Sunday, May 07 2023
Created:
LLM
Read karpathy/nanoGPT: The simplest, fastest repository for training/finetuning medium-sized GPTs. to understand how transformers work
“Chat.lmsys.org” - Chat with Open Large Language Models Has multiple Open Large Language Models that you can chat with. Some of them listed are - Vicuna, LLaMa, Koala, Alpaca, Dolly, ChatGLM and StableLM.
bigcode/starcoder · Hugging Face StarCoder is the first large model (15B) which is both high performance (beating the like of PaLM, LLaMa, CodeGen or OpenAI code-crushman-001 on code generation) and also trained only on carefully vetted data (only permissive code licences, comprehensive PII removing) If you need a clean and high performance code generation model, check it out Even though it can act as a chat model with a dialog prompt, it’s not a real chat model though as we haven’t done dialog finetuning yet so don’t use it like ChatGPT. You should rather use it as the basis for internal finetuning or for code completion e.g. with the vscode extension - HF Code autocomplete
Thread on the design of texttodesign.ai using LangChainAI and AWSAmplify; Ollie The Dev | Building Text To Design: A Full Stack Journey with AWS Amplify, React, and Figma API. See the video on texttodesign.ai’s home page for the demo of how the chat is used to generate the design (web design from what I can tell). This appears to be a good example of how to put together
- LangChainAI - the AI component
- AWSAmplify - the infrastructure
- Figma - existing platform to extend
- TextToDesign - the integrator to extend the plaform’s capability