<aside>
ℹ️ מידע על העמוד
- מקורות מידע - Based on recommendations from Semantic Scholar folks
- עדכניות - קיץ 2023
</aside>
How to write good prompts
https://github.com/dair-ai/Prompt-Engineering-Guide
Prompt Engineering Course by FastAI
Prompting methods
Automatic prompt design
GPT-specific prompting advice
Evaluating prompts
How to plan prompting-based experiments for baselines
How to pick few-shots examples
How to pick few-shots order of examples
Instruction-tuning vs few-shots
Useful packages / platforms
- DSPy - a framework designed to help users create, manage, and optimize prompts for LLMs in a structured, declarative manner. It allows users to define prompts, conditions, and transformations declaratively, making it easier to experiment with and refine prompts for different tasks. DSPy helps streamline the process of prompt engineering, enabling more effective and efficient interactions with LLMs. It is very powerful and commonly used, though it is known to have a difficult learning curve and “making users feel stupid”..
- Guidance - a tool from Microsoft seems like the most sophisticated approach for writing and maintaining complex prompts that only need to interact with a language model. Highly recommended by S2 folks. one of them said - “Guidance has a slightly steep learning curve; but just after 1 hour reading this it will transform the way for you writing any prompts in the future.”
- MiniChain - A tiny library for coding with large language models. Another recommendation from the S2 folks, described as “in my opinion, the most elegant and well designed code out there” .
- PyLLMs - a minimal Python library to connect to LLMs (OpenAI, Anthropic, Google, AI21, Cohere, Aleph Alpha, HuggingfaceHub) with a built-in model performance benchmark. Built to allow you to fast prototyping and evaluating different models. Another recommendation from the S2 folks, described as “nice if you don't need much prompting guidance, but wanna be able to switch between providers”. Another comparable option is EasyLLM that was created by a HuggingFace employee, so it has nice support for HuggingFace models. Another comparable option is **lightLLM** by BerriAI.
- **promptlayer -** a platform for GPT prompt engineering. allows you to log requests to the OpenAI API (and than see on their web GUI each prompt and GPT’s response in various formats). Allows you to create templates for prompts.
- **W&B prompts** - A platform that allows intuitive visualizations that allow you to interrogate every step of a program that interacts with LLMs. Also connected to OpenAI API and allows you to analyze API calls with dedicated dashboards.