GIZ/ADAPMIT-multilabel-bge_f

ADAPMIT-multilabel-bge

This model is a fine-tuned version of BAAI/bge-base-en-v1.5 on the on the Policy-Classification dataset. It achieves the following results on the evaluation set:

• Loss: 0.3101
• Precision-micro: 0.9058
• Precision-samples: 0.8647
• Precision-weighted: 0.9058
• Recall-micro: 0.9305
• Recall-samples: 0.8693
• Recall-weighted: 0.9305
• F1-micro: 0.9180
• F1-samples: 0.8622
• F1-weighted: 0.9180

Model description

The purpose of this model is to predict multiple labels simultaneously from a given input data. Specifically, the model will predict 2 labels - AdaptationLabel, MitigationLabel - that are relevant to a particular task or application

Intended uses & limitations

More information needed

Training and evaluation data

• Training Dataset: 12538

  Class	Positive Count of Class
  AdaptationLabel	5439
  MitigationLabel	6659

• Validation Dataset: 1190

  Class	Positive Count of Class
  AdaptationLabel	533
  MitigationLabel	604

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 4.08e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 300
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss	Precision-micro	Precision-samples	Precision-weighted	Recall-micro	Recall-samples	Recall-weighted	F1-micro	F1-samples	F1-weighted
0.3368	1.0	784	0.2917	0.8651	0.8450	0.8664	0.9138	0.8542	0.9138	0.8888	0.8437	0.8890
0.1807	2.0	1568	0.2549	0.9092	0.8643	0.9094	0.9156	0.8571	0.9156	0.9124	0.8571	0.9123
0.0955	3.0	2352	0.2988	0.9069	0.8660	0.9072	0.9252	0.8655	0.9252	0.9160	0.8613	0.9160
0.0495	4.0	3136	0.3101	0.9058	0.8647	0.9058	0.9305	0.8693	0.9305	0.9180	0.8622	0.9180
label	precision	recall	f1-score	support
:-------------:	:---------:	:-----:	:------:	:------:
AdaptationLabel	0.910	0.928	0.919	533.0
MitigationLabel	0.902	0.932	0.917	604.0

Environmental Impact

Carbon emissions were measured using CodeCarbon.

• Carbon Emitted: 0.04051 kg of CO2
• Hours Used: 0.994 hours

Training Hardware

• On Cloud: yes
• GPU Model: 1 x Tesla T4
• CPU Model: Intel(R) Xeon(R) CPU @ 2.00GHz
• RAM Size: 12.67 GB

Framework versions

• Transformers 4.38.1
• Pytorch 2.1.0+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2