lmind_nq_train6000_eval6489_v1_recite_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_recite_qa_v3 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4565
• Accuracy: 0.7919

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
1.7331	1.0	529	1.4271	0.6365
1.3037	2.0	1058	1.0687	0.6846
0.8818	3.0	1587	0.8142	0.7216
0.6397	4.0	2116	0.6636	0.7470
0.4735	5.0	2645	0.5547	0.7667
0.3798	6.0	3174	0.5002	0.7764
0.3409	7.0	3703	0.4850	0.7801
0.3054	8.0	4232	0.4691	0.7835
0.2803	9.0	4761	0.4637	0.7859
0.2637	10.0	5290	0.4532	0.7877
0.2661	11.0	5819	0.4668	0.7879
0.2513	12.0	6348	0.4647	0.7893
0.2424	13.0	6877	0.4615	0.7897
0.2499	14.0	7406	0.4546	0.7894
0.235	15.0	7935	0.4668	0.7896
0.2317	16.0	8464	0.4510	0.7913
0.2225	17.0	8993	0.4497	0.7915
0.2358	18.0	9522	0.4475	0.7916
0.2253	19.0	10051	0.4529	0.7918
0.2172	20.0	10580	0.4565	0.7919

Framework versions

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