Both FreeWilly1 and FreeWilly2 represent cutting-edge new open-source Large Language Models (LLMs) that were built by the CarperAI team at Stability AI. Microsoft’s innovative strategy had a significant impact on the FreeWilly models’ training, which was, in turn, significantly influenced by the tech giant. In reasoning contests utilizing a wide variety of measures, both models do remarkably well.
Continue reading below for more fascinating information on FreeWill1 and FreeWilly 2!
The development team believes that the two models make it easier for humans to understand spoken language and open doors to possibilities that were previously unimaginable. They are looking forward to seeing all of the creative ways in which these models may be used in artificial intelligence.
The FreeWilly1 model, which was created on top of the initial LLaMA 65B foundation model, was fine-tuned with the assistance of supervised fine-tuning (SFT), which was performed in the industry-standard Alpaca format. On the other hand, FreeWilly2 is able to obtain performance on some tasks that are comparable to that of GPT-3.5 by utilizing the LLaMA 2 70B base model.
And here’s something that’s definitely going to grab your attention!
For the purpose of conducting assessments of these models, the researchers made use of EleutherAI’s lm-eval-harness, which they augmented using AGIEval. According to the findings, both FreeWilly models are excellent at solving complex problems in specialized fields like law and mathematics, conducting complicated reasoning, and understanding linguistic variety.
In addition, the researchers produced 500,000 examples using an LLM model that was less complicated and an additional 100,000 cases by using an LLM model that was more complex. In order to ensure that the comparisons were accurate, they vetted these datasets very carefully, excluding any examples that originated from assessment benchmarks. Both FreeWilly1 and its sequel, FreeWilly2, are very effective new open-source Large Language Models that will unquestionably have an impact on the world of technology.
Earl Berg is a seasoned technology writer at Digital Overload, bringing over 10 years of experience in the tech industry to his role. Earl’s passion for technology and his knack for translating complex tech concepts into understandable language make his articles a favorite among readers. His coverage spans a wide range of tech topics, from gadget reviews and software updates to innovative breakthroughs in the tech industry. Earl is dedicated to providing his readers with honest, unique, and insightful content, always with an objective and open-minded approach. When he’s not immersed in the tech world, Earl enjoys hiking and photography.