"There has been a shift in the #AI space: some models, like #ChatGPT & #Gemini, have evolved into entire web platforms spanning multiple use cases & access points. Other large language models like #LLaMa or #OLMo, though technically speaking they share a basic architecture, don’t actually fill the same role. They are intended to live in the background as a service or component, not in the foreground as a name brand." https://techcrunch.com/2024/04/19/too-many-models/
With all the valid concern around #llm and #genai power and water usage, I thought I'd start a blog series on tiny LLMs. Let's see what they can do on real tasks on very power efficient hardware.
The impact from smaller opensource LLMs like Llama3-8B and Phi-3 could be large. They are not necessarily the best and smartest models but can be easily integrated in software on every device and platform. Also they can be finetuned, improved with RAG to function better for specific tasks and in specific contexts. Exciting times ahead. #opensource#LLM#AI#Llama#Phi
⚠️ @forrestbrazeal on the inside threat to OSS
🍴Vicki Boykis says Redis is forked
👻 @johnonolan says Ghost is federating
🦙 Meta Engineering announces Llama 3
❓ @eieio's questions to ask when you don't want to work
🎙 hosted by @jerod
Meta released today Llama 3, the next generation of the Llama model. LLama 3 is a state-of-the-art open-source large language model. Here are some of the key features of the model: 🧵👇🏼
A major release to Ollama - version 0.1.32 is out. The new version includes:
✅ Improvement of the GPU utilization and memory management to increase performance and reduce error rate
✅ Increase performance on Mac by scheduling large models between GPU and CPU
✅ Introduce native AI support in Supabase edge functions
After months of work and $10 million, Databricks has unveiled DBRX - the world's most potent publicly available open-source large language model.
DBRX outperforms open models like Meta's Llama 2 across benchmarks, even nearing the abilities of OpenAI's closed GPT-4. Novel architectural tweaks like a "mixture of experts" boosted DBRX's training efficiency by 30-50%.
Please, use #AI to generate tons of #content that you otherwise couldn't.
But for the love of all that is holy, pay attention to what you are putting out. Read the output. If it doesn't say exactly what you would say, edit it! Make changes. Regenerate. Go through the process of making it good.
I truly don't think people hate AI content. They hate lazy content.
The Code Llama 34b model isn't half bad! Been toying around with it integrated into clion having it explain my own code to me and generate small functions and it's been so far around 90% successful, with most of the errors being minor, the bug detection does have a decent amount of false positives though. I also like that it's aware enough of api's to give doc links
Bonus points for it going off on a tangent once on why console applications are better than gui.
So #Steeve got a major upgrade recently. He moved from a #gptneo (2.4B) model to a #llama2 (7B) model. Trained on 300k messages from our private chat history, Steeve is way more capable of following the conversation now. He used to have some "favorite phrases" he would say a lot, and I'm seeing less of that. His vision and reading models also got upgraded, so he gets more detail about the links and memes we share. Long live Steeve! :steeve: