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

Summary

Training and aligning large language models and small language models requires datasets tailored for complex reasoning, preference tuning, multi-turn questioning, and format conversion. To eliminate the need for bespoke data preparation scripts, SyGra provides a low-code and no-code Python framework for automated dataset creation, transformation, and alignment. The library integrates with diverse inference backends, including vLLM, Hugging Face TGI, Triton, and Ollama, allowing teams to focus on prompt engineering rather than pipeline infrastructure. It supports data workflows ranging from converting PDFs and knowledge bases into question-answering pairs to filtering low-quality samples and generating Direct Preference Optimization datasets. These plug-and-play workflows reduce manual curation effort while accelerating model fine-tuning and retrieval-augmented generation pipelines.

Context

Training and aligning LLMs and SLMs requires structured, high-quality, and task-specific datasets, such as reasoning chains, Direct Preference Optimization pairs, and multi-turn questions. Raw or open data rarely matches required formats, forcing practitioners to manually filter samples, expand contexts for RAG, translate across languages, or extract structured Q&A data from PDFs and knowledge bases.

Approach / What changed

SyGra is a low-code/no-code Python library and framework that automates dataset creation, transformation, quality filtering, and alignment. It delegates heavy pipeline engineering so users can focus on prompt engineering, offering native support for inference backends like vLLM, Hugging Face TGI, Triton, and Ollama.

Takeaways

• SyGra provides a low-code and no-code Python framework that lets developers build complex datasets by focusing on prompt engineering rather than custom pipeline scripts.
• The framework natively integrates with multiple inference backends, including vLLM, Hugging Face TGI, Triton, and Ollama.
• Supported dataset tasks include turning knowledge bases or PDFs into Q&A formats, generating DPO preference pairs from SFT data, quality filtering, and expanding contexts for RAG pipelines.

Related reading

• Hugging Face · Dec 4, 2025

  We Got Claude to Fine-Tune an Open Source LLM

  Hugging Face Skills equips coding agents like Claude Code, OpenAI Codex, and Google's Gemini CLI to manage end-to-end language model fine-tuning. Using the hf-llm-trainer skill, an agent validates dataset formats, selects cloud hardware, configures authentication, and submits jobs to Hugging Face Jobs. Supported techniques include Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Group Relative Policy Optimization (GRPO) for models ranging across various parameter sizes. The integration incorporates Trackio for real-time monitoring and automates LoRA configuration for larger models. Once training completes, the agent pushes artifacts to the Hugging Face Hub and can convert models to GGUF format for local execution.

  ben burtenshaw, shaun smith

• Hugging Face · Nov 19, 2025

  Apriel-H1: The Surprising Key to Distilling Efficient Reasoning Models