mohar07/qwen3-0.6b-kg-triplets

TEXT GENERATIONConcurrent Unit Cost:1Model Size:0.8BQuant:BF16Context Size:32kTool Calling:SupportedPublished:May 28, 2026License:mitArchitecture:Transformer Open Weights Featherless Exclusive Cold

mohar07/qwen3-0.6b-kg-triplets is a LoRA finetuned Qwen3-0.6B model, developed by Mohar Das, specifically designed for ontology-constrained knowledge graph extraction. This 0.6 billion parameter model, with a 2048-token context length, generates structured triplets (source -> relation -> target) with normalized entities and calibrated confidence weights. It excels at producing graph-ingestable JSON output for direct use in graph databases and GraphRAG pipelines, addressing common inconsistencies found in general LLM extractions.

Loading preview...

Qwen3-0.6B-KG-Triplets: Ontology-Constrained Knowledge Graph Extraction

This model is a LoRA finetuned version of the Qwen3-0.6B base model, developed by Mohar Das, specialized for highly structured knowledge graph extraction. Unlike general instruction-tuned LLMs, it focuses on generating triplets (source -> relation -> target) that are directly ingestible into graph databases and GraphRAG pipelines.

Key Capabilities & Differentiators

  • Ontology-Constrained Output: Strictly adheres to a predefined set of 20 relation types, preventing out-of-schema relations.
  • Normalized Entities: Produces consistent and reusable entity names, avoiding fragmented or inconsistent extractions.
  • Calibrated Confidence Weights: Assigns meaningful confidence scores to relations, improving the utility of extracted data.
  • Graph-Ingestable JSON: Outputs structured JSON designed for minimal post-processing before database ingestion.
  • Targeted Training: Finetuned on a custom dataset derived from Wikipedia and arXiv, featuring hard negatives and curriculum ordering to enhance precision and schema adherence.
  • Small Footprint: Achieves stable, specialized extraction behavior with only 0.6 billion parameters and a 2048-token context length.

Intended Use Cases

This model is ideal for applications requiring precise and structured information extraction for:

  • Knowledge Graph Construction
  • GraphRAG pipelines
  • Automated Knowledge Graph Population
  • Document-to-Graph Conversion

While prioritizing schema correctness and grounding, the model intentionally trades some recall for high precision and adherence to the defined ontology. It is primarily trained on English content.

Finetunes
1 models