lmind_hotpot_train8000_eval7405_v1_reciteonly_qa_meta-llama_Llama-2-7b-hf_5e-4_lora2

This model is a fine-tuned version of meta-llama/Llama-2-7b-hf on the tyzhu/lmind_hotpot_train8000_eval7405_v1_reciteonly_qa dataset. It achieves the following results on the evaluation set:

• Loss: 8.0667
• Accuracy: 0.4450

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: 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	Validation Loss	Accuracy
1.0076	1.0	250	1.0128	0.7184
0.8686	2.0	500	1.0307	0.7170
0.6975	3.0	750	1.1069	0.7135
0.5421	4.0	1000	1.1970	0.7097
0.386	5.0	1250	1.3194	0.7065
0.2836	6.0	1500	1.4507	0.7033
0.1939	7.0	1750	1.5671	0.7020
0.1547	8.0	2000	1.6786	0.7007
0.1181	9.0	2250	1.7524	0.7000
0.1093	10.0	2500	1.7871	0.6994
0.0917	11.0	2750	1.8465	0.6995
0.0937	12.0	3000	1.8788	0.6992
0.0854	13.0	3250	1.9002	0.6985
0.9593	14.0	3500	5.8170	0.4891
1.8603	15.0	3750	2.0887	0.6453
7.7554	16.0	4000	6.9929	0.4434
7.3836	17.0	4250	8.0698	0.4387
7.044	18.0	4500	6.9576	0.4427
6.8665	19.0	4750	7.4200	0.4471
7.9799	20.0	5000	8.0667	0.4450

Framework versions

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