sadnblueish/apache-hadoop-mddocs-instruct
收藏Hugging Face2026-03-25 更新2026-03-29 收录
下载链接:
https://hf-mirror.com/datasets/sadnblueish/apache-hadoop-mddocs-instruct
下载链接
链接失效反馈官方服务:
资源简介:
---
dataset_info:
features:
- name: instruction
dtype: string
- name: response
dtype: string
- name: hadoop_version
dtype: string
- name: source
dtype: string
- name: section_title
dtype: string
- name: system
dtype: string
splits:
- name: train
num_bytes: 14904765
num_examples: 12438
- name: test
num_bytes: 784902
num_examples: 655
download_size: 4883687
dataset_size: 15689667
configs:
- config_name: default
data_files:
- split: train
path: hadoop_3.4.3/train-*
- split: test
path: hadoop_3.4.3/test-*
license: apache-2.0
task_categories:
- question-answering
- feature-extraction
- text-generation
language:
- en
pretty_name: Apache Hadoop v3.4.3
size_categories:
- 10K<n<100K
---
# Dataset Card for sadnblueish/apache-hadoop-mddocs-instruct
Domain Knowledge Synthetic Dataset of Apache Hadoop v3.4.3.
## Dataset Details
### Dataset Description
AI Cognitive SFT dataset for domain knowledge of Apache Hadoop version 3.4.3.
Ollama hosted Deepseek-Coder-16B:Q4 was used to augment the dataset via multi-step Markdown ingestion pipeline.
A LoRA adapter of Qwen2.5-Coder-7B was Fine Tuned with this dataset:
| | Avg Train Loss | Final Train Loss | Eval Loss | Train Runtime (s) | flash_attn |
| ------| -------------- | ---------------- | --------- | ----------------- | -------------|
| bf16 | 0.5232 | 0.2942 | 0.7667 | 2886 | flash-attn-3 |
### Dataset Sources
- **Repository:** https://github.com/apache/hadoop/tree/branch-3.4.3
数据集信息:
特征字段:
- 字段名:instruction,数据类型:字符串
- 字段名:response,数据类型:字符串
- 字段名:hadoop_version,数据类型:字符串
- 字段名:source,数据类型:字符串
- 字段名:section_title,数据类型:字符串
- 字段名:system,数据类型:字符串
数据集划分:
- 划分名称:train,字节数:14904765,样本数:12438
- 划分名称:test,字节数:784902,样本数:655
下载大小:4883687,总数据集大小:15689667
配置项:
- 配置名称:default,数据文件:
- 训练集划分:hadoop_3.4.3/train-*
- 测试集划分:hadoop_3.4.3/test-*
许可证:apache-2.0
任务类别:
- 问答任务
- 特征提取任务
- 文本生成任务
语言:英语
数据集展示名:Apache Hadoop v3.4.3
样本量区间:10000 < 样本数 < 100000
# 数据集卡片:sadnblueish/apache-hadoop-mddocs-instruct
Apache Hadoop 3.4.3 版本领域知识合成数据集。
## 数据集详情
### 数据集描述
面向Apache Hadoop 3.4.3版本领域知识的AI认知监督微调(Supervised Fine-Tuning, SFT)数据集。本数据集通过多步Markdown文档摄取流水线,借助Ollama部署的Deepseek-Coder-16B:Q4模型完成数据增强。
本数据集用于微调Qwen2.5-Coder-7B的低秩适配器(Low-Rank Adaptation, LoRA),微调结果如下:
| | 平均训练损失 | 最终训练损失 | 评估损失 | 训练运行时长(秒) | Flash Attention (flash_attn) |
| ------| -------------- | ---------------- | --------- | ----------------- | -------------|
| bf16 | 0.5232 | 0.2942 | 0.7667 | 2886 | flash-attn-3 |
### 数据集来源
- **代码仓库**:https://github.com/apache/hadoop/tree/branch-3.4.3
提供机构:
sadnblueish



