lmind_nq_train6000_eval6489_v1_recite_qa_v3_Qwen_Qwen1.5-4B_5e-5_lora2

This model is a fine-tuned version of Qwen/Qwen1.5-4B on the tyzhu/lmind_nq_train6000_eval6489_v1_recite_qa_v3 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5804
• Accuracy: 0.7754

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: 5e-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
1.8478	1.0	529	1.6699	0.6080
1.7862	2.0	1058	1.6003	0.6164
1.6531	3.0	1587	1.5363	0.6251
1.5515	4.0	2116	1.4608	0.6343
1.4038	5.0	2645	1.3876	0.6456
1.2751	6.0	3174	1.3186	0.6553
1.1475	7.0	3703	1.2514	0.6637
1.0282	8.0	4232	1.1740	0.676
0.9067	9.0	4761	1.1004	0.6870
0.8202	10.0	5290	1.0408	0.6964
0.7007	11.0	5819	0.9592	0.7084
0.6259	12.0	6348	0.8998	0.7191
0.553	13.0	6877	0.8332	0.7295
0.4948	14.0	7406	0.7799	0.7387
0.4221	15.0	7935	0.7330	0.7466
0.3911	16.0	8464	0.6805	0.7551
0.3377	17.0	8993	0.6475	0.7620
0.3179	18.0	9522	0.6195	0.7680
0.288	19.0	10051	0.5962	0.7723
0.2605	20.0	10580	0.5804	0.7754

Framework versions

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