Officially announcing: Finch × Baidu AI Cloud We're partnering with @Baidu_Inc AI Cloud to advance the AI agent economy, combining its AI capabilities and industry expertise with Finch's platform and developer ecosystem. Our collaboration begins with model integration through Qianfan, Baidu AI Cloud’s MaaS platform. Together, we’ll explore new business models and industry applications for AI agents, and build an open, thriving ecosystem where developers, businesses, and partners can create value. We’re building the agent economy, together.

The Meet Manus 2.0 event series is going global - online and in person. Join a Founder Briefing, Roadshow, or a Fellow-led Community Meetup to see the product in action and hear the thinking behind it. First up: a Founder Briefing webinar with Co-founder & CPO @hidecloud on Sep 30 at 10 PM SGT. RSVP: https://luma.com/essfrk2k More events in the thread below ↓

New historic NanoGPT record at 39.9s (-27.7s) from @DevenPzak , obliterating the prior record of 67.6s! This record introduces a new paradigm of thinking to NanoGPT: instead of optimizing matmuls or adding more expressive operations, optimize at the individual flop level with incredibly clever engineering and ML judgement. If a flop is low value on a particular step, skip it. Specifically: -(~8s) Sampled softmax. If a token doesn’t appear in a batch, skip its lm_head fwd/bwd some fraction of the time. -Sparse values. Only run an optimizer step for ngram embeddings that occurred in the batch. Set beta1 to zero to enable this. Beta2 is applied retroactively when the row is later used. -Sparse updates. Only update ngram and value embeddings once every 4 steps instead of once every 2. -Sparse communication. Shard the n-gram table across GPUs, and only pass the rows receiving updates on each step. -Sparse optimizer states. For the n-gram table, reduce from 2 floats in Adam optimizer per param, to 1 float per 768 params. -Hand-rolled flash attention for 64 dim heads. There are several additions that add accuracy too: -(~4s) EMA during last 300 steps, combined with lifting final_lr to 0.3 instead of 0.15. -(~1s) A new optimizer, Anvil2, which expands muon via a second tracked momentum buffer, improves the ortho coefficients, and modifies the cautious weight decay application. -A couple additional dynamic skip connections in the network. The most striking consequence of the ‘flop aware paradigm’ is you can grow parameters arbitrarily large, only limited by the available memory, since you can selectively choose how to expend flops on those parameters on each step. NanoGPT has kept active parameters below 124M, but total is unbounded, and has grown to 640M through embedding sparsity over the last year. This PR takes that to its logical conclusion on the 8xH100, scaling up to 65B sparse embedding parameters, which accounts for 25% of the PR’s gains. At frontier scale, where one is not bounded by an 8xH100, one could imagine where this paradigm could lead. https://github.com/KellerJordan/modded-nanogpt/pull/360 As this was a very notable PR, I spoke with Deven for an hour to learn how he did it. Here’s his story on the changes: https://hyperstition.cc/training-nanogpt-in-39-9-seconds

So excited to welcome @theworldlabs and @drfeifei to the @AMD family! I’ve always been a huge fan of Fei-Fei and her pioneering research in AI. Together, we’ll combine World Labs’ deep expertise in AI and world models with AMD’s compute leadership to power the future of AI and strengthen the open AI ecosystem. Can’t wait for all we’ll accomplish!