
Jesse Bunch
1.4K posts

Jesse Bunch
@bunchjesse
Dad. Husband. Commercial Pilot. Engineer. Productivity Applications at .


Difference in quality between NASA and SpaceX’s launch livestreams.

Calling it now: all these agent coding TUIs are a phase and it will be short lived. Most devs will be back in GUIs and IDEs in a few months.


nanochat can now train GPT-2 grade LLM for <<$100 (~$73, 3 hours on a single 8XH100 node). GPT-2 is just my favorite LLM because it's the first time the LLM stack comes together in a recognizably modern form. So it has become a bit of a weird & lasting obsession of mine to train a model to GPT-2 capability but for much cheaper, with the benefit of ~7 years of progress. In particular, I suspected it should be possible today to train one for <<$100. Originally in 2019, GPT-2 was trained by OpenAI on 32 TPU v3 chips for 168 hours (7 days), with $8/hour/TPUv3 back then, for a total cost of approx. $43K. It achieves 0.256525 CORE score, which is an ensemble metric introduced in the DCLM paper over 22 evaluations like ARC/MMLU/etc. As of the last few improvements merged into nanochat (many of them originating in modded-nanogpt repo), I can now reach a higher CORE score in 3.04 hours (~$73) on a single 8XH100 node. This is a 600X cost reduction over 7 years, i.e. the cost to train GPT-2 is falling approximately 2.5X every year. I think this is likely an underestimate because I am still finding more improvements relatively regularly and I have a backlog of more ideas to try. A longer post with a lot of the detail of the optimizations involved and pointers on how to reproduce are here: github.com/karpathy/nanoc… Inspired by modded-nanogpt, I also created a leaderboard for "time to GPT-2", where this first "Jan29" model is entry #1 at 3.04 hours. It will be fun to iterate on this further and I welcome help! My hope is that nanochat can grow to become a very nice/clean and tuned experimental LLM harness for prototyping ideas, for having fun, and ofc for learning. The biggest improvements of things that worked out of the box and simply produced gains right away were 1) Flash Attention 3 kernels (faster, and allows window_size kwarg to get alternating attention patterns), Muon optimizer (I tried for ~1 day to delete it and only use AdamW and I couldn't), residual pathways and skip connections gated by learnable scalars, and value embeddings. There were many other smaller things that stack up. Image: semi-related eye candy of deriving the scaling laws for the current nanochat model miniseries, pretty and satisfying!

Running Kimi K2.5 on my desk. Runs at 24 tok/sec with 2 x 512GB M3 Ultra Mac Studios connected with Thunderbolt 5 (RDMA) using @exolabs / MLX backend. Yes, it can run clawdbot.



in opencode v1.1.11 you can now use your ChatGPT Plus/Pro plans in OpenCode /connect to set it up

Introducing: The New Pyramid

Announcing a profound partnership between National Design Studio and Sec Kennedy, Sec Rollins, CMS Administrator Oz, and FDA Commissioner Makary to bring the new Dietary Guidelines to life in the form of the New Pyramid. Shocking how misleading the one hanging in my second grade classroom was. Even more surprising would have been to tell my younger self one day I’d help redesign it. Eat whole foods, and enjoy realfood.gov @ndstudio @SecKennedy @SecRollins @DrOzCMS @DrMakaryFDA @PressSec @WhiteHouse @RapidResponse47








