pokutuna/llm2025-main
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Feb 7, 2026License:apache-2.0Architecture:Transformer Open Weights Cold
pokutuna/llm2025-main is a 4 billion parameter instruction-tuned causal language model, fine-tuned from Qwen/Qwen3-4B-Instruct-2507. Developed by pokutuna, this model is specifically optimized for generating structured outputs such as JSON, YAML, XML, TOML, and CSV. It utilizes NEFTune noise regularization and a specialized training objective focusing on assistant output loss to enhance accuracy for structured data generation tasks.
Loading preview...
Overview
pokutuna/llm2025-main is a 4 billion parameter language model, fine-tuned from Qwen/Qwen3-4B-Instruct-2507, with a context length of 40960 tokens. Its primary objective is to excel at generating structured outputs, including JSON, YAML, XML, TOML, and CSV formats.
Key Capabilities
- Structured Output Generation: Specifically optimized for producing accurate and well-formed structured data.
- Targeted Training: Employs an assistant-only loss mechanism during training, focusing on the quality of the model's direct output for structured tasks, while masking intermediate Chain-of-Thought reasoning.
- Generalization: Incorporates NEFTune noise regularization to improve the model's ability to generalize across various structured output scenarios.
- Data Diversity: Trained on a curated set of datasets from the LLM2025 competition, including
u-10bei/structured_data_with_cot_datasetanddaichira/structured-3k-mix-sft, preprocessed for rule-based structuring.
Good For
- Data Extraction: Ideal for use cases requiring the conversion of unstructured text into structured formats.
- API Integration: Generating API requests or responses in JSON/YAML.
- Configuration Files: Creating or modifying configuration files in TOML or XML.
- Data Reporting: Producing CSV outputs from natural language instructions.