Here is an 80-line Python script of how to use the @GoogleDeepMind Gemini 2.0 Flash Live API for real-time conversations. 🗣️

1. Copy the code from the gist
2. "pip install pyaudio websockets"
2. "GEMINI_API_KEY={TOKEN} python live_test.py"
3. Talk to Gemini it is fun! Simple SimpleGeminiVoice Client gist.github.com/philschmid/cb8…

How can we teach LLMs to be factual, correct, and more reliable? 🤔
RAG is one approach to adding information to the prompt. But, always retrieving can lead to bad responses😔

Self-RAG proposes a new method to teach LLMs when to retrieve information and how to use it.🤯

🧶 🛠 𝗜𝗺𝗽𝗹𝗲𝗺𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻
1️⃣ Create a “critique” dataset with when to use retrieval and what is relevant using, e.g., GPT-4
2️⃣ Train critique model on the synthetic dataset
3️⃣ Use prompts, critique model and retriever to generate RAG dataset in an offline way

Amazon Bedrock Is Now Generally Available. Below is a Thread from my findings looking at the published sources and trying it out.

Let's start with a highlight. 🏞

🧶 Quotas for using different models, not adjustable at the moment.

Is Llama 2 special or just a better iteration of Llama 1? 🤔 Over the weekend, I had time to read the paper in which Meta released. 📖

Below are some of my findings, which you might have missed📝

🧵 1/6 🧠 A 34B version may come later after more testing
⚖️ The 7B model used a 285x token to parameter ratio, with loss still decreasing.
💰 Training the 7B would cost ~$1M in AWS compute (5$ per A100 on AWS on-demand)
🛫 Llama Chat was started before Llama 2 finished training

🧵2/6

OpenLLaMA 13B was released and competitive with its original counterpart from MetaAI. 🚀🎉 Two months ago, the OpenLM research initiative started to create a permissively licensed open-source reproduction of Meta AI’s LLaMA! 🛫

👉 huggingface.co/openlm-researc…
🧵 1/4 Last week the team released the 13B weights under Apache 2.0 with evaluations on the lm-evaluation-harness by EleutherAI🔓
OpenLLaMA matches @Meta LLaMA with an avg score of 0.57, making it a perfect replacement for all your commercial use cases🥊

huggingface.co/openlm-researc…
🧵 2/4

Finally had the time to read the "The False Promise of Imitating Proprietary LLMs.” paper in detail. 📚✨  Below are some of my key takeaways: 📝

🔍 Objective:
- The paper aimed to evaluate the effectiveness of models trained on GPT outputs.

🧵 1/4 💻Implementation
- collected datasets imitating ChatGPT for specific tasks or broadly imitating its behavior (0.3M–150M tokens).
- Fine-tuned LLMs (GPT-2 and LLaMA)
- Evaluated with Humans and GPT-4 (blind pairwise comparisons with ChatGPT) and on canonical NLP benchmarks
🧵 2/4

Introducing StarChat Beta β 🤖 Your new coding buddy 🙌Attention all coders and developers 💻 

You can write in plain English, and it will understand your queries, offer explanations, and provide step-by-step guidance to solve coding problems 🤯

👉 huggingface.co/spaces/Hugging…
🧵1/4 StarChat can help you:
🙋🏻‍♂️ Answer coding questions in over 80 languages, including Python, Java, C++ and more!
🧠 Explain concepts and help debug your code
📊 Generate sample code for data visualizations and plots in Python
💬 Iterate together to solve your coding errors

🧵2/4