|
--- |
|
language: |
|
- en |
|
tags: |
|
- table-to-text |
|
- tabular |
|
datasets: |
|
- totto |
|
--- |
|
|
|
# BLOOM (0.56B) fine-tuned on ToTTo for Table-to-text |
|
|
|
This model is a fine-tuned version of [bigscience/bloom-560m](https://huggingface.co/bigscience/bloom-560m) on the **ToTTo** [dataset](https://huggingface.co/datasets/totto). |
|
|
|
|
|
## The model |
|
|
|
It is a 560M params version of [**BLOOM**](https://bigscience.huggingface.co/blog/bloom) |
|
|
|
## The dataset |
|
|
|
**ToTTo** is an open-domain English table-to-text dataset with over 120,000 training examples that proposes a controlled generation task: given a Wikipedia table and a set of highlighted table cells, produce a one-sentence description. |
|
|
|
During the dataset creation process, tables from English Wikipedia are matched with (noisy) descriptions. Each table cell mentioned in the description is highlighted and the descriptions are iteratively cleaned and corrected to faithfully reflect the content of the highlighted cells. |
|
|
|
|
|
### Evaluation results |
|
|
|
| Metric | Value | |
|
|:-------:|:-----:| |
|
| rouge1 | 0.56 | |
|
| rouge2 | 0.33 | |
|
| rougeL | 0.48 | |
|
| rougeLsum | 0.48 | |
|
|
|
|
|
## Usage |
|
|
|
```py |
|
from datasets import load_dataset |
|
from transformers import BloomTokenizerFast, BloomForCausalLM |
|
|
|
valid_dataset = load_dataset('totto', split='validation') |
|
|
|
from preprocess import preprocess # This file is included in the repo |
|
|
|
# Now we linearize the tables |
|
valid_dataset = valid_dataset.map(preprocess) |
|
|
|
model_ckpt = "mrm8488/bloom-560m-finetuned-totto-table-to-text" |
|
|
|
tokenizer = BloomTokenizerFast.from_pretrained(ckpt) |
|
model = BloomForCausalLM.from_pretrained(ckpt).to("cuda") |
|
|
|
|
|
def explain_hl_cells(text): |
|
inputs = tokenizer(text, return_tensors='pt') |
|
input_ids = inputs.input_ids.to("cuda") |
|
attention_mask = inputs.attention_mask.to("cuda") |
|
output = model.generate(input_ids, attention_mask=attention_mask, max_length=2048, eos_token_id=tokenizer.eos_token_id) # num_beams=3, temperature=1.9 |
|
|
|
return tokenizer.decode(output[0], skip_special_tokens=False) |
|
|
|
example = valid_dataset[1] |
|
|
|
print(explain_hl_cells(example['linearized_table']) |
|
``` |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.21.2 |
|
- Pytorch 1.12.1+cu113 |
|
- Datasets 2.4.0 |
|
- Tokenizers 0.12.1 |
