lmind_nq_train6000_eval6489_v1_doc_qa_v3_Qwen_Qwen1.5-4B_3e-5_lora2

This model is a fine-tuned version of Qwen/Qwen1.5-4B on the tyzhu/lmind_nq_train6000_eval6489_v1_doc_qa_v3 dataset. It achieves the following results on the evaluation set:

• Loss: 2.4758
• Accuracy: 0.5594

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: 3e-05
• train_batch_size: 1
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 20.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
1.8498	1.0	529	1.6299	0.5722
1.821	2.0	1058	1.6039	0.5750
1.7317	3.0	1587	1.6205	0.5749
1.6831	4.0	2116	1.6653	0.5729
1.5981	5.0	2645	1.7394	0.5695
1.5258	6.0	3174	1.8378	0.5663
1.4781	7.0	3703	1.8968	0.5656
1.4482	8.0	4232	2.0232	0.5636
1.4036	9.0	4761	2.0670	0.5609
1.3778	10.0	5290	2.1624	0.562
1.3242	11.0	5819	2.2012	0.5615
1.3135	12.0	6348	2.2194	0.5634
1.2761	13.0	6877	2.3155	0.5607
1.266	14.0	7406	2.3657	0.5605
1.1995	15.0	7935	2.3795	0.5593
1.1944	16.0	8464	2.4200	0.5590
1.1433	17.0	8993	2.4230	0.5598
1.1323	18.0	9522	2.4370	0.5591
1.1049	19.0	10051	2.4679	0.5601
1.056	20.0	10580	2.4758	0.5594

Framework versions

• PEFT 0.5.0
• Transformers 4.40.2
• Pytorch 2.3.0
• Datasets 2.19.1
• Tokenizers 0.19.1