---
language:
- ru
- en
license: apache-2.0
base_model: igorktech/rugpt3-joker-150k
tags:
- not-for-all-audiences
- art
- humour
- jokes
- generated_from_trainer
model-index:
- name: zeio/wit
  results: []
datasets:
- zeio/baneks
metrics:
- loss
widget:
- text: 'Купил мужик шляпу'
  example_title: hat
- text: 'Пришла бабка к врачу'
  example_title: doctor
- text: 'Нашел мужик подкову'
  example_title: horseshoe
---

<p align="center">
    <img src="https://i.ibb.co/zP7j7ng/wit-logo.png"/>
</p>

# wit

This model is a fine-tuned version of [igorktech/rugpt3-joker-150k][base] on the [baneks][dataset] dataset for 1 epoch. It achieved `2.0391` loss during training.
Model evaluation has not been performed.

## Model description

The model is a fine-tuned variant of the [igorktech/rugpt3-joker-150k][base] architecture with causal language modeling head.

## Intended uses & limitations

The model should be used for studying abilities of natural language models to generate jokes.

## Training and evaluation data

The model is trained on a list of anecdotes pulled from a few vk communities (see [baneks][dataset] dataset for more details).

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 0.0005
- train_batch_size: 8
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 8
- total_train_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 1000
- num_epochs: 10

### Training results

| Train Loss | Epoch |
|:----------:|:-----:|
| 2.0391     | 10    |

### Framework versions

- Transformers 4.34.0
- Pytorch 2.1.0
- Datasets 2.12.0
- Tokenizers 0.14.1

[base]: https://huggingface.co/igorktech/rugpt3-joker-150k
[dataset]: https://huggingface.co/datasets/zeio/baneks