So-called reasoning AI models are becoming easier—and cheaper—to develop.
on Friday, NovaSky, a team of researchers based out of UC Berkeley's Sky Computing Lab, has released a reasonably priced model that competes with the Sky-T1-32B-Preview. Early version of OpenAI's o1 By many key standards, Sky-T1 appears to be the first truly open-source rational example in the sense that it could be. Replicated from scratch.; The team released the data set used to train it and the necessary training codes.
“Surprisingly, the Sky-T1-32B-Preview is trained for under $450,” the team wrote. Blog post“Demonstrates that it is possible to affordably and efficiently replicate high-level reasoning capabilities.”
I don't think I can afford $450. But it hasn't been long before the price went up to train a model with comparable performance. Often millions of dollars apart..
Unlike most AIs; Reasoning models effectively check themselves for truth. It helps them avoid some of the pitfalls that models typically run into.. Reasoning patterns typically take seconds to minutes longer to arrive at answers — compared to non-normal reasoning patterns. The advantage is, They are physics, They tend to be more confident in areas such as science and math.
The NovaSky team uses another form of reasoning. Alibaba's QwQ-32B-Preview.to generate initial training data for Sky-T1; The data mixture was then “curated” and leveraged by OpenAI. GPT-4o-mini To reformat the data into a more workable format. Training the 32 billion-billion-parameter Sky-T1 took about 19 hours using 8 Nvidia H100 GPUs. (Limits are roughly related to a model's problem-solving ability.)
According to the NovaSky team, Sky-T1 outperformed an earlier preview version of o1 in MATH500, a collection of “competitive-level” math challenges. The model also outperforms o1's preview on a difficult problem set from LiveCodeBench, a code evaluation.
However, Sky-T1 is physics that a PhD graduate would probably know. Failed over o1 preview in GPQA-Diamond with biology and chemistry questions.
Important to note is OpenAI. GA edition of o1 It is a more powerful model than o1's preview version, and OpenAI is expected to release a reasoning model with better performance. o3In the weeks ahead,
But the NovaSky team says Sky-T1 is just the beginning of their journey to develop open source models with advanced reasoning capabilities.
“Going forward, we seek to develop more efficient models that maintain robust reasoning performance, and to explore advanced techniques that further improve the performance and accuracy of the models during testing,” the team wrote in the post. “Stay tuned as we make progress on these exciting initiatives.”