flammenai/flame-kindling-v1

flame-kindling-v1是一个小型、专业化的监督微调（SFT）数据集，旨在将一个3B级别的指令模型微调成一个能够从自由文本种子生成严格JSON格式输出的角色设计器。数据集包含400个种子到设计角色的配对，这些配对经过Claude Sonnet 4.5的蒸馏、严格验证、去重处理，并由Qwen3.5-27B进行一致性评估。数据集支持多种语言，主要用于flammen.ai的Create-a-Flame流程，也可用于其他需要生成JSON格式角色的项目。

flame-kindling-v1 is a small, opinionated SFT dataset designed to finetune a 3B-class instruct model into a character designer that emits a strict JSON schema from a free-text seed. It contains 400 (seed → DesignedFlame) pairs distilled from Claude Sonnet 4.5, validated against a strict pydantic schema, deduplicated, and judged by Qwen3.5-27B for coherence. The dataset supports multiple languages and is primarily intended for flammen.ais Create-a-Flame pipeline, but can also serve as a starter for any project needing a small JSON-emitting character generator.