gemma-4-26B-A4B-it

recent items

• I implemented a modern LLM in 700 lines of C (www.reddit.comhttps) 3w

  I’ve been working on a small project called gemma4.c. The idea is pretty simple: you can download a modern language model, compile one 700-line C file, and have it generate text on an ordinary CPU.

  ↯ Gemma↯ Gemma 4↯ Gemma 4↯ Gemma 4↯ Gemma 4↯ Gemma 4gemma

• Gemma 4 E2B inference in 700 lines of C (github.com via hn) +4 3w

  gemma4.c Gemma 4 E2B CPU inference in 700 lines of pure C. An educational project made to understand how LLM inference works.

  ↯ Gemma↯ Gemma 4↯ Gemma 4↯ Gemma 4↯ Gemma 4↯ Gemma 4gemma

• 12 abliterated Gemma 4 12B variants, one base, 165 GPU hours - Abliterlitics (www.reddit.com via reddit) 3w

  I ran 11 uncensored variants of Gemma 4 12B that I grabbed from huggingface, sorting by downloads. 10 full abliterations plus 2 LoRA adapters which were requested to be added in the comparison, against the official base.

  ↯ Gemma↯ Gemma 4gemma

• What it takes to run Gemma 4 E4B on an iPhone (amoli.app via hn) +1 3w

  I spent the last several months building a photo journal that runs a full multimodal LLM — Gemma 4 E4B — entirely on an iPhone. No cloud inference, no accounts, no analytics, full privacy; the App Store privacy label is "Data Not Collected…

  ↯ Gemma↯ Gemma 4gemma

• Best model for 16gb ram Mac (www.reddit.com via reddit) 3w

  Hi everybody! Every now and then these days, we’re seeing really huge open-weight models popping up.

  ↯ Llama↯ Gemma↯ Gemma 4gemmallama

• Ling Tiny, King of Speed (www.reddit.comhttps) 3w

  Ling Tiny has now replaced Gemma4-12B in my rig as an auxiliary model doing hindsight operations. This is on a 4060Ti, which is a reasonable GPU available out there, and the speed is phenomenal.

  ↯ Vllm↯ Gemma 4vllm

• I fine tuned Gemma 4 12B for a 2.7x improvement on tool calling because I can't fit anything else comfortably into my 16 GBs of Vram (huggingface.co via reddit) 3w

  Gemma 12B is obviously a very well trained model, I always thought the fine tuning they did on it wasn't really cut out for agentic coding. From my own experiences it struggles to use the tools it's given from Github Copilot and is also ve…

  ↯ Copilot↯ Ollama↯ Llama↯ Gemma↯ Gemma 4ollamagemmacopilot+2

• Your own GGUF (www.reddit.com via reddit) 4w

  Hello, I have a few questions that I can't seem to find a clear answer to. Does it make sense to make your own GGUF?

  ↯ Llama↯ Gemma↯ Gemma 4gemmallama

• Gemma (github.com via hn) +2 4w

  Gemma is Google DeepMind's family of lightweight, state-of-the-art open models. Contents Start Here Models Core Models Variants Inference Local Hosted Fine-Tune Tutorials Demos and Applications Gemma 4 Good Challenge Gemma in Space Researc…

  ↯ Deepmind↯ Gemma↯ Gemma 4deepmindgemma

• Apertura A from-scratch Objective-C++ / MLX rebuild of Gemma-4 for Apple Silicon (github.com via hn) +11 4w

  Apertura A from-scratch Objective-C++ / MLX rebuild of Google's Gemma-4 for Apple Silicon — built to be inspected, observed, and experimented with. Most language models are run behind glass: you send a prompt and get an answer, with no way…

  ↯ Gemma↯ Gemma 4gemma

• Ask HN: Are third party gguf/mmproj safe on Llama in production env (news.ycombinator.com) +2 5w

  So i had been building a side project screenmind ,sort of local ai desktop assistant, and hit a confusing multimodel failure... texts worked,but vision and image failed no crash just returning <unsused49>.

  ↯ Llama↯ Gemma↯ Gemma 4gemmallama

• Claude Code plugin for translating from Claudish to English (www.reddit.comhttps) 5w

  I (like thousands of other engineers) got tired of trying to understand Claude's writing. ASD-STE100 instructions didn't change much, so I decided to approach the problem in the 2026 style: use an LLM.

  ↯ Ollama↯ Gemma↯ Gemma 4ollamagemmaclaude-code

• Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac (github.com via hn) +21653 7w

  Hi HN, I built a specialized inference engine for running 4-bit Gemma 4 26B-A4B-IT on any M-series Mac using about 2 GB of RAM. It is called TurboFieldfare and is written in Swift and Metal.

  ↯ Gemma↯ Gemma 4gemma

• Show HN: Gemma 4 26B A4B running on an iPhone 17 Pro via model paging (noemaai.com via hn) +2 7w

  I wanted to share a use case for Noema Overfit, a model-paging system available today in the Noema app. For disclosure, I founded Noema and I am part of the team that has helped develop the system.

  ↯ Gemma↯ Gemma 4gemma

• Benchmarking Fine-tuning and Retrieval Strategies for a Multimodal Language Model on the NRC Reactor Operator Licensing Examination (arxiv.org) 7w

  The integration of large language models (LLMs) into the nuclear power industry requires outputs grounded in domain-specific knowledge. This study evaluates a 31-billion-parameter open-weight multimodal model (Gemma 4 31B-IT) on its capaci…

  ↯ Fine Tuning↯ Gemma↯ Gemma 4operatorgemmafine-tuning

• Command-hook compliance gate: Cursor rules + Ollama, no cloud judge (www.reddit.com via reddit) 7w

  Cursor type: "prompt" hooks always use Cursor’s own model. No model / baseUrl — you can’t point them at Ollama.

  ↯ Ollama↯ Gemma 4ollamacursor

• Show HN: Cuts Long Horizon Inference Costs by 50% via external KV Cache Offload (github.com via hn) +20 7w

  Hey HN, we’re the developers of OpenLake, an open source storage engine for offloading LLM KV caches from GPU memory into a shared tier of RAM and NVMe. We built OpenLake because KV caches are outgrowing GPU memory.

  ↯ Gemma↯ Gemma 4gemma

• F*ck it, Vox is now free and the source code is public (www.reddit.com via reddit) 9w

  I'm making Vox (a voice-to-text app for macOS and Windows that I built with Claude) free for everyone, and the repo is now open. You hold a hotkey, talk, release, and the transcribed + polished text lands in your clipboard (or gets pasted…

  ↯ Gemma↯ Gemma 4gemmacursor

• Running Gemma 4 26B at 5 tokens/SEC on a 13-year-old Xeon with no GPU (www.neomindlabs.com via hn) +226 9w

  June 8, 2026 · By Ryan Findley Running Gemma 4 26B at 5 tokens/sec on a 13-year-old Xeon with no GPU There’s a server in my basement that has no business running a modern language model. It’s a repurposed HP StoreVirtual storage box, rough…

  ↯ Gemma↯ Gemma 4gemma

• Show HN: BYO AI free notetaking with optional screen reading for OpenClaw/hermes (stagewhisper.io via hn) +31 9w

  I've built fully on-device (macOS) meeting transcription/summaries/action-items app. It uses parakeet and gemma 4 or OpenClaw or Hermes Agent to drive main functionalities.

  ↯ Gemma↯ Gemma 4gemmaopenclaw

• Show HN: Godot running Gemma 4 inference in GDScript and Vulkan compute shaders (github.com via hn) +1 9w

  Experimental Godot 4.7 project running Gemma 4 inference entirely in GDScript and Vulkan compute shaders.

  ↯ Gemma↯ Gemma 4gemma

• Show HN: Cactus v2 – On-device AI with cloud fallback (news.ycombinator.com) +1 10w

  Hi HN, Roman and Henry here from Cactus (https://github.com/cactus-compute/cactus). We just shipped the biggest upgrade to our on-device inference platform: - Built-in model confidence-based routing to hand off inference runs to the cloud…

  ↯ Gemma↯ Gemma 4gemma

• Gemma 4 Technical Report (arxiv.org) 10w

  We introduce Gemma 4, a new generation of open-weight, natively multimodal language models in the Gemma model family. Designed to advance compute efficiency and reasoning, the Gemma 4 model suite features dense and Mixture-of-Experts archi…

  ↯ Gemma↯ Gemma 4gemma

• does fable use gemma 4-12b to run tests (www.reddit.comhttps) 11w

  I was auditing my codes using the now back fable 5 and it kept failing to run runtime tests and this the error i got. so is anthropic now using gemma-4-12b-agentic-fable5-composer2.5-v2-3.5x-tau2 to run tests?

  ↯ Gemma↯ Gemma 4gemmaanthropicagentic

• Faster Gemma 4 on MLX with multi-token prediction (ollama.com via hn) +2 11w

  Gemma 4 is now significantly faster in Ollama 0.31 on Apple Silicon via multi-token prediction (MTP), powered by MLX. Performance is now up to 90% faster when used with coding agents, as measured using the Aider polyglot benchmark.

  ↯ Ollama↯ Gemma↯ Gemma 4aiderollamagemma

• Hugging Face and Cerebras bring Gemma 4 to real-time voice AI (huggingface.co) 11w

  HF Realtime Voice Voice chat over WebSocket against a HF speech-to-speech The result is a speech-to-speech experience that feels dramatically more natural. Instead of waiting for an AI to respond, conversations flow with the responsiveness…

  ↯ Gemma↯ Gemma 4gemma

• Gemma 4 on Cerebras - The Fastest Inference Is Now Multimodal (www.cerebras.ai via hn) +31 11w

  Gemma 4 on Cerebras—The Fastest Inference is Now Multimodal Gemma 4 31B is now running at over 1,800 tokens per second on Cerebras Inference. This multimodal model unlocks an entirely new class of applications, from computer use to image-d…

  ↯ Gemma↯ Gemma 4gemma

• I brought Claude-style artifacts to local models (www.reddit.comhttps) 12w

  One thing I miss when using local models is the artifact experience from Claude. With Claude, if you ask for a dashboard, chart, diagram, or landing page, you actually get the thing rendered in the chat.

  ↯ Gemma↯ Gemma 4↯ Gemma 4↯ Gemma 4↯ Gemma 4↯ Gemma 4↯ Gemma 4gemma

• In Claude Code I fine-tuned Gemma 4 (E2B, Q4_K_M) and got it running 100% on-device in an iOS app — a little sea-creature companion you actually talk to. Offline, no servers, beta's open. (www.reddit.comhttps) 12w

  Requirements: iPhone with A17 Pro or newer (8 GB RAM floor for the model), iOS 26+. TestFlight beta is open to anyone with a compatible device.

  ↯ Gemma↯ Gemma 4gemmaclaude-code