Last call to register for the ElevenLabs hackathon!
Join us for a weekend of app coding with voice AI on July 28-30 to win cash and free access prizes!
Register: https://t.co/wSOqyfQ4t0
Check the tutorials from @lablabai to learn about some of the amazing apps you can build!
Introducing FreeWilly1 and FreeWilly2 - The latest groundbreaking LLMs from Stability AI's and @carperai lab!
⭐Open access and remarkable versatility.
#AI#LLMs#OpenAccess#StabilityAI
https://t.co/K5KHCg9aIt
Another LLM candy just dropped. 🍬
Stability. AI just released to Huggingface two models comparable to ChatGPT!🔥
--- TL;DR ---
Stability AI just announced FreeWilly1 and its successor FreeWilly2, two powerful new, open access, Large Language Models (LLMs).
Both models demonstrate exceptional reasoning ability across varied benchmarks.
FreeWilly1 leverages the original LLaMA 65B foundation model and was carefully fine-tuned with a new synthetically-generated dataset using Supervised Fine-Tune (SFT) in standard Alpaca format.
Similarly, FreeWilly2 leverages the LLaMA 2 70B foundation model to reach a performance that compares favorably with GPT-3.5 for some tasks.
Both models are research experiments, and are released to foster open research under a non-commercial license. 🤝
--- Data ---
The training for the FreeWilly models was directly inspired by the methodology pioneered by Microsoft in its paper: "Orca: Progressive Learning from Complex Explanation Traces of GPT-4.” While the data generation process is similar, it differs in its data sources.
Their variant of the dataset, containing 600,000 data points (roughly 10% of the dataset size the original Orca paper used), was created by prompting language models with high-quality instructions from the following datasets created by Enrico Shippole:
1. COT Submix Original (https://t.co/QYaVBzu9lG) 📚
2. NIV2 Submix Original (https://t.co/eY1xfUt079) 📘
3. FLAN 2021 Submix Original (https://t.co/pIqjnbAlny) 📙
4. T0 Submix Original (https://t.co/KWqQupT5c8) 📕
With this approach, they generated 500k examples with one simpler LLM model and an additional 100k with a more sophisticated LLM model.
To ensure fair comparisons, they carefully filtered these datasets and removed examples that originated from evaluation benchmarks. Despite training on 0.1 the sample size of the original Orca paper (significantly reducing the cost and carbon footprint of training the model compared to the original paper), the resulting FreeWilly models demonstrate exceptional performance across various benchmarks – validating their approach to synthetically generated datasets. 👏
--- Evaluation ---
To internally evaluate these models, they used EleutherAI’s lm-eval-harness, to which they added AGIEval.
Both FreeWilly models excel in many areas, including intricate reasoning, understanding linguistic subtleties, and answering complex questions related to specialized domains, e.g. Law and mathematical problem-solving.🏆
--- Links ---
1. Blog - https://t.co/P3ATWvCbX7
2. FreeWilly 1 (LLaMA 1) - https://t.co/10FggzNAhl
3. FreeWilly 2 (LLaMA 2) - https://t.co/w8ZjcLsQyh
GM from my daily swim spot in Key West.
This place allows me to hang out with fish that have literally grown to know me. Squid that accept me into their pod and wave their arms at me. Momma & baby manatees that gave me 45 minutes to join them on a lazy swim. Even stung by a ray.
14 yo and 11 yo showed up at my office door asking for startup advice. I told them that the way to get really rich was to get good at programming now and start the startup in a few years, when knowing how to program had started to make good ideas visible.
I feel personally attacked @daltonc 🤣
Does it count if I immediately used @bing & @openai to help teach myself Python, MLPs, Deep Learning & tutor me during the AI @MicrosoftLearn courses I take?
P. S. Launch is in a few days, locals & tourists are exciting to get on it!
The messages of excitement for the launch of @KeySaverKW is the gas to keep my fire going
I'm running on empty atm my teeney team is new to tech, mostly self taught & yet overcame every hurdle. Huge thanks to @OpenAI@Azure@MicrosoftLearn@Microsoft
Just an example of how advertising will be changed forever.
By fine-tuning my AI she now responds to users questions adding the companies we've specially selected to be mentioned.
I said " Hi, Mermaid Molly, it's me Ayla. I want to party at the sandbar, where should I go?
Just another history making moment 👀
Advirtising w KeySaver will be immersive, entertaining and informative... Providing value.
Powered by @OpenAI Gpt4, @elevenlabs