lmind_hotpot_train8000_eval7405_v1_reciteonly_qa_Qwen_Qwen1.5-4B_3e-4_lora2

This model is a fine-tuned version of Qwen/Qwen1.5-4B on the tyzhu/lmind_hotpot_train8000_eval7405_v1_reciteonly_qa dataset. It achieves the following results on the evaluation set:

• Loss: 2.7813
• Accuracy: 0.6609

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
1.4488	1.0	250	1.4958	0.6770
1.3142	2.0	500	1.5007	0.6772
1.1176	3.0	750	1.5507	0.6756
0.9253	4.0	1000	1.6442	0.6728
0.7213	5.0	1250	1.7736	0.6701
0.5718	6.0	1500	1.8863	0.6682
0.4232	7.0	1750	2.0245	0.6660
0.3334	8.0	2000	2.1773	0.6642
0.2433	9.0	2250	2.2681	0.6632
0.2076	10.0	2500	2.3732	0.6629
0.1632	11.0	2750	2.4368	0.6623
0.1491	12.0	3000	2.5182	0.6617
0.1275	13.0	3250	2.5680	0.6619
0.1273	14.0	3500	2.6412	0.6613
0.1129	15.0	3750	2.6497	0.6617
0.1129	16.0	4000	2.6932	0.6614
0.102	17.0	4250	2.7003	0.6612
0.1109	18.0	4500	2.7033	0.6614
0.0997	19.0	4750	2.7139	0.6613
0.1012	20.0	5000	2.7813	0.6609

Framework versions

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