merve

I was thinking of buying a teenage engineering sampler as I love their stuff then I remembered I have a rabbit r1 I'm thinking of installing llama.cpp on it but also doing s2s apps tbf initially I wasn't as interested but now they offer you to vibe code apps, use community creations on device or jailbreak which to me is very cool probably need to upgrade the os first disclaimer: this was sent to me for free but it's not a paid ad or whatever, I do not do paid ads and never did

Fun and crazy napkin math: Meta, NVIDIA, Alibaba, DeepSeek, Moonshot, Mistral and ~30 other labs have shared close to $1B worth of training compute on the Hugging Face Hub Free to download End of 2023 it was under $40M Wikipedia was written in volunteer hours. This AI age commons is written in GPU-hours and data (it's a floor btw: about half comes from official compute numbers rest is educated guess and it still miss the ~900k community fine-tunes) Space with data to explore: thomwolf-hub-dollars.static.hf.space

Learn about how to get multipliers in local LLM speed-ups using speculative decoding in this livestream with @mervenoyann, @MaartenGr and @_EldarKurtic youtube.com/watch?v=H4azmWcqO5g

Big shoutout to all of our partners for making Nemotron 3.5 Super VL their own through post-training 🙌 Here’s what they’ve built 🧵

Beam is now available across platforms, with open weights under the Apache 2.0 license - in collaboration with an ecosystem of 40+ partners. We're releasing: - Weights, tokenizer, and reference implementation - BF16, FP8 and NVFP4 weights and a DFlash2 drafter - The open source harness we built and trained Beam on: Mirror CLI Weights and model card on Hugging Face: huggingface.co/reflection

And of course Muse Spark Open works great with llama.cpp 🦙🔥 Behold this amazing demo courtesy of @mervenoyann, running on a Mac (M3 Ultra 512). This is a frontier model that runs on your (beefy) desktop. Local AI is only getting better 🚀

Yesterday and today I worked on the new "decision" pipeline to support all `jev`-like models in `transformers`. This will bring good standards in the mayhem that the ecosystem is now 👀 Meaning support for all jev/clef/von/etc via the simple and efficient `transformers serve`

So many models launching this week! And we have one of our own coming...⌛️

Learn about how to get multipliers in local LLM speed-ups using speculative decoding in this livestream with @mervenoyann, @MaartenGr and @_EldarKurtic youtube.com/watch?v=H4azmWcqO5g

I turned @googlegemma EmbeddingGemma 2, an embedding model, into a speech-to-text model, and it turned out to be one of the fastest STT models you can run on device. On a base M3 MacBook Air it transcribes an hour of audio in about 13 seconds: 288× real time. The idea came in March with Gemini Embedding 2. It puts text, images, audio and video into one shared space, so a spoken word lands right next to the written word. If that's true, you should be able to read a transcript straight out of the embedding model. I couldnt tried it then, but last week EmbeddingGemma 2 came out as an open model, so I tried it.