匿名用户2024年07月31日
17阅读
开发技术Tensorflow
所属分类ai、generated_from_train、lora、llama-factory
开源地址https://modelscope.cn/models/Tianmasikong/JDIGSTEST1
授权协议other

作品详情

legalinstrumenttrain

This model is a fine-tuned version of ./Llama-2-7b-hf on the criminal1_10k dataset. It achieves the following results on the evaluation set:

  • Loss: 0.3369

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: 5e-05
  • trainbatchsize: 2
  • evalbatchsize: 2
  • seed: 42
  • gradientaccumulationsteps: 2
  • totaltrainbatch_size: 4
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lrschedulertype: cosine
  • lrschedulerwarmup_steps: 20
  • num_epochs: 3.0
  • mixedprecisiontraining: Native AMP

Training results

Training Loss Epoch Step Validation Loss
0.6239 0.15 100 0.6105
0.4555 0.3 200 0.4880
0.4614 0.44 300 0.4438
0.3975 0.59 400 0.4208
0.3747 0.74 500 0.3984
0.4318 0.89 600 0.3888
0.3629 1.04 700 0.3766
0.3729 1.19 800 0.3685
0.3675 1.33 900 0.3632
0.4056 1.48 1000 0.3570
0.3222 1.63 1100 0.3522
0.2821 1.78 1200 0.3489
0.3431 1.93 1300 0.3448
0.2885 2.07 1400 0.3429
0.262 2.22 1500 0.3413
0.3168 2.37 1600 0.3394
0.3183 2.52 1700 0.3380
0.3021 2.67 1800 0.3372
0.2748 2.81 1900 0.3369
0.3175 2.96 2000 0.3369

Framework versions

  • PEFT 0.10.0
  • Transformers 4.39.2
  • Pytorch 2.2.2+cu121
  • Datasets 2.18.0
  • Tokenizers 0.15.2
声明:本文仅代表作者观点,不代表本站立场。如果侵犯到您的合法权益,请联系我们删除侵权资源!如果遇到资源链接失效,请您通过评论或工单的方式通知管理员。未经允许,不得转载,本站所有资源文章禁止商业使用运营!
下载安装【程序员客栈】APP
实时对接需求、及时收发消息、丰富的开放项目需求、随时随地查看项目状态

评论