|
--- |
|
tags: |
|
- generated_from_trainer |
|
model-index: |
|
- name: training_bert |
|
results: [] |
|
license: mit |
|
language: |
|
- en |
|
metrics: |
|
- perplexity |
|
--- |
|
|
|
<!-- This model card has been generated automatically according to the information the Trainer had access to. You |
|
should probably proofread and complete it, then remove this comment. --> |
|
|
|
# training_bert |
|
|
|
This model is a fine-tuned version of [Bert Base Uncased](https://huggingface.co/) on dataset composed of different jobs posted in several job platforms and thousands of resumes. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 4.0495 |
|
|
|
## Model description |
|
|
|
Pretraining done on bert base architecture. |
|
|
|
## Intended uses & limitations |
|
|
|
This model can be used to generate contextual embeddings for textual data used in Applicant Tracking Systems such as resumes, jobs and cover letters. |
|
The embeddings can be further used to perform other NLP downstream tasks such as classification, Named Entity Recognition and so on. |
|
|
|
|
|
## Training and evaluation data |
|
THe training corpus is developed using about 40000 resumes and 2000 jobs posted scrapped from different job portals. This is a preliminary dataset |
|
for the experimentation. THe corpus size is about 2.35 GB of textual data. Similary evaluation data contains few resumes and jobs making about 12 mb of textual data. |
|
|
|
## Training procedure |
|
For the pretraining of masked language model, Trainer API from Huggingface is used. The pretraining took about 6 hrs 40 mins. |
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 5e-05 |
|
- train_batch_size: 128 |
|
- eval_batch_size: 128 |
|
- seed: 42 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: linear |
|
- num_epochs: 5 |
|
|
|
### Training results |
|
|
|
| Training Loss | Epoch | Step | Validation Loss | |
|
|:-------------:|:-----:|:-----:|:---------------:| |
|
| 6.7862 | 0.11 | 500 | 6.9461 | |
|
| 5.9428 | 0.22 | 1000 | 6.4640 | |
|
| 5.5463 | 0.33 | 1500 | 6.2736 | |
|
| 5.1871 | 0.44 | 2000 | 5.8517 | |
|
| 4.896 | 0.55 | 2500 | 5.6070 | |
|
| 4.6557 | 0.66 | 3000 | 5.4669 | |
|
| 4.4832 | 0.77 | 3500 | 5.3318 | |
|
| 4.3368 | 0.88 | 4000 | 5.2414 | |
|
| 4.1887 | 0.99 | 4500 | 5.0666 | |
|
| 4.053 | 1.1 | 5000 | 4.9532 | |
|
| 3.9653 | 1.21 | 5500 | 4.8288 | |
|
| 3.8865 | 1.33 | 6000 | 4.6741 | |
|
| 3.8294 | 1.44 | 6500 | 4.7943 | |
|
| 3.7565 | 1.55 | 7000 | 4.7336 | |
|
| 3.673 | 1.66 | 7500 | 4.4760 | |
|
| 3.6447 | 1.77 | 8000 | 4.5856 | |
|
| 3.5808 | 1.88 | 8500 | 4.6133 | |
|
| 3.5329 | 1.99 | 9000 | 4.4766 | |
|
| 3.4916 | 2.1 | 9500 | 4.5085 | |
|
| 3.4392 | 2.21 | 10000 | 4.5306 | |
|
| 3.4333 | 2.32 | 10500 | 4.5433 | |
|
| 3.3905 | 2.43 | 11000 | 4.1829 | |
|
| 3.3701 | 2.54 | 11500 | 4.2976 | |
|
| 3.3345 | 2.65 | 12000 | 4.2817 | |
|
| 3.2815 | 2.76 | 12500 | 4.3146 | |
|
| 3.2689 | 2.87 | 13000 | 4.2634 | |
|
| 3.2401 | 2.98 | 13500 | 4.0907 | |
|
| 3.2068 | 3.09 | 14000 | 4.1130 | |
|
| 3.2097 | 3.2 | 14500 | 4.2001 | |
|
| 3.1627 | 3.31 | 15000 | 4.0852 | |
|
| 3.1647 | 3.42 | 15500 | 4.0383 | |
|
| 3.1294 | 3.53 | 16000 | 3.9377 | |
|
| 3.1166 | 3.64 | 16500 | 4.0733 | |
|
| 3.1028 | 3.75 | 17000 | 3.8429 | |
|
| 3.0903 | 3.86 | 17500 | 4.1127 | |
|
| 3.0877 | 3.98 | 18000 | 3.8605 | |
|
| 3.0407 | 4.09 | 18500 | 3.8482 | |
|
| 3.0452 | 4.2 | 19000 | 4.0345 | |
|
| 3.0496 | 4.31 | 19500 | 3.8602 | |
|
| 3.0229 | 4.42 | 20000 | 4.2268 | |
|
| 3.0157 | 4.53 | 20500 | 3.8028 | |
|
| 3.0037 | 4.64 | 21000 | 3.8668 | |
|
| 2.9992 | 4.75 | 21500 | 3.9542 | |
|
| 3.016 | 4.86 | 22000 | 3.9090 | |
|
| 2.9804 | 4.97 | 22500 | 4.0495 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.25.1 |
|
- Pytorch 1.8.0+cu111 |
|
- Datasets 2.7.1 |
|
- Tokenizers 0.13.2 |