Image created with Gemini. Image prompt: A flat saturated acrylic illustration of a small backyard scene assembled from a grid of slightly offset overlapping photo panels, showing a turquoise swimming pool with hand-drawn white squiggle ripples in the foreground and a beige desktop server tower sitting on bright green lawn beside it with a cable running to an outdoor outlet, hard-edged shapes with no shading, high noon light with almost no shadows, terracotta pink patio tiles and hot yellow sun, cheerful open composition with a high horizon.
It’s very cool that Apple shipped a 20B parameter on-device. You can’t put 20B parameters in RAM at any reasonable precision. To make it work they are using pretty exotic architecture by today’s standards. A small model predicts from the query (or prompt) which experts to load
https://x.com/awnihannun/status/2064202168618422396
DiffusionGemma is out 🔥 it’s compute-bound so 4x faster compared to other Gemma-4 models (1k tok/s on H100) 💨 also great on coding, generate and iterate on any code from 3D generation to front-end ⤵️
https://x.com/mervenoyann/status/2064753402064601181
Awesome to see this innovation in text diffusion. DiffusionGemma is lightning fast, 4x faster than other Gemma 4 models! Congrats to @bodonoghue85 and the team who worked so hard on this – excited to see what people build with it!
https://x.com/demishassabis/status/2064873362799600042
DiffusionGemma is an open, experimental model that brings our text diffusion research to Gemma 4. It’s a racehorse 🏇achieving up to 4x faster inference by generating entire blocks of text simultaneously vs predicting token-by-token (word-by-word) output!
https://x.com/sundarpichai/status/2064744343743922189
Introducing DiffusionGemma
https://blog.google/innovation-and-ai/technology/developers-tools/diffusion-gemma-faster-text-generation/
Five labs, five minds: building a multi-model finance drama on small models
https://huggingface.co/blog/build-small-hackathon/thousand-token-wood-sim-v2
Last time around Apple released a lot of information about how their AI version of Siri worked between local and cloud models, not so much this time It is nice to have a Gemma-like model on device, but it is extremely limited unless it can call a smarter cloud model when needed.
https://x.com/emollick/status/2064052841367392536
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
https://x.com/_lewtun/status/2064386398090576236
We believe openness drives innovation. Both fp8 and nf4 checkpoints are in our repo, with the nf4 variant fitting on a single 24 GB GPU. Huggingface:
https://t.co/SNEh1yZsq8 Github:
https://t.co/mYuEfS2L9n Blog:
https://x.com/ideogram_ai/status/2062956472489922584
🧠 Gemma 4 QAT checkpoints are out, and vLLM is Google’s recommended way to serve them! Open-source inference is at its best when one engine spans research and production — glad vLLM’s is the recommendation for Gemma 4 QAT. Get started 👇
https://x.com/vllm_project/status/2062938949560283216
Building super fast experiences with Gemma just got easier. Gemma 4 MTP is now officially merged into llama.cpp. Developers can now pair MTP with Gemma 4 QAT for a fast, lightweight setup.
https://x.com/googlegemma/status/2064030477628182814
bullish on Gemini
https://x.com/OfficialLoganK/status/2063819854348697681
Gemma 4 now runs 2x faster with MTP GGUFs! Run locally on just 6GB RAM. ⚡️ MTP enables Google Gemma 4 run ~1.4-2.2× faster with no accuracy loss. Gemma 4 12B MTP can run at 162 t/s vs. 52 t/s without MTP. 31B reaches 101 t/s. GGUFs + Guide:
https://x.com/UnslothAI/status/2065107734916432189
Gemma 4 Quantization-Aware Training (QAT) weights are now available on Ollama! They reduce memory requirements while maintaining model quality. E2B: ollama run gemma4:e2b-it-qat E4B: ollama run gemma4:e4b-it-qat 12B: ollama run gemma4:12b-it-qat 26B: ollama run
https://x.com/ollama/status/2062965815864066079
Gemma 4 with quantization-aware training
https://blog.google/innovation-and-ai/technology/developers-tools/quantization-aware-training-gemma-4/
Gemma goes diffusion! DiffusionGemma with up to 1000+ tokens per second! 🌬️ – Built on Gemma 4 as a 26B MoE model. – 3.8B parameters during inference. – Generates text in 256-token blocks in parallel. – Fits within 18 GB VRAM limits when quantized. – Apache 2.0
https://x.com/_philschmid/status/2064745464252055647
Gemma-4 QAT just dropped! We found if you naively convert from QAT Q4_0 BF16, you will lose accuracy since the conversion to llama.cpp has a different lattice. Unsloth dynamic GGUFs recovers most of it! 26B-A4B: 85.6% top-1 % from 70.2% (+15.4%) 31B: 96.7% from 87.9% (+8.8%)
https://x.com/danielhanchen/status/2062933017430315481
Google releases DiffusionGemma.✨ The new 26B-A4B diffusion text model runs locally on 18GB RAM. It supports high-speed text generation, thinking, image, video and 256K context. Run and train via Unsloth Studio. GGUF:
https://t.co/ZH0dCJQ59P Guide:
https://x.com/UnslothAI/status/2064743714875220118
Introducing Gemma 4 QAT 🤏 – Quantization aware training to reduce models’ precision while preserving quality – Introducing a new mobile quantization format that reduces memory footprint of E2B to 1GB – Q4 for all your favorite libraries ✨
https://x.com/osanseviero/status/2062933011415392482
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!
https://x.com/googlegemma/status/2064374874962117084
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://x.com/osanseviero/status/2064375902046245219
llama.cpp just added video input support 👀 You can now enjoy Gemma 4 video understanding capabilities in your chat completions endpoint and via mtmd-cli
https://x.com/osanseviero/status/2063985470489448887
Meet DiffusionGemma ⚡ Our latest experimental open model (Apache 2.0) that generates text up to 4x faster. Instead of predicting and typing just one word at a time like most language models, it drafts and refines entire blocks of text simultaneously. Here’s how it works 🧵 ↓
https://x.com/Google/status/2064741293163418032
More Gemma 4! New QAT Gemma 4 checkpoints with similar performance while using ~4x less memory! It comes with a new mobile quantization format that reduces memory footprint of Gemma 4 E2B to just 1GB. Quantization-Aware Training (QAT) simulates low-precision operations during
https://x.com/_philschmid/status/2063990553826439378
We just dropped Gemma 4 Quantization-Aware Training (QAT) checkpoints on Hugging Face! All Gemma 4 model sizes and their drafters are now optimized with QAT to cut memory requirements and maximize on-device performance!
https://x.com/googlegemma/status/2062928831229665566
Congrats to @GoogleDeepMind on DiffusionGemma 🎉 A 26B diffusion language model on the Gemma4 backbone, and the first dLLM natively supported in vLLM. It denoises 256-token blocks in parallel instead of generating one token at a time: 1200+ output tok/s at batch size 1 on a
https://x.com/vllm_project/status/2064753414735900835
this model is the opposite of mythos. Its small, cost effective, apache 2.0, and locally deployable. This is the way LLMs should go. small, open source, transparent and sovereign vs large, expensive, proprietary and hegemonic
https://x.com/nickfrosst/status/2064396337404096809
We made DiffusionGemma run via llama.cpp locally! It works well with Unsloth GGUFs and you can run it in realtime visualization mode or normal chat CLI mode! See our docs
https://t.co/IslbgeCs7Z on how to set it up!
https://x.com/danielhanchen/status/2064760001567306232
DiffusionGemma is our new experimental open model with up to 4x faster output on dedicated GPUs. Instead of predicting word-by-word, it generates entire blocks of text simultaneously. This lets the model self-correct and format complex markdown in real time.
https://x.com/GoogleDeepMind/status/2064741061352636762
DiffusionGemma is so fast that we had to slow down the videos so people could see what was happening
https://x.com/osanseviero/status/2065041448135770436
Meet DiffusionGemma! An experimental open model that explores a fast approach to text generation, released under an Apache 2.0 license. Moving beyond sequential, token-by-token processes to generate entire blocks of text simultaneously. Here’s what’s new with DiffusionGemma: 👇
https://x.com/googlegemma/status/2064741002204545467
Leave a Reply