Menu
"深度阅读和分析 PDF 论文。支持 PDF 转 Markdown、智能摘要、关键信息提取、问答式学习、笔记生成。"
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
"自动化论文检索、翻译、导入和学习闭环系统。每日从 arXiv 检索论文、自动翻译标题和摘要、导入 Zotero、生成 Obsidian 笔记和学习报告。"
"生成各种类型的图表和流程图：流程图、时序图、类图、甘特图、思维导图等。支持 Mermaid、PlantUML、Graphviz 等格式。"
"GitHub CLI (gh) 自动化操作。管理仓库、Issues、Pull Requests、Workflows、Releases 等。支持所有 gh 命令。"
"深度反思学习过程，巩固知识，发现盲点，优化学习策略。基于笔记内容生成反思报告、学习建议、知识图谱。"
"从 arXiv 自动检索论文、翻译标题摘要、导入 Zotero、生成 Obsidian 笔记。支持关键词检索、自动去重、中文翻译。"
"生成论文学习报告：每日总结、每周总结、月度报告、季度报告、年度总结。统计学习进度、分析研究方向、追踪知识积累。"
Based on SOC occupation classification
| name | PDF Reader |
| description | "深度阅读和分析 PDF 论文。支持 PDF 转 Markdown、智能摘要、关键信息提取、问答式学习、笔记生成。" |
Use this skill when you want to:
Triggers:
阅读 PDF: D:\Papers\attention-is-all-you-need.pdf
阅读 arXiv 论文: 2303.12345
这篇论文的主要贡献是什么？
python {workspace}/PaperVault/scripts/pdf-reader.py --pdf path/to/paper.pdf
Converts PDF to structured Markdown:
Input: PDF file
Output: {paper-name}.md
Conversion includes:
Example Output:
---
title: "Attention Is All You Need"
arxiv_id: "1706.03762"
date_read: 2024-03-25
reading_time: 45min
difficulty: ⭐⭐⭐☆☆
---
# Attention Is All You Need
> [!abstract] 摘要
> The dominant sequence transduction models are based on complex recurrent or convolutional neural networks...
## 1. Introduction
Recurrent neural networks, long short-term memory and gated recurrent neural networks...
### 1.1 Background
The Transformer uses multi-headed self-attention...
## 2. Model Architecture
### 2.1 Encoder and Decoder Stacks
**Encoder**: The encoder is composed of a stack of N = 6 identical layers...
**Decoder**: The decoder is also composed of a stack of N = 6 identical layers...
## 3. Attention
$$
\text{Attention}(Q, K, V) = \text{softmax}\left(\frac{QK^T}{\sqrt{d_k}}\right)V
$$
## 4. Experiments
| Model | BLEU | Training Time |
|-------|------|---------------|
| Transformer (big) | 28.4 | 3.5 days |
| Transformer (base) | 27.3 | 12 hours |
## Key Insights
1. Self-attention allows modeling of dependencies regardless of distance
2. Multi-head attention enables attending to information from different positions
3. Positional encoding is necessary since the model contains no recurrence
## Questions
- [ ] Why divide by sqrt(d_k)?
- [ ] How does multi-head attention work in detail?
- [ ] What are the computational complexity trade-offs?
## References
1. Vaswani et al. (2017) - This paper
2. Bahdanau et al. (2015) - Attention mechanism
3. Gehring et al. (2017) - Convolutional sequence models
Generates multiple levels of summaries:
1-sentence summary: "Proposes Transformer, a purely attention-based architecture for sequence transduction."
1-paragraph summary: "The paper introduces the Transformer model, which relies entirely on self-attention mechanisms without recurrence or convolution. It achieves state-of-the-art performance on machine translation tasks with significantly reduced training time."
Key points summary:
Extracts structured information:
Contributions:
{
"contributions": [
"Proposed Transformer architecture",
"Introduced multi-head self-attention",
"Achieved SOTA on WMT 2014",
"Reduced training time significantly"
]
}
Methods:
{
"methods": [
{
"name": "Scaled Dot-Product Attention",
"equation": "Attention(Q,K,V) = softmax(QK^T / sqrt(d_k))V",
"purpose": "Compute attention weights efficiently"
},
{
"name": "Multi-Head Attention",
"description": "Run multiple attention operations in parallel",
"benefit": "Attend to information from different positions"
}
]
}
Experiments:
{
"experiments": [
{
"task": "Machine Translation",
"dataset": "WMT 2014 English-German",
"metric": "BLEU",
"result": 28.4,
"baseline": 26.4
}
]
}
Interactive question answering:
User: "What is the main innovation?" Assistant: "The main innovation is replacing recurrent layers with self-attention mechanisms, allowing for more parallelization and better modeling of long-range dependencies."
User: "How does self-attention work?" Assistant: "Self-attention computes relationships between all positions in a sequence simultaneously. For each position, it computes attention weights with respect to all other positions..."
User: "What are the limitations?" Assistant: "Based on the paper and subsequent research: 1) Quadratic memory complexity for long sequences, 2) Position encoding may not generalize well, 3) Limited to fixed-length contexts..."
Generates comprehensive notes:
Concept Notes:
# Self-Attention
## Definition
Self-attention is a mechanism that relates different positions of a single sequence...
## How It Works
1. Compute Query, Key, Value matrices
2. Calculate attention weights
3. Apply weights to values
4. Output weighted sum
## Why It Matters
- Captures long-range dependencies
- Highly parallelizable
- Interpretable attention patterns
## Applications
- Machine translation
- Text summarization
- Image generation
## Related Concepts
- [[Attention Mechanism]]
- [[Transformer]]
- [[Multi-Head Attention]]
Method Notes:
# Multi-Head Attention
## Overview
Runs multiple self-attention operations in parallel...
## Implementation
```python
class MultiHeadAttention(nn.Module):
def __init__(self, d_model, num_heads):
super().__init__()
self.num_heads = num_heads
self.d_k = d_model // num_heads
# ...
## Reading Workflow
### Level 1: Quick Scan (5-10 min)
- Title and abstract
- Introduction
- Conclusion
- Key figures
**Output**: High-level understanding
### Level 2: Standard Read (30-60 min)
- All sections
- Important equations
- Key experiments
- Method details
**Output**: Detailed notes + questions
### Level 3: Deep Dive (2-4 hours)
- Every section in detail
- Derive equations
- Reproduce experiments
- Related work
**Output**: Comprehensive understanding + implementation
## Advanced Features
### Batch Processing
Process multiple PDFs:
```bash
python pdf-reader.py --dir D:\Papers\ --batch
Compare multiple papers:
对比分析这两篇论文的异同
python pdf-reader.py --zotero --query "attention mechanism"
python pdf-reader.py --pdf paper.pdf --save-to-zotero
Create prompts/custom-summary.txt:
请从以下角度总结这篇论文：
1. 研究动机
2. 核心创新
3. 技术方案
4. 实验验证
5. 局限性
Create templates/custom-template.md:
# {title}
## 一句话总结
{one_sentence_summary}
## 核心观点
{key_points}
## 技术细节
{technical_details}
## 我的思考
<!-- Your reflections -->
Error: Cannot extract text from PDF
Solutions:
--ocr flagWarning: Many formatting errors
Solutions:
--enhance flagError: API rate limit exceeded
Solution: Wait or use different API key
| Key | Action |
|---|---|
n | Next section |
p | Previous section |
q | Ask question |
s | Save note |
h | Highlight text |
f | Find in paper |
Esc | Exit |
paper-fetcher - Get papers to readpaper-summarizer - Generate summarieslearning-reflector - Reflect on readingzotero-local - Manage PDF library