llama-7b-finnish-instruct-v0.2_En__size_52_epochs_10_2024-06-21_06-03-14_3556408

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: 0.5385
• Accuracy: 0.577
• Chrf: 0.668
• Bleu: 0.528
• Sacrebleu: 0.5
• Rouge1: 0.582
• Rouge2: 0.356
• Rougel: 0.564
• Rougelsum: 0.562
• Meteor: 0.471

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: 52
• training_steps: 520

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Chrf	Bleu	Sacrebleu	Rouge1	Rouge2	Rougel	Rougelsum	Meteor
0.3836	4.0	52	1.1835	0.578	0.44	0.258	0.3	0.345	0.154	0.305	0.303	0.327
1.3385	8.0	104	0.9983	0.578	0.493	0.309	0.3	0.34	0.183	0.328	0.327	0.34
0.4067	12.0	156	0.8706	0.578	0.522	0.378	0.4	0.411	0.228	0.387	0.387	0.402
0.3689	16.0	208	0.7509	0.576	0.586	0.417	0.4	0.476	0.254	0.459	0.454	0.406
0.3043	20.0	260	0.7277	0.578	0.586	0.42	0.4	0.479	0.253	0.451	0.454	0.409
0.7212	24.0	312	0.6785	0.576	0.62	0.471	0.5	0.528	0.312	0.508	0.502	0.431
0.0984	28.0	364	0.6221	0.575	0.626	0.474	0.5	0.551	0.303	0.518	0.519	0.436
0.57	32.0	416	0.6060	0.571	0.657	0.503	0.5	0.551	0.32	0.53	0.522	0.434
0.0689	36.0	468	0.5514	0.577	0.657	0.514	0.5	0.57	0.34	0.55	0.534	0.459
0.1901	40.0	520	0.5385	0.577	0.668	0.528	0.5	0.582	0.356	0.564	0.562	0.471

Framework versions

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