datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
storyweaver-writing-zh
StoryWeaver 中文写作质量评测集
12 道按写作失效模式反推设计的中文创作题、4 个参赛者写出的 48 篇章节、432 条逐维度两两判决（含裁判完整推理原文）。
来自 StoryWeaver 的写作质量评测轨道。榜单：https://storyweaver.cn/benchmark-writing.html
核心结论
接系统比换一代底模更管用。同一底模接上多 Agent 系统后的胜率：k2.5 **75.1%**、k2.6 **60.2%**；而 k2.5(系统) 对 k2.6(裸) 是 70.3%，反过来只有 37.2%——系统加持能把旧一代底模抬过裸的新一代底模。系统档拿下 22 个维度里的 20 个榜首，包括全部 9 个负向维度。
k2.5 与 k2.6 之间 54.7%，落在噪音带内，不构成结论。
题目怎么设计的
每道题咬住 rubric 里的一个维度或负向维度，用硬约束逼出功力：… See the full description on the dataset page: https://huggingface.co/datasets/godwei123/storyweaver-writing-zh.Python_GOD_Coder_Omniforge_AI_12k
Python GOD Coder Omniforge AI 12k
Creator: Within Us AI
A 12,000-row mixed-format Python coding dataset designed as a sharpening corpus for building a small but dangerous Python specialist.
This dataset is intentionally focused on the practical behaviors that matter for a modern Python coding model:
implementation with tests
strict code-only instruction following
debugging and repair
refactoring for readability and production readiness
next-token code completion… See the full description on the dataset page: https://huggingface.co/datasets/WithinUsAI/Python_GOD_Coder_Omniforge_AI_12k.godot-lora-dataset
Godot LORA Dataset
GDScript training dataset for fine-tuning code models on Godot engine development.
Dataset Info
Total samples: 476
Train split: 428
Validation split: 48
Format: JSONL (instruction, input, output)
Language: GDScript (Godot 4.x)
Languages: German instructions, GDScript code
Sources
godotengine/godot-demo-projects
GDQuest/godot-open-rpg
GDQuest/godot-3d-dodge-the-creeps
bitbrain/beehave (behavior trees)
limboai/limboai (AI for Godot)… See the full description on the dataset page: https://huggingface.co/datasets/matzejo/godot-lora-dataset.storyweaver-chunking-zh
StoryWeaver 中文叙事切分语料
12 篇中文叙事短文，按切分算法的失效模式反推设计，用于比较 chunking 策略在长篇小说创作场景下的表现。
来自 StoryWeaver 的切分质量评测轨道。榜单：https://storyweaver.cn/benchmark-chunking.html
这批语料的特别之处
它不是真实小说的随机采样，而是每篇专门写来触发某一类切分失效：
对话密集、闪回嵌套、隐晦转场、双线交替、同场景多视角、同场景话题漂移……
这么设计是为了让不同算法拉开差距——真实文本里这些情况稀疏出现，抽样评测容易得到"各方法差不多"的钝化结论。代价是分布有偏，解读结果时必须计入这一点。
文件
corpus.jsonl（12 行）
字段
说明
doc_id
篇名（中文，如 夜班、面馆）
text
全文
chars
字数
leaderboard.json
首轮评测的汇总结果：6… See the full description on the dataset page: https://huggingface.co/datasets/godwei123/storyweaver-chunking-zh.Kishor_V2_53K_LLM_Prompt-Response_Pairs
Kishor V2: 53K Prompt-Response Dataset
Kishor V2 is a diverse and compact dataset curated for training small to medium-sized language models. It includes 53,000 structured prompt-response pairs across multiple domains to simulate human-like dialogue, reasoning, and general intelligence.
📦 File
KishorV2_dataset.jsonl: Main dataset in JSON Lines format.
📂 Format
Each line is a JSON object with:
{
"type": "qa" | "dialogue" | "quote" | "fact" | "reasoning" |… See the full description on the dataset page: https://huggingface.co/datasets/GODELEV/Kishor_V2_53K_LLM_Prompt-Response_Pairs.chat_his