	Rouge1	Rouge2	Rougel	Rougelsum	Meteor
0.4205	4.0	104	1.0350	0.577	0.44	0.277	0.3	0.317	0.149	0.288	0.297	0.355
0.2068	8.0	208	0.9393	0.578	0.501	0.338	0.3	0.391	0.206	0.357	0.353	0.362
0.1289	12.0	312	0.8516	0.575	0.548	0.38	0.4	0.439	0.225	0.414	0.416	0.395
0.7509	16.0	416	0.7754	0.578	0.578	0.42	0.4	0.461	0.259	0.441	0.44	0.431
0.0951	20.0	520	0.7348	0.574	0.606	0.459	0.5	0.513	0.291	0.484	0.483	0.446
0.1201	24.0	624	0.6261	0.577	0.637	0.484	0.5	0.541	0.321	0.516	0.505	0.437
1.0287	28.0	728	0.5589	0.574	0.668	0.51	0.5	0.559	0.334	0.535	0.51	0.458
0.0522	32.0	832	0.5623	0.574	0.659	0.526	0.5	0.573	0.349	0.552	0.541	0.449
0.1314	36.0	936	0.5154	0.576	0.682	0.534	0.5	0.588	0.36	0.573	0.576	0.465
0.0449	40.0	1040	0.5080	0.574	0.692	0.544	0.5	0.598	0.374	0.581	0.585	0.465

Framework versions

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