Tessar Largest

🚀 Tessar (Large-Sized Model)

Tessar is an advanced table reasoning model developed by SVECTOR. It builds upon groundbreaking research and expands the boundaries of neural table understanding, offering high - precision solutions for table - based question answering.

✨ Features

• Advanced neural SQL execution capabilities
• State - of - the - art performance on complex table question answering
• Robust handling of nuanced and multi - step queries
• Fine - tuned on the WikiTableQuestions dataset

📦 Installation

The README does not provide specific installation steps, so this section is skipped.

💻 Usage Examples

Basic Usage

from transformers import TessarTokenizer, BartForConditionalGeneration
import pandas as pd

# Load Tessar model and tokenizer
tokenizer = TessarTokenizer.from_pretrained("SVECTOR-CORPORATION/Tessar-largest")
model = BartForConditionalGeneration.from_pretrained("SVECTOR-CORPORATION/Tessar-largest")

# Prepare sample table data
data = {
    "year": [1896, 1900, 1904, 2004, 2008, 2012],
    "city": ["athens", "paris", "st. louis", "athens", "beijing", "london"]
}
table = pd.DataFrame.from_dict(data)

# Ask a specific query
query = "In which year did beijing host the Olympic Games?"
encoding = tokenizer(table=table, query=query, return_tensors="pt")

# Generate answer
outputs = model.generate(**encoding)

# Decode and print result
print(tokenizer.batch_decode(outputs, skip_special_tokens=True))
# Expected output: [' 2008.0']

Advanced Usage

There is no advanced usage code example in the original README, so this part is skipped.

📚 Documentation

Model Description

Tessar (Textual SQL Analysis and Reasoning) is a sophisticated neural model for table - based question answering. It implements an innovative neural SQL executor to interpret and reason over complex tabular data precisely.

The model uses the BART architecture, with a bidirectional encoder and an autoregressive decoder. This design enables Tessar to capture complex contextual relationships in tabular data and generate accurate, context - relevant answers.

Intended Uses

Tessar is powerful for solving complex table - based questions in various domains. Here are some example questions:

Evaluation

For comprehensive evaluation scripts and benchmarking, refer to the SVECTOR documentation and research repositories.

Performance Highlights

• Exceptional accuracy on complex table reasoning tasks
• Robust handling of multi - step and contextual queries
• State - of - the - art performance on the WikiTableQuestions dataset

Citation

If you use Tessar in your research the SVECTOR implementation:

@inproceedings{svector2025tessar,
    title={{Tessar}: Advanced Neural Table Reasoning},
    author={{SVECTOR Team}},
    year={2025},
    institution={SVECTOR Research}
}

Contact and Support

For further information, support, or collaboration opportunities, contact SVECTOR's research team at research@svector.co.in.

📄 License

This model is released under the MIT License. Review the licensing terms before use.