lmind_nq_train6000_eval6489_v1_recite_qa_v3_Qwen_Qwen1.5-4B_3e-4_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: 0.4311
• Accuracy: 0.7981

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: 0.0003
• 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
1.7637	1.0	529	1.4995	0.6288
1.3986	2.0	1058	1.1711	0.6720
0.9515	3.0	1587	0.8766	0.7148
0.642	4.0	2116	0.6720	0.7478
0.4362	5.0	2645	0.5458	0.7697
0.3201	6.0	3174	0.4751	0.7823
0.2652	7.0	3703	0.4510	0.7887
0.2263	8.0	4232	0.4372	0.7914
0.2035	9.0	4761	0.4335	0.7940
0.1913	10.0	5290	0.4322	0.7950
0.188	11.0	5819	0.4379	0.7945
0.1777	12.0	6348	0.4279	0.7957
0.1723	13.0	6877	0.4326	0.7956
0.1767	14.0	7406	0.4329	0.7967
0.1666	15.0	7935	0.4396	0.7962
0.1642	16.0	8464	0.4391	0.7965
0.1575	17.0	8993	0.4405	0.7967
0.1634	18.0	9522	0.4265	0.7976
0.1593	19.0	10051	0.4323	0.7978
0.153	20.0	10580	0.4311	0.7981

Framework versions

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