Fabrix-AI-Inc/Triton-VX-Qwen3.5-2B-DPO
Fabrix-AI-Inc/Triton-VX-Qwen3.5-2B-DPO is a 2.3 billion parameter Qwen3.5-based language model, specifically domain-adapted and DPO-aligned for Fabrix.ai's knowledge retrieval and documentation citation. It excels at generating precise, single-card citations for Fabrix.ai's internal knowledge base, distinguishing it from general-purpose LLMs. This model is optimized for use cases requiring accurate and concise internal documentation referencing.
Loading preview...
Triton-VX-Qwen3.5-2B-DPO: Specialized Citation Model
This model, developed by Fabrix-AI-Inc, is a 2.3 billion parameter language model based on the Qwen3.5 architecture. It is uniquely domain-adapted and DPO-aligned to serve as an assistant for Fabrix.ai's internal knowledge retrieval, documentation citation, and custom dashboard generation.
Key Capabilities & Alignment:
- Precise Citation Generation: Specifically trained to output repo-relative knowledge base paths, prioritizing single-card citations (e.g.,
kb/cards/*.md). - Two-Stage Alignment: Utilizes Supervised Fine-Tuning (SFT) on Fabrix technical documentation and Direct Preference Optimization (DPO) to eliminate multi-line citations and enforce conciseness.
- High Preference Accuracy: Achieves 100% DPO Preference Accuracy and 90.9% exact match on preferred citations on the Fabrix citation benchmark.
Ideal Use Cases:
- Internal Documentation Retrieval: Generating accurate and concise citations for Fabrix.ai's knowledge base.
- Developer Assistance: Aiding developers in quickly finding relevant internal documentation for widgets, dashboards, and pipelines.
- Automated Citation: Automating the process of linking user queries to specific internal documentation cards.