lmind_nq_train6000_eval6489_v1_reciteonly_qa_v3_Qwen_Qwen1.5-4B_5e-4_lora2

This model is a fine-tuned version of Qwen/Qwen1.5-4B on the tyzhu/lmind_nq_train6000_eval6489_v1_reciteonly_qa_v3 dataset. It achieves the following results on the evaluation set:

• Loss: 2.8927
• Accuracy: 0.5834

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: 0.0005
• 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
1.745	0.9973	187	1.6686	0.6096
1.4255	2.0	375	1.6989	0.6087
1.0295	2.9973	562	1.8159	0.6052
0.6982	4.0	750	1.9994	0.5996
0.4584	4.9973	937	2.2058	0.5949
0.3048	6.0	1125	2.3636	0.5928
0.2175	6.9973	1312	2.5218	0.5905
0.1719	8.0	1500	2.6292	0.5909
0.1355	8.9973	1687	2.7028	0.5892
0.1302	10.0	1875	2.7502	0.5876
0.1281	10.9973	2062	2.7715	0.5875
0.1331	12.0	2250	2.7947	0.5862
0.1435	12.9973	2437	2.7769	0.5867
0.1468	14.0	2625	2.7598	0.5862
0.1449	14.9973	2812	2.8147	0.5855
0.1404	16.0	3000	2.8564	0.5859
0.121	16.9973	3187	2.8381	0.5863
0.1198	18.0	3375	2.8844	0.5841
0.1198	18.9973	3562	2.9040	0.5834
0.1215	19.9467	3740	2.8927	0.5834

Framework versions

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