lmind_hotpot_train8000_eval7405_v1_recite_qa_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_hotpot_train8000_eval7405_v1_recite_qa dataset. It achieves the following results on the evaluation set:

• Loss: 0.6450
• Accuracy: 0.7555

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	Validation Loss	Accuracy
1.0799	1.0	1089	1.0260	0.7169
1.0425	2.0	2178	0.9992	0.7192
1.0165	3.0	3267	0.9771	0.7211
0.9772	4.0	4357	0.9578	0.7230
0.9463	5.0	5446	0.9372	0.7247
0.9022	6.0	6535	0.9148	0.7269
0.8621	7.0	7624	0.8946	0.7289
0.8297	8.0	8714	0.8746	0.7310
0.7947	9.0	9803	0.8553	0.7329
0.7436	10.0	10892	0.8319	0.7350
0.7032	11.0	11981	0.8134	0.7371
0.6781	12.0	13071	0.7954	0.7388
0.6416	13.0	14160	0.7733	0.7411
0.5968	14.0	15249	0.7518	0.7434
0.5642	15.0	16338	0.7366	0.7453
0.5442	16.0	17428	0.7135	0.7475
0.5038	17.0	18517	0.6964	0.7494
0.4795	18.0	19606	0.6801	0.7515
0.4508	19.0	20695	0.6601	0.7537
0.42	20.0	21780	0.6450	0.7555

Framework versions

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