J Feldman

Today we're thrilled to announce the first end-to-end platform for Reinforcement Fine-Tuning. With just a dozen labeled data points, you can outperform #OpenAI o1 and #DeepSeekR1 on complex tasks. Built on the #GRPO methodology that DeepSeek-R1 popularized, our platform delivers exceptional results. In our real-world PyTorch to Triton transpilation case study, we achieved 3x higher accuracy than OpenAI o1 and DeepSeek-R1 when writing GPU code. Check out the thread below to learn how you can adapt an #opensource #LLM to your use cases with unmatched efficiency. #rft

Want AI that speaks your language? Fine-tuning is the spark you need. Essentially, you’re tailoring an off-the-shelf model to your precise goals. The catch? Historically, it took an avalanche of labeled data—thousands of samples—to train your ‘old dog’ to perform a brand-new trick. I was getting tired of labeling thousands of samples and the cost. And then I looked into Reinforcement Fine-Tuning (RFT) from @predibase . 🧵1/n With RFT, a lightweight opensource LLM can rapidly evolve into an exceptional problem-solving machine. 📚 Reinforcement Method Predibase's Reinforcement Fine-Tuning (RFT) addresses the constraints of classic supervised approaches. It systematically applies a reward function to guide model updates. - Advanced policy gradient methods that shorten iteration cycles. - This setup surpasses standard fine-tuning by providing precise feedback and higher accuracy with minimal labeled data. ⚙️ Minimal Labeled Data - RFT excels with fewer than 100 labeled examples. - It discards the old requirement for massive datasets by validating responses against a reward metric. - This approach cuts data-collection costs significantly. - Small, controlled feedback loops sharpen performance in logical or multi-hop tasks. 🔗 Chain-of-Thought Boost Chain-of-thought integration refines step-by-step reasoning. RFT checks partial correctness and then readjusts updates to strengthen valid outputs. This self-correcting mechanism limits error propagation in arithmetic or combinatorial tasks. Iterative feedback enables the model to refine its own reasoning. 🚀 Integration and Impact The RFT pipeline is seamlessly integrated within the Predibase platform. - Debugging tools help trace reward distribution, and distributed training supports larger-scale tasks. - Models can be deployed or tracked in a serverless manner without hardware overhead. - Adaptive reward shaping and automated checkpointing accelerate development. - RFT extends fine-tuning capabilities for LLMs in limited-data domains. - This eliminates hefty labeling expenses while sustaining performance gains. - It directly tackles situations where correctness is measurable but labeled data is scarce. ------ The below image is from their official technical report (link in comment). RFT leads the pack at each data scale. Note the jump in performance when training with just 10 or 100 samples.

Huge release from @Predibase today -- the first end-to-end platform for Reinforcement Fine-Tuning Bringing the techniques that power DeepSeekR1 to any open source model and data

New #WirelessQ! Read how: 🔥 how #IoT promises quicker detection of wildfires 💰 sensors tech will continue to become cheaper, more advanced and more widely available 🧸#Wireless tech helps create innovative products to engage and educate children bit.ly/3nroXmT📚

@united waited on hold for an hour after being told 5 minute wait time only to have the agent hang up on me in the first 30 seconds. I understand flights are being cancelled but at least provide the customer support staffing to help your customers.

Had another great time talking with @embeddedfm, this time about firmware developer productivity within small and large organizations.

Our CustoMoose collection is dropping next week! Read about the details on our latest blog post: @moosetraxnft/customoose-is-here-95ecddf53578" target="_blank" rel="nofollow noopener">medium.com/@moosetraxnft/… Join us on Discord if you have any lingering questions about the drop!

Find out how Bluetooth Audio Sharing will bring new audio experiences to consumers all over the world: bit.ly/3jtKM62 #LEAudio

Join our partner webinar with @Memfault to see how #nRF91, #nRF53, and #nRF52 Series developers can access Memfault’s platform via the #nRFConnect SDK for free bit.ly/3Ab886c 👇

@OwnTheMomentNFT @JeudyJustice @Suburban_Eric

🎥SHOWCASE GIVEAWAY🎥 We understand influencers have an advantage. That's not fair so we're doing something about it. Whatever pack the OTM showcase wins, we'll gift the Moments to YOU! ✅RT this tweet ✅Like the showcase ✅Tag 2 friends in comments nbatopshot.com/showcases/7c99…

Need to build a CLI shell for your firmware project? Check out this guide from the Memfault team: interrupt.memfault.com/blog/firmware-…

Curious how breakpoints work when you're debugging? The Memfault team gives us an overview and explains Cortex-M hardware breakpoints. interrupt.memfault.com/blog/cortex-m-…

EXCLUSIVE: Privacy-centric web browser @Brave is raising a Series A round at a valuation of roughly $133 million, sources tell CoinDesk. ow.ly/qKom30oKHGr @BradyDale reports

Just 10 years ago first #Bitcoin transaction was sent and it’s astonishing to see industry growth and how many new assets have gained traction. In this unique visualization by DataLight, you can track the top-10 crypto assets by market cap, from crypto’s early days until today.

In a busy office and need to take a phone call? Next time, use this device.

Remember how we just announced the debut of Flashy Ways this morning? Well we found out at 10 am that it just won a Bronze Medal 🥉 at the 2018 New York Craft Brewers Competition and Governor’s Excelsior Cup! Stoooooooooked!!!!!

Buying beer with 0.001 Bitcoin using PundiX

#NPXS circulating supply has been verified and updated on @CoinMarketCap. You can check more detail info on coinmarketcap.com. #pundix #Cryptocurrency

242 wins. 21 seasons. Bartolo Colon is still going strong on his 45th birthday.