CodeLlama-13b-Instruct-hf_Fi__translations_size_104_epochs_10_2024-06-22_02-50-05_3557641

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: 1.4179
• Accuracy: 0.035
• Chrf: 0.719
• Bleu: 0.636
• Sacrebleu: 0.6
• Rouge1: 0.678
• Rouge2: 0.497
• Rougel: 0.645
• Rougelsum: 0.675
• Meteor: 0.594

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
0.3439	4.0	104	2.0524	0.036	0.581	0.472	0.5	0.556	0.309	0.5	0.55	0.493
0.194	8.0	208	2.0784	0.036	0.55	0.435	0.4	0.519	0.275	0.479	0.513	0.466
0.2001	12.0	312	2.1078	0.061	0.564	0.456	0.5	0.544	0.291	0.504	0.538	0.502
0.5322	16.0	416	1.7052	0.037	0.64	0.527	0.5	0.6	0.366	0.563	0.598	0.524
0.1677	20.0	520	1.8442	0.037	0.606	0.493	0.5	0.566	0.334	0.529	0.557	0.496
0.1649	24.0	624	1.6364	0.037	0.66	0.558	0.6	0.627	0.408	0.593	0.621	0.535
0.6799	28.0	728	1.5108	0.034	0.692	0.597	0.6	0.647	0.452	0.608	0.644	0.541
0.1644	32.0	832	1.4941	0.035	0.709	0.623	0.6	0.671	0.478	0.634	0.667	0.553
0.2382	36.0	936	1.4313	0.035	0.715	0.633	0.6	0.681	0.494	0.645	0.679	0.556
0.1956	40.0	1040	1.4179	0.035	0.719	0.636	0.6	0.678	0.497	0.645	0.675	0.594

Framework versions

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