CodeLlama-13b-Instruct-hf_En__components_size_252_epochs_10_2024-06-21_16-38-40_3556558

This model is a fine-tuned version of codellama/CodeLlama-13b-Instruct-hf on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.9627
• Accuracy: 0.475
• Chrf: 0.304
• Bleu: 0.214
• Sacrebleu: 0.2
• Rouge1: 0.462
• Rouge2: 0.235
• Rougel: 0.433
• Rougelsum: 0.455
• Meteor: 0.515

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
0.0529	4.0	252	3.7602	0.473	0.065	0.0	0.0	0.043	0.0	0.037	0.041	0.197
0.0603	8.0	504	2.7810	0.481	0.17	0.117	0.1	0.299	0.204	0.292	0.295	0.44
0.076	12.0	756	2.6938	0.474	0.187	0.115	0.1	0.298	0.202	0.292	0.296	0.41
1.4593	16.0	1008	2.7991	0.48	0.187	0.084	0.1	0.322	0.185	0.316	0.321	0.345
0.1332	20.0	1260	2.4287	0.475	0.237	0.146	0.1	0.344	0.19	0.335	0.344	0.425
0.3344	24.0	1512	2.2035	0.477	0.26	0.158	0.2	0.401	0.196	0.388	0.401	0.43
0.019	28.0	1764	2.1072	0.474	0.272	0.145	0.1	0.372	0.198	0.352	0.367	0.46
0.1253	32.0	2016	2.0500	0.475	0.293	0.209	0.2	0.448	0.241	0.426	0.448	0.516
0.0306	36.0	2268	1.9983	0.475	0.298	0.196	0.2	0.426	0.22	0.405	0.419	0.511
0.0274	40.0	2520	1.9627	0.475	0.304	0.214	0.2	0.462	0.235	0.433	0.455	0.515

Framework versions

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