add guidance
This commit is contained in:
parent
7db8c049e8
commit
ebb5167683
|
|
@ -1,6 +1,6 @@
|
||||||
---
|
---
|
||||||
layout: default
|
layout: default
|
||||||
title: "Example"
|
title: "Apps"
|
||||||
nav_order: 5
|
nav_order: 5
|
||||||
has_children: true
|
has_children: true
|
||||||
---
|
---
|
||||||
|
|
@ -1,8 +1,8 @@
|
||||||
---
|
---
|
||||||
layout: default
|
layout: default
|
||||||
title: "Essay"
|
title: "Essay"
|
||||||
parent: "Example"
|
parent: "Apps"
|
||||||
nav_order: 1
|
nav_order: 2
|
||||||
---
|
---
|
||||||
|
|
||||||
# Summarization + QA agent for Paul Graham Essay
|
# Summarization + QA agent for Paul Graham Essay
|
||||||
|
|
|
||||||
|
|
@ -0,0 +1,37 @@
|
||||||
|
---
|
||||||
|
layout: default
|
||||||
|
title: "Essay"
|
||||||
|
parent: "Apps"
|
||||||
|
nav_order: 1
|
||||||
|
---
|
||||||
|
|
||||||
|
# LLM System Design Guidance
|
||||||
|
|
||||||
|
**Use LLMs to help with system design and implementation wherever possible.**
|
||||||
|
|
||||||
|
1. **Understand Requirements**
|
||||||
|
- Clarify the app’s needs and workflow.
|
||||||
|
- Determine data access (e.g., from files or databases).
|
||||||
|
|
||||||
|
2. **High-Level Flow Design**
|
||||||
|
- Represent the process as a *Nested Directed Graph*.
|
||||||
|
- Identify possible branching for *Node Action*.
|
||||||
|
- Identify data-heavy steps for *Batch*.
|
||||||
|
|
||||||
|
3. **Shared Memory Structure**
|
||||||
|
- For small apps, in-memory data is sufficient.
|
||||||
|
- For larger or persistent needs, use a database.
|
||||||
|
- Define schemas or data structures and plan how states will be stored and updated.
|
||||||
|
|
||||||
|
4. **Implementation**
|
||||||
|
- Rely on LLMs for coding tasks.
|
||||||
|
- Start with minimal, straightforward code (e.g., avoid heavy type checking initially).
|
||||||
|
|
||||||
|
5. **Optimization**
|
||||||
|
- *Prompt Engineering:* Provide clear instructions and examples to reduce ambiguity.
|
||||||
|
- *Task Decomposition:* Break complex tasks into manageable steps.
|
||||||
|
|
||||||
|
6. **Reliability**
|
||||||
|
- *Structured Output:* Verify outputs match the desired format, and increase `max_retries`.
|
||||||
|
- *Test Cases:* Create tests for parts of the flow with clear inputs/outputs.
|
||||||
|
- *Self-Evaluation:* For unclear areas, add another Node for LLMs to review the results.
|
||||||
|
|
@ -60,7 +60,9 @@ We model the LLM workflow as a **Nested Directed Graph**:
|
||||||
- [(Advanced) Multi-Agents](./multi_agent.md)
|
- [(Advanced) Multi-Agents](./multi_agent.md)
|
||||||
- Evaluation
|
- Evaluation
|
||||||
|
|
||||||
## Example Projects
|
## Example LLM Apps
|
||||||
|
|
||||||
|
[LLM System Design Guidance](./guide.md)
|
||||||
|
|
||||||
- [Summarization + QA agent for Paul Graham Essay](./essay.md)
|
- [Summarization + QA agent for Paul Graham Essay](./essay.md)
|
||||||
- More coming soon...
|
- More coming soon...
|
||||||
|
|
|
||||||
Loading…
Reference in New Issue