metadata

tags:
  - token-classification
  - named-enity-recognition
datasets:
  - DFKI-SLT/few-nerd
model-index:
  - name: span-marker-robert-base
    results: []
license: apache-2.0
language:
  - en
library_name: span-marker
pipeline_tag: token-classification

span-marker-robert-base

This model is a fine-tuned version of roberta-base on few-nerd dataset using SpanMarker an module for NER.

Usage

  from span_marker import SpanMarkerModel
  
  model = SpanMarkerModel.from_pretrained("krinal/span-marker-robert-base")
  
  ner_result = model.predict("Argentine captain Lionel Messi won Golden Ball at FIFA world cup 2022")

Training and evaluation data

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-05
train_batch_size: 4
eval_batch_size: 4
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 8
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.1
num_epochs: 1

Evaluation

It achieves the following results on the evaluation set:

Loss: 0.0214
Overall Precision: 0.7642
Overall Recall: 0.7947
Overall F1: 0.7791
Overall Accuracy: 0.9397

Training results

Training Loss	Epoch	Step	Validation Loss	Overall Precision	Overall Recall	Overall F1	Overall Accuracy
0.0214	0.08	100	0.0219	0.7641	0.7679	0.7660	0.9330
0.0199	0.16	200	0.0243	0.7442	0.7679	0.7559	0.9348
0.0179	0.24	300	0.0212	0.7730	0.7580	0.7654	0.9361
0.0188	0.33	400	0.0225	0.7616	0.7710	0.7662	0.9343
0.0149	0.41	500	0.0240	0.7537	0.7783	0.7658	0.9375
0.015	0.49	600	0.0230	0.7540	0.7829	0.7682	0.9362
0.0137	0.57	700	0.0232	0.7746	0.7538	0.7640	0.9319
0.0123	0.65	800	0.0218	0.7651	0.7879	0.7763	0.9393
0.0103	0.73	900	0.0223	0.7688	0.7964	0.7824	0.9397
0.0108	0.82	1000	0.0209	0.7763	0.7816	0.7789	0.9397
0.0116	0.9	1100	0.0213	0.7743	0.7879	0.7811	0.9398
0.0119	0.98	1200	0.0214	0.7653	0.7947	0.7797	0.9400

Framework versions

Transformers 4.30.2
Pytorch 2.0.1+cu118
Datasets 2.13.1
Tokenizers 0.13.3
span-marker 1.2.3