lmind_hotpot_train8000_eval7405_v1_doc_qa_Qwen_Qwen1.5-4B_lora2

This model is a fine-tuned version of Qwen/Qwen1.5-4B on the tyzhu/lmind_hotpot_train8000_eval7405_v1_doc_qa dataset. It achieves the following results on the evaluation set:

• Loss: 3.6298
• Accuracy: 0.5108

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: 0.0001
• 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	Accuracy	Validation Loss
1.7619	0.9998	1089	0.5172	2.2994
1.648	1.9995	2178	0.5210	2.2683
1.4941	2.9993	3267	0.5214	2.3185
1.3627	4.0	4357	0.5190	2.4249
1.2234	4.9998	5446	0.5152	2.5963
1.1107	5.9995	6535	0.5130	2.7933
0.9891	6.9993	7624	0.5119	2.9422
0.919	8.0	8714	0.5077	3.1141
0.833	8.9998	9803	0.5084	3.1755
0.7635	9.9977	10890	0.5085	3.3117
0.6899	10.9998	11979	3.3147	0.5072
0.6427	11.9995	13068	3.4025	0.5101
0.604	12.9993	14157	3.3905	0.5103
0.5507	14.0	15247	3.4740	0.5088
0.5099	14.9998	16336	3.4772	0.5085
0.478	15.9995	17425	3.5259	0.5088
0.4545	16.9993	18514	3.5391	0.5094
0.427	18.0	19604	3.5887	0.5095
0.4083	18.9998	20693	3.5945	0.5097
0.3818	19.9977	21780	3.6298	0.5108

Framework versions

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