Alternatives
Products that do what Running Gemma-4 26B at 124 tokens/SEC on a CPU, no GPU does
I wanted to know how fast a 26B mixture-of-experts model could run on a desktop CPU with no GPU. Got ~40 tok/s single-stream (lossless) and ~124 batched. The surprising part was the byte budget: for this model you compress the output head (32% of per-token bytes), not the experts (16%). The writeup has the bandwidth roofline and the dead-ends; the repo has the reproducible recipe. Happy to answer questions. Repo: https://github.com/arun-prasath2005/gemma4-cpu-moe
- 1OS
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. I have always adored on-device AI. It feels like magic that you can run a powerful NN on your Mac or iPhone. So I wanted to push the limits a bit and run a model whose weights don’t fit in memory. The model’s 4-bit quantized weights occupy roughly 14 GB, which makes running it with conventional inference tools almost impossible on an 8 GB or even 16 GB Mac once the OS, applications, and KV cache are…
Jul 2026 · github.com
- 2
Run multimodal AI locally with an encoder-free architecture
Jun 2026 · blog.google
- 3
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
27d ago · cactuscompute.com
- 4CH
Hey HN, Henry & Roman here from Cactus. A small, on-device model is fast and private, but sometimes wrong, but frontier models are getting expensive pretty fast. So, we post-trained Gemma 4 E2B post-trained to know when it's wrong. Every response comes with a confidence score between 0 and 1. Developers can accept the on-device when it's high, hand off to a bigger cloud model when it's low. By routing only 15-35% of queries to Gemini 3.1 Flash-Lite, Gemma-4-E2B matches Gemini 3.1 Flash-Lite on most benchmarks. - ChartQA: 15-20% - LibriSpeech: 25-30% - MMBench, GigaSpeech, MMAU: 30-35% -…
Jul 2026 · github.com
- 5
- 6AT
A 3.16M-parameter INT4 transformer running entirely in the on-chip memory of a Xilinx Kria KV260. Zero DRAM in the token loop, 59,965 tok/s on the fabric, bit-exact. Chat with it live.
27d ago · mikeayles.com
- 7
General Compute▲315AI models that run on an inference cloud optimized for speed
May 2026 · generalcompute.com
- 8FT
Aug 2026 · github.com
- 9
- 10DD
We recently used DeepSeek V4 Flash as a teacher for finance tasks with GPT-OSS-120B. Distillation works well on this problem. At a constrained 8k token budget, our self-distilled 120B scores 83.61% on FinanceReasoning, above Kimi K3 (81.93%) and Inkling (65.13%). We released the 20B open weights. With V4 as the teacher though, we realized it would be timely to measure if the censorship characteristic of it transferred to the distilled version of the base model. tl;dr it didn't, the teacher answered politically sensitive questions 7 SDs differently than expected, but the distilled model's…
Jul 2026 · ctgt.ai
- 11
- 12RR
runNburn is an Apache-2.0 Rust inference engine for quantized GGUF models that are too big for your fast memory. The core idea: weights stay file-backed (mmap), host residency stays under an explicit byte budget (--ram-budget), and GPU caches are sized from detected free/total VRAM — never from device-name presets. There is no conversion step, no sidecar cache files, no silent requantization. The GGUF on disk is the single source of truth. The result that made me want to post this: Tencent's Hy3 (295B total / 21B active sparse MoE, a single 97.8 GiB Q2_K GGUF) runs on my desktop…
Jul 2026 · github.com
- 13AN
The core question: how did HP's scientific calculators actually work at the gate level? That rabbit hole led to building one from scratch. The architectural decision everything else follows from: a decimal calculator should store numbers as BCD — one decimal digit per 4-bit nibble. A standard byte-oriented CPU (Z80, 6502) fights that layout constantly. So I designed a small custom CPU in Verilog where 4 bits is the natural data width and memory is nibble addressable. What the project covers: - Custom CPU: Harvard architecture, 12-bit ISA, 8-state execution FSM, hardware stack guard with a…
May 2026 · github.com
- 14IM
I made my first macOS utility app that ships with a bundled Gemma 4 model, specifically the Gemma E4B one. It made my app DMG have 5.3 GB in size, but I think it is a small size for the power that this free local model can provide. It runs fine on CPU, but can also run on Apple Silicon GPU, although I did not notice any performance improvements with GPU (tested on a M5 chip). I think these local lightweight and multimodal models will open multiple possibilities for new software tools where privacy is essential.
May 2026 · snapname.app
- 15CL
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. A single 256K token conversation on Gemma 4 31B produces approximately 43GB of KV state, more than half the memory of an 80GB H100. The problem becomes even harder across a cluster: a prefix cached on one GPU host is unavailable when the next request lands on a different GPU, forcing the new GPU to repeat work the fleet has already completed. Once the KV cache is offloaded,…
Jul 2026 · github.com
- 16
- 17AN
Kimi K3 has 2.78 trillion parameters and ships as 1.42 TB of weights. It clearly does not fit in the memory of a laptop. But K3 is a Mixture-of-Experts model. For each token, only a small fraction of its 896 experts per layer is activated. That changes the problem: the entire model does not need to be resident in RAM, as long as the weights required by each token can be reached quickly enough. We built WASTE — the Weight-Aware Streaming Tensor Engine — to explore that idea. WASTE keeps the dense, repeatedly used part of the model resident in memory, stores the routed experts in an…
Jul 2026
- 18
High performance storage engine for efficient LLM inference and GPU Training.
16h ago · theopenlake.com
- 19IE
Quick note on how it works and how I've done my batch embedding engine IgniteMS. The whole thing runs as one process using Rust, reading input, tokenizing, packing batches, keeping the queue full. TensorRT handles inference. Python is only as a wrapper. I built it this way because when you use more than couple of GPUs, the GPUs stop being the problem. CPU cannot feed them fast enough. One A100 can go through batches faster than Python can tokenize and feed, so the GPU just sits there idle waiting for work. Most of my time went into optimizing this. At 8 GPUs that was basically the entire…
Jun 2026 · github.com
- 20
- 21MA
I've been working on training this small vision language model for the last month - excited to release the first prototype today! It is based on SigLIP (image encoder), Phi-1.5 (text model) and trained using the LLaVa-1.5 training dataset. It runs reasonably fast on CPU with ~8GB of RAM in full 32-bit precision. There's plenty of room to speed it up and reduce memory consumption by quantizing the model. I posted a video of it running on my M2 Macbook Air (on CPU not MPS, so performance should be comparable on other hardware) on Twitter to demonstrate inference speed:…
2023 · github.com
- 22I4
It's our new text-to-image model: a 9.3B single-stream diffusion transformer trained entirely from scratch. We focused heavily on controllability through structured JSON prompts, with strong text rendering, spatial awareness through bounding box guidance, and color palette control. It has the best text rendering of any open-weight model we've tested so far, and the NF4 quantized checkpoint runs on a single 24GB GPU. For more technical details and examples see our blog post: https://ideogram.ai/blog/ideogram-4.0/ We will be happy to answer any questions :)
Jun 2026 · github.com
- 23S1
I wanted to build an inference provider for proprietary AI models, but I did not have a huge GPU farm. I started experimenting with Serverless AI inference, but found out that coldstarts were huge. I went deep into the research and put together an engine that loads large models from SSD to VRAM up to ten times faster than alternatives. It works with vLLM, and transformers, and more coming soon. With this project you can hot-swap entire large models (32B) on demand. Its great for: Serverless AI Inference Robotics On Prem deployments Local Agents And Its open source. Let me know if anyone…
Nov 2025 · github.com
- 24IR
Jun 2026 · github.com
Ranked by how close each launch is in meaning, then by votes. Refine with a description →