BACnet-Klassifizierung-Gewerke-bert-base-german-cased

This model is a fine-tuned version of bert-base-german-cased on the gart-labor "klassifizierung_gewerke" dataset. It achieves the following results on the evaluation set:

Loss: 0.0394
F1: [0.96296296 0.8 0.97297297 1. 0.99469027 0.98979592 0.98969072]

Model description

This model makes it possible to classify the components of the technical building equipment described with the BACnet standard into different trades. The model is based on a German-language data set.

Intended uses & limitations

The model divides descriptive texts into the following building services trades: Waste_water_water_gas_systems, Other_systems, Building_automation, Refrigeration_systems, Air_technical_systems, Heavy_current_systems and Heat_supply_systems

Training and evaluation data

The model is based on a German-language data set.

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-05
train_batch_size: 8
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 16
total_train_batch_size: 128
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 5.0

Training results

Training Loss	Epoch	Step	Validation Loss	F1
0.4309	0.99	45	0.0736	[0.89655172 0.84210526 0.97297297 0.98901099 0.9929078 0.99492386 0.98701299]
0.0722	1.99	90	0.0511	[0.92307692 0.875 0.96 1. 0.99295775 0.98979592 0.98714653]
0.0431	2.99	135	0.0460	[1. 0.8 0.97297297 1. 0.99469027 0.98979592 0.99224806]
0.0313	3.99	180	0.0365	[1. 0.84210526 0.97297297 1. 0.99646643 0.98979592 0.99224806]
0.0238	4.99	225	0.0394	[0.96296296 0.8 0.97297297 1. 0.99469027 0.98979592 0.98969072]

Framework versions

Transformers 4.21.1
Pytorch 1.12.0+cu113
Datasets 2.4.0
Tokenizers 0.12.1