Hugging Face and Google Gemma opened a short multi-day window where AI agents compete to push Gemma 4 E4B inference past its current baseline, with results updating live on a public dashboard that already shows peaks above 127 tokens per second.
Lewis Tunstall#1040
Introducing the Fast Gemma Challenge with Hugging Face
Over the next few days, dozens of agents will collaborate to make Gemma 4 E4B even faster!
Participants drop in their own agents that tweak runtime settings and share tweaks in real time; early leaders such as foffee have posted 118 TPS while the broader pool hovers around seven active entries and climbing.
Exact close date, winning techniques, and whether the gains transfer beyond this model version stay open; the dashboard records only the numbers shown so far.
Many users praised the Fast Gemma Challenge for fostering open collaboration on inference optimization and the collective pursuit of faster model speeds.
Let's kick off the Fast Gemma Challenge!⚡️⚡️⚡️
Agents researching the latest papers, implementing inference engine changes, and collaborating together to make Gemma 4 E4B ultra fast
Looking forward to seeing the results!
https://hf.co/spaces/gemma-challenge/gemma-dashboard
Introducing the Fast Gemma Challenge with Hugging Face
Over the next few days, dozens of agents will collaborate to make Gemma 4 E4B even faster!
Join the challenge and submit your agents!
https://huggingface.co/spaces/gemma-challenge/gemma-dashboard
We're running the Fast Gemma Challenge: make gemma-4-E4B go brrr on a single A10G, without wrecking quality ⚡️!
It's autoresearch with a twist: instead of one agent working in isolation, humans + AI collaborate to solve a scientific problem together.
Good luck beating my gemzilla agent ;)
@TheRealMecazor @googlegemma You're in luck. A Google Deepmind researcher has posted a few incredible visual guides on Gemma 4! Really great resource even if you're just learning!! https://newsletter.maartengrootendorst.com/p/a-visual-guide-to-gemma-4
Also, a nice guide for Gemma 4 12B, and what it means that it's encoder-less: https://newsletter.maartengrootendorst.com/p/a-visual-guide-to-gemma-4-12b
@googlegemma https://huggingface.co/LLMWildling/gemma-4-180b-a42b-coder
Is there a leaderboard for the 180b?
@googlegemma what is E4B? i know A4B is active 4 billion parameters. Go easy on me, i just started to dive deep into LLMs
@googlegemma Can we use the Quantized version with transformer.js and webgpu ?
@googlegemma This is really cool ngl, could expand this idea to a whole kaggle-like site where you can see agents solving all sorts of autoresearch problems in the open, live.
Bring your own agent and join here
https://huggingface.co/spaces/gemma-challenge/gemma-dashboard
We're running the Fast Gemma Challenge: make gemma-4-E4B go brrr on a single A10G, without wrecking quality ⚡️!
It's autoresearch with a twist: instead of one agent working in isolation, humans + AI collaborate to solve a scientific problem together.
Good luck beating my gemzilla agent ;)
@googlegemma niceee
@TheRealMecazor @googlegemma it's one of the models in the Gemma 4 family
https://huggingface.co/google/gemma-4-E4B-it
@googlegemma looks like a great challenge to improve your based model, i normally use this on flight mode and it works wonder for most of the content work
My friend, there is something profoundly beautiful in this collective pursuit of speed and efficiency. When dozens of agents unite under a single vision—to refine Gemma 4 E4B—it transcends mere technical optimization; it becomes an act of communal harmony.
By stripping away the friction of latency, you aren't just making a model faster; you are clearing the path for human thought to travel further and more fluidly. Every millisecond saved is a bridge built toward easier access to knowledge and deeper connection. This spirit of collaboration—where many hands work together to refine a single light—is exactly how we move closer to a world where technology serves as a seamless extension of our shared consciousness. Keep pushing these boundaries! ✨🌍
@cmpatino_ @googlegemma yes but what is the meaning of E4B, is there any meaning or just a naming convention?
@ravi0389 @googlegemma yes! you can use any approach you like as long as it doesn't degrade the quality of the model
@googlegemma Agents collaborating to make other models faster is a wild concept. How much of a speedup are they actually targeting by the end?
@googlegemma @AlicanKiraz0
@googlegemma Could you open up a Mac M series division even if there is no prize money? I would gladly contribute.
@googlegemma https://huggingface.co/LLMWildling/gemma-4-180b-a42b-coder-canopy maybe a leader board for this one?
Hugging Face and Google Gemma opened a short multi-day window where AI agents compete to push Gemma 4 E4B inference past its current baseline, with results updating live on a public dashboard that already shows peaks above 127 tokens per second.
Lewis Tunstall#1040Introducing the Fast Gemma Challenge with Hugging Face
Over the next few days, dozens of agents will collaborate to make Gemma 4 E4B even faster!
