CodeLlama-13b-Instruct-hf_En__size_52_epochs_10_2024-06-21_06-37-42_3556410

This model is a fine-tuned version of codellama/CodeLlama-13b-Instruct-hf on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.6986
• Accuracy: 0.052
• Chrf: 0.682
• Bleu: 0.599
• Sacrebleu: 0.6
• Rouge1: 0.651
• Rouge2: 0.434
• Rougel: 0.601
• Rougelsum: 0.644
• Meteor: 0.54

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
0.5377	4.0	52	1.6208	0.06	0.637	0.525	0.5	0.619	0.387	0.564	0.608	0.488
0.932	8.0	104	2.1202	0.05	0.554	0.452	0.5	0.569	0.313	0.515	0.566	0.485
0.3679	12.0	156	1.9634	0.049	0.606	0.488	0.5	0.594	0.371	0.549	0.59	0.491
0.3454	16.0	208	1.9613	0.053	0.601	0.487	0.5	0.571	0.325	0.524	0.566	0.504
0.3294	20.0	260	1.8641	0.05	0.638	0.536	0.5	0.611	0.388	0.568	0.604	0.516
0.5272	24.0	312	1.8354	0.052	0.644	0.535	0.5	0.609	0.368	0.559	0.603	0.531
0.1871	28.0	364	1.7705	0.054	0.659	0.568	0.6	0.627	0.41	0.586	0.624	0.54
0.4867	32.0	416	1.7689	0.052	0.665	0.571	0.6	0.63	0.406	0.579	0.624	0.562
0.1634	36.0	468	1.6964	0.052	0.682	0.601	0.6	0.658	0.444	0.605	0.65	0.538
0.3475	40.0	520	1.6986	0.052	0.682	0.599	0.6	0.651	0.434	0.601	0.644	0.54

Framework versions

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