Artigenz-Coder-DS-6.7B_En__translations_size_104_epochs_10_2024-06-22_03-26-15_3557997

This model is a fine-tuned version of Artigenz/Artigenz-Coder-DS-6.7B on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 3.1141
• Accuracy: 0.06
• Chrf: 0.499
• Bleu: 0.407
• Sacrebleu: 0.4
• Rouge1: 0.494
• Rouge2: 0.242
• Rougel: 0.449
• Rougelsum: 0.488
• Meteor: 0.401

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
0.1365	4.0	104	1.1838	0.046	0.714	0.6	0.6	0.676	0.459	0.613	0.668	0.522
0.1026	8.0	208	1.3421	0.045	0.699	0.569	0.6	0.66	0.437	0.601	0.648	0.482
0.1001	12.0	312	1.3957	0.047	0.724	0.621	0.6	0.701	0.482	0.63	0.685	0.528
0.4589	16.0	416	1.6948	0.046	0.702	0.601	0.6	0.694	0.473	0.62	0.681	0.51
0.1812	20.0	520	2.5671	0.077	0.59	0.47	0.5	0.605	0.346	0.526	0.591	0.403
0.1966	24.0	624	2.5118	0.066	0.607	0.502	0.5	0.607	0.357	0.544	0.601	0.428
0.9528	28.0	728	2.7303	0.055	0.567	0.465	0.5	0.577	0.325	0.52	0.567	0.429
0.2147	32.0	832	2.9680	0.055	0.529	0.435	0.4	0.541	0.285	0.489	0.533	0.402
0.367	36.0	936	3.1490	0.067	0.508	0.417	0.4	0.516	0.264	0.469	0.509	0.392
0.2157	40.0	1040	3.1141	0.06	0.499	0.407	0.4	0.494	0.242	0.449	0.488	0.401

Framework versions

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