lmind_nq_train6000_eval6489_v1_recite_qa_v3_meta-llama_Llama-2-7b-hf_3e-5_lora2

This model is a fine-tuned version of meta-llama/Llama-2-7b-hf on the tyzhu/lmind_nq_train6000_eval6489_v1_recite_qa_v3 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7173
• Accuracy: 0.7434

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: 3e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• 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
1.3186	1.0	529	0.6640	1.2075
1.2869	2.0	1058	0.6686	1.1716
1.2221	3.0	1587	0.6729	1.1375
1.1692	4.0	2116	0.6770	1.1135
1.1179	5.0	2645	0.6808	1.0888
1.0551	6.0	3174	0.6846	1.0604
1.0184	7.0	3703	0.6893	1.0354
0.9519	8.0	4232	0.6929	1.0097
0.8969	9.0	4761	0.6963	0.9885
0.8428	10.0	5290	0.6993	0.9753
0.7945	11.0	5819	0.7055	0.9353
0.7459	12.0	6348	0.7097	0.9101
0.6852	13.0	6877	0.7145	0.8795
0.6452	14.0	7406	0.7182	0.8614
0.602	15.0	7935	0.7230	0.8327
0.5465	16.0	8464	0.7273	0.8036
0.5233	17.0	8993	0.7305	0.7916
0.4896	18.0	9522	0.7641	0.7349
0.4549	19.0	10051	0.7421	0.7390
0.42	20.0	10580	0.7173	0.7434

Framework versions

• Transformers 4.34.0
• Pytorch 2.1.0+cu121
• Datasets 2.18.0
• Tokenizers 0.14.1