StructuraAI / README.md
talha2134's picture
Upload folder using huggingface_hub
1ac4819 verified
|
Raw History Blame Contribute Delete
1.96 kB
---
base_model: Qwen/Qwen3-0.6B
library_name: transformers
model_name: StructuraAI
tags:
- generated_from_trainer
- sft
- trl
- json
- information-extraction
- text-generation
---
# StructuraAI
StructuraAI is a fine-tuned language model based on
Qwen3-0.6B. It is designed to automatically extract
structured information from unstructured and messy text
and return the extracted information as JSON.
Unlike conventional extraction systems that require
explicit instructions for every request, StructuraAI
aims to identify available information automatically.
## Features
- Automatic information extraction.
- Converts unstructured text into JSON.
- Extracts only information explicitly mentioned.
- Omits attributes that are not present in the input.
- Avoids intentionally inferring missing information.
## Example
### Input
Ahmed is 22 years old and works at Google as an AI
Engineer for 3 years.
### Output
```json
{
"name": "Ahmed",
"age": 22,
"company": "Google",
"designation": "AI Engineer",
"experience": 3
}
```
## Base Model
Qwen/Qwen3-0.6B
## Training Procedure
This model was fine-tuned using Supervised Fine-Tuning
(SFT) with the TRL library.
The training dataset consists of synthetically generated
examples designed for information extraction from
unstructured text.
## Intended Use
StructuraAI is intended for:
- Information extraction.
- Converting messy text into structured JSON.
- Preparing unstructured data for downstream applications.
- Prototyping automated data-processing workflows.
## Limitations
- The model may occasionally generate invalid JSON.
- It may confuse attributes with similar meanings.
- Extraction accuracy depends on the wording and quality
of the input.
- Outputs should be validated before production use.
## Frameworks
- Transformers
- TRL
- PyTorch
- Hugging Face Datasets
## Acknowledgements
Built upon Qwen3-0.6B and trained using the Hugging Face
TRL ecosystem.