deepseek-coder-6.7b-instruct_Fi__translations_size_104_epochs_10_2024-06-21_23-44-44_3557640

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: 2.9391
• Accuracy: 0.027
• Chrf: 0.526
• Bleu: 0.433
• Sacrebleu: 0.4
• Rouge1: 0.512
• Rouge2: 0.258
• Rougel: 0.471
• Rougelsum: 0.507
• Meteor: 0.438

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
0.1525	4.0	104	1.1253	0.014	0.725	0.595	0.6	0.702	0.477	0.625	0.693	0.513
0.1027	8.0	208	1.2219	0.014	0.711	0.58	0.6	0.684	0.45	0.611	0.673	0.51
0.1307	12.0	312	1.3506	0.025	0.709	0.606	0.6	0.689	0.472	0.616	0.679	0.545
1.0101	16.0	416	1.6526	0.017	0.706	0.614	0.6	0.696	0.494	0.629	0.687	0.532
0.177	20.0	520	2.0077	0.028	0.663	0.558	0.6	0.659	0.439	0.594	0.653	0.46
0.2181	24.0	624	2.4044	0.02	0.615	0.507	0.5	0.623	0.393	0.565	0.617	0.474
0.9954	28.0	728	2.5812	0.025	0.585	0.483	0.5	0.579	0.329	0.523	0.569	0.452
0.28	32.0	832	2.8603	0.034	0.548	0.462	0.5	0.543	0.299	0.499	0.538	0.424
0.3616	36.0	936	2.9194	0.031	0.532	0.436	0.4	0.522	0.266	0.476	0.516	0.437
0.3022	40.0	1040	2.9391	0.027	0.526	0.433	0.4	0.512	0.258	0.471	0.507	0.438

Framework versions

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