Alternatives
Products that do what UL-SMF – Open-source linear-complexity ~300x KV-cache compression does
The Unified Latent-State Memory Fabric (UL-SMF) is a hardware-software co-designed memory compression fabric that solves the memory bottleneck in long-context Transformer inference. By combining FSQ with dynamic 16-dimensional latent mapping, UL-SMF compresses Key-Value (KV) cache tensors by up to 3
- 1
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
- 2AT
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
- 3
High performance storage engine for efficient LLM inference and GPU Training.
16h ago · theopenlake.com
- 4OA
I built an experiment that uses an overfitted transformer and arithmetic coding to compress individual files. Instead of training the model to generalize, I train a 900KB transformer to memorize a single file and predict the next byte. Those predictions are fed into an arithmetic coder to produce the compressed output. On a 100MB NYC taxi CSV, it compresses to about 7MB (~0.5 bits/byte). On a 100MB slice of enwik9, it compresses to about 21MB (~1.68 bits/byte). It's pretty slow right now (roughly 20–30 minutes of training and 45 minutes each for compression and decompression on my…
Jun 2026
- 5
- 6
- 7
An OKF-backed Model Context Protocol (MCP) server delivering persistent long-term memory and SQLite FTS5 search for AI agents. - fellowgeek/mcp-memory
24d ago · github.com
- 8CL
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
- 9RG
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
Jun 2026 · apeg.dev
- 10MW
2017 · github.com
- 11
- 12ML
Aug 2026 · github.com
- 13RR
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
- 14UF
May 2026 · github.com
- 15FN
"I wanted to see if I could optimize the dequantization bottleneck during 4-bit LLM inference. By writing a custom kernel in Triton to optimize memory access patterns, I managed to get up to a 1.41x speedup over the standard bitsandbytes implementation. Check out the source code and benchmarks, feedback is highly appreciated!"
Jul 2026 · github.com
- 16MA
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
- 17
Local, gradient-free neuro-symbolic memory engine combining Hyperdimensional Computing (HDC/VSA), Hebbian plasticity, and graph triples for offline AI. - roandejager/Hillock
7d ago · github.com
- 18LS
Hi HN, I built llm.sql, an LLM inference framework that reimagines the LLM execution pipeline as a series of structured SQL queries atop SQLite. The motivation: Edge LLMs are getting better, but hardware remains a bottleneck, especially RAM (size and bandwidth). When available memory is less than the model size and KV cache, the OS incurs page faults and swaps pages using LRU-like strategies, resulting in throughput degradation that's hard to notice and even harder to debug. In fact, the memory access pattern during LLM inference is deterministic - we know exactly which weights are needed…
Apr 2026
- 19MR
I needed to stuff twice the telemetry through the same 115 kbaud line on a Cortex-M0+ that only had 8 kB flash left. Micro-RLE is the smallest drop-in I could come up with: 264 B of Thumb code, 36 B of state, no malloc, worst-case 14 cycles/byte and still lossless for every 8-bit pattern. On the usual sensor streams (ADC, IMU, GPS) it’s 33-70 % smaller than raw output and boots in < 600 µs, so you can fire-and-forget from main() before the PLL even locks. Repo is a single .c file and a 3-function API—replace the weak emit() hook with your UART / DMA / ring-buffer and you’re…
Nov 2025 · github.com
- 20SL
I'd like to share a little toy project of mine, a really simple image codec that can do lossy to full lossless image compression with complete scalability at a byte level granularity - you can compress an image just once, even fully losslessly if needed, and then get any lossy version possible by simply stopping decompression at any offset in the compressed data. This "encode onde, serve many" approach is especially interesting for providing downscaled low quality image previews (LQIP) in as tight a storage budget as possible, and then allowing seamless, transparent refinement as deemed…
2024 · github.com
- 21AN
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
- 22TE
2021 · github.com
- 23UM
Hi HN, We’ve been building memU(https://github.com/NevaMind-AI/memU), an open-source, general-purpose memory framework for AI agents. It supports dual-mode retrieval: classic RAG and LLM-based direct file reading. Most multimodal memory systems either embed everything into vectors or treat non-text data as attachments. These work, but at scale it becomes hard to explain why certain context was retrieved and what evidence it relies on. memU takes a different approach: since models reason in language, multimodal memory should converge into structured, queryable text, while…
Jan 2026 · github.com
- 24IO
Hey folks, I’m the creator of WFGY — a semantic reasoning framework for LLMs. After open-sourcing it, I did a full technical and value audit — and realized this engine might be worth $8M–$17M based on AI module licensing norms. If embedded as part of a platform core, the valuation could exceed $30M. Too late to pull it back. So here it is — fully free, open-sourced under MIT. --- ### What does it solve? Current LLMs (even GPT-4+) lack *self-consistent reasoning*. They struggle with: - Fragmented logic across turns - No internal loopback or self-calibration - No modular thought units - Weak…
2025 · github.com
Ranked by how close each launch is in meaning, then by votes. Refine with a description →