KVSplit – Run 2-3x longer contexts on Apple Silicon
I discovered that in LLM inference, keys and values in the KV cache have very different quantization sensitivities. Keys need higher precision than values to maintain quality. I patched llama.cpp to enable different bit-widths for keys vs. values on Apple Silicon. The results are surprising: - K8V4 (8-bit keys, 4-bit values): 59% memory reduction with only 0.86% perplexity loss - K4V8 (4-bit keys, 8-bit values): 59% memory reduction but 6.06% perplexity loss - The configurations use the same number of bits, but K8V4 is 7× better for quality This means you can run LLMs with 2-3× longer…
KVSplit – Run 2-3x longer contexts on Apple Silicon launched on May 16, 2025 with 272 votes, #106 of 7,905 launches that month and more than 99% of that year's launches. The site was still online in September 2026, 1 year on.
In plain words
KVSplit is a modification to llama.cpp that enables independent quantization of keys and values in the KV cache for Apple Silicon processors. By using higher precision for keys and lower precision for values, it reduces memory usage by up to 59% while maintaining model quality, allowing users to run language models with 2-3 times longer context windows on Mac computers. The tool adds command-line flags to control quantization bit-widths separately for each component.
written from the facts on this page · September 2026
From the sources
Run larger LLMs with longer contexts on Apple Silicon by using differentiated precision for KV cache quantization. KVSplit enables 8-bit keys & 4-bit values, reducing memory by 59% with <1% quality loss. Includes benchmarking, visualization, and one-command setup. Optimized for M1/M2/M3 Macs with Metal support. - dipampaul17/KVSplit
In the maker’s words, at launch
I discovered that in LLM inference, keys and values in the KV cache have very different quantization sensitivities. Keys need higher precision than values to maintain quality. I patched llama.cpp to enable different bit-widths for keys vs. values on Apple Silicon. The results are surprising: - K8V4 (8-bit keys, 4-bit…
Does a similar job
all alternatives →
Speeding up LLM inference 2x times (possibly)2024 · asciinema.org · ▲419Here's a project I've been working on for the last few months. It's a new (I think) algorithm, that allows to adjust smoothly - and in real time - how many calculations you'd like to do during inference of an LLM model. It seems that it's possible to do just 20-25% of weight multiplications instead of all of them, and still get good inference results. I implemented it to run on M1/M2/M3 GPU. The mmul approximation itself can be pushed to run 2x fast before the quality of output collapses. The inference speed is just a bit faster than Llama.cpp's, because the rest of implementation…
80% faster, 50% less memory, 0% loss of accuracy Llama finetuning2023 · github.com · ▲385Hi HN! I'm just sharing a project I've been working on during the LLM Efficiency Challenge - you can now finetune Llama with QLoRA 5x faster than Huggingface's original implementation on your own local GPU. Some highlights: 1. Manual autograd engine - hand derived backprop steps. 2. QLoRA / LoRA 80% faster, 50% less memory. 3. All kernels written in OpenAI's Triton language. 4. 0% loss in accuracy - no approximation methods - all exact. 5. No change of hardware necessary. Supports NVIDIA GPUs since 2018+. CUDA 7.5+. 6. Flash Attention support via Xformers. 7. Supports 4bit and 16bit…
Tiny-vLLM – high performance LLM inference engine in C++ and CUDAMay 2026 · github.com · ▲205Build your own high performance LLM inference engine in C++ and CUDA - a smaller version of vLLM - jmaczan/tiny-vllm
We made our own inference engine for Apple Silicon2025 · github.com · ▲186We wrote our inference engine on Rust, it is faster than llama cpp in all of the use cases. Your feedback is very welcomed. Written from scratch with idea that you can add support of any kernel and platform.
Klarity – Open-source tool to analyze uncertainty/entropy in LLM output2025 · github.com · ▲132We've open-sourced Klarity - a tool for analyzing uncertainty and decision-making in LLM token generation. It provides structured insights into how models choose tokens and where they show uncertainty. What Klarity does: - Real-time analysis of model uncertainty during generation - Dual analysis combining log probabilities and semantic understanding - Structured JSON output with actionable insights - Fully self-hostable with customizable analysis models The tool works by analyzing each step of text generation and returns a structured JSON: - uncertainty_points: array of {step, entropy,…
- TFThe fastest way to run Mixtral 8x7B on Apple Silicon Macs2024 · ▲18
I’d originally launched my app: Private LLM[1][2] on HN around 10 months ago, with a single RedPajama Chat 3B model. The app has come a long way since then. About a month ago, I added support for 4-bit OmniQuant quantized Mixtral 8x7B Instruct model, and it seems to outperform Q4 models at inference speed and Q8 models at text generation quality, while consuming only about 24GB of RAM[3] at 8k context length. The trick is: a) to use a better quantization algorithm and b) to use unquantized embeddings and the MoE gates (the overhead is quite small). Other notable features include many more…
More ai this month
the category →
Paintings by AI models: programs run through a simulation of oil paint.
AI · 20d ago · stillwet.art
PixVerse R2▲381A real-time world model you can explore and change
AI · 28d ago · world.pixverse.video
Source: Show HN post ↗ · launch text quoted in part
Launched alongside, May 2025
the whole month →
Tapflow 2.0▲1,125We built Tapflow for devs, designers, PMs, and tech folks who’ve been sitting on internal docs, templates, and systems — but never productized them. Upload a PDF, notes, or workshop deck, and Tapflow drafts a structured product (with chapters, teaser and pricing plans). It’s good for AI kits, onboarding guides, design systems, and frameworks. You can sell through Stripe, LemonSqueezy, or PayPal. Real-time collaboration is in early beta. No site builder, no LMS — clean design and just focused tools to ship. Fully bootstrapped and just launched on PH today →…
Work · 2025 · tapflow.ai
- C9
Life & fun · 2025 · felixrieseberg.github.io
Appwrite Sites▲990Appwrite 2.0 brings a new engine, a rebuilt Console, five database types, an S3 API, an OAuth 2.1 server, and organization-level Domains and Firewall.
Dev tools · 2025 · appwrite.io
