deepseek-coder-6.7b-instruct_En__size_52_epochs_10_2024-06-21_06-20-33_3556409

This model is a fine-tuned version of deepseek-ai/deepseek-coder-6.7b-instruct on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.4340
• Accuracy: 0.042
• Chrf: 0.734
• Bleu: 0.608
• Sacrebleu: 0.6
• Rouge1: 0.707
• Rouge2: 0.494
• Rougel: 0.637
• Rougelsum: 0.693
• Meteor: 0.534

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
0.1233	4.0	52	1.1674	0.027	0.726	0.601	0.6	0.681	0.458	0.612	0.674	0.539
0.5834	8.0	104	1.2639	0.032	0.708	0.57	0.6	0.686	0.458	0.617	0.679	0.483
0.1938	12.0	156	1.2723	0.032	0.708	0.574	0.6	0.684	0.457	0.609	0.673	0.479
0.1681	16.0	208	1.2437	0.036	0.719	0.595	0.6	0.697	0.469	0.619	0.682	0.524
0.176	20.0	260	1.4102	0.037	0.699	0.565	0.6	0.666	0.435	0.588	0.652	0.507
0.4563	24.0	312	1.3416	0.039	0.717	0.586	0.6	0.69	0.452	0.609	0.678	0.521
0.114	28.0	364	1.3758	0.041	0.728	0.602	0.6	0.703	0.478	0.618	0.683	0.524
0.4204	32.0	416	1.4116	0.042	0.727	0.598	0.6	0.705	0.476	0.621	0.689	0.545
0.1118	36.0	468	1.4229	0.042	0.734	0.607	0.6	0.709	0.497	0.64	0.694	0.528
0.2482	40.0	520	1.4340	0.042	0.734	0.608	0.6	0.707	0.494	0.637	0.693	0.534

Framework versions

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