SyGra: The One-Stop Framework for Building Data for LLMs and SLMs

When we think about building a model - be it a Large Language Model (LLM) or a Small Language Model (SLM) - the first thing we need is data. While a vast amount of open data is available, it rarely comes in the exact format required to train or align models. In practice, we often face scenarios where the raw data isn't enough. We need data that is more structured, domain-specific, complex, or aligned with the task at hand. Let's look at some common situations:

Complex Scenarios Missing

 You start with a simple dataset, but the model fails on advanced reasoning tasks. How do you generate more complex datasets to strengthen performance?

Knowledge Base to Q&A

 You already have a knowledge base, but it's not in Q&A format. How can you transform it into a usable question-answering dataset?

From SFT to DPO

 You've prepared a supervised fine-tuning (SFT) dataset. But now you want to align your model using Direct Preference Optimization (DPO). How can you generate preference pairs?

Depth of Questions

 You have a Q&A dataset, but the questions are shallow. How can you create in-depth, multi-turn, or reasoning-heavy questions?

Domain-Specific Mid-Training

 You possess a massive corpus but need to filter and curate data for mid-training on a specific domain.

PDFs and Images to Documents

 Your data lives in PDFs or images, and you need to convert them into structured documents for building a Q&A system.

Boosting Reasoning Ability

 You already have reasoning datasets, but want to push models toward better "thinking tokens" for step-by-step problem-solving.

Quality Filtering

 Not all data is good data. How do you automatically filter out poor-quality samples and keep only the high-value ones?

Small to Large Contexts

 Your dataset has small chunks of context, but you want to build larger-context datasets optimized for RAG (Retrieval-Augmented Generation) pipelines.