CodeQwen1.5-7B-Chat_Finetuned_dataset_size_52_epochs_10_2024-06-08_19-19-20

This model is a fine-tuned version of Qwen/CodeQwen1.5-7B-Chat on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.7232
• Accuracy: 0.075
• Chrf: 0.862
• Bleu: 0.791
• Sacrebleu: 0.8
• Rouge1: 0.832
• Rouge2: 0.698
• Rougel: 0.805
• Rougelsum: 0.819
• Meteor: 0.755

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
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Chrf	Bleu	Sacrebleu	Rouge1	Rouge2	Rougel	Rougelsum	Meteor
0.068	4.0	52	0.5372	0.073	0.834	0.754	0.8	0.809	0.657	0.761	0.795	0.729
0.4041	8.0	104	0.4620	0.073	0.85	0.782	0.8	0.829	0.691	0.79	0.817	0.751
0.0639	12.0	156	0.4250	0.074	0.855	0.784	0.8	0.822	0.684	0.791	0.812	0.74
0.0439	16.0	208	0.4491	0.075	0.862	0.791	0.8	0.832	0.698	0.802	0.819	0.744
0.029	20.0	260	0.4818	0.075	0.862	0.797	0.8	0.84	0.711	0.815	0.828	0.742
0.1251	24.0	312	0.5486	0.075	0.86	0.791	0.8	0.836	0.703	0.81	0.823	0.721
0.0044	28.0	364	0.5942	0.075	0.868	0.797	0.8	0.844	0.713	0.818	0.834	0.73
0.092	32.0	416	0.6645	0.075	0.866	0.802	0.8	0.839	0.71	0.812	0.827	0.756
0.0011	36.0	468	0.7190	0.075	0.859	0.792	0.8	0.83	0.694	0.806	0.821	0.748
0.0021	40.0	520	0.7232	0.075	0.862	0.791	0.8	0.832	0.698	0.805	0.819	0.755

Framework versions

• PEFT 0.7.1
• Transformers 4.37.0
• Pytorch 2.2.0+cu121
• Datasets 2.19.2
• Tokenizers 0.15.2