Alternatives
Products that do what KV-psi, using Linux PSI to to trim an LLM KV cache does
I thought it'd be interesting to use Linux PSI (Pressure Stall Information) for an LLM runtime to trim the KV cache. This is mainly useful imo for edge devices like the Jetson Orin super nano kit which have unified memory. I haven't benched much, but plan to do so more over time and see if I can make a real use of it as I run local LLMs. Let me know if it makes sense :P (I of course vibed this idea)
- 1
I made this after seeing someone posit the idea online yesterday over lunch then spent some time refining it. So far it's pretty impressive IMO! Right now I am running Qwen3-30B-A3B on my 24gb unified memory m4 MacBook Pro at 50 tok/sec and this should definitely not be working for such a large model on my middling hardware. Things are detailed in the README to get up and running and DESIGN.md has details on all the choices and such made along the way.
23d ago · github.com
- 2ML
Time to first token is 39% faster Agent wall times decrease by 46% No swaps Tracks your resource usage in real-time and adjusts how the model runs so that it works perfectly on your device. Implements KV cache sizing, prefix caching, live RAM pressure management, context trimming, KV quantization, and more. Built a ton of features
Jun 2026 · autotunellm.com
- 3PU
After seeing a cool demo of a hack on Twitter, I built a cross platform version of it that works well and uses streaming. From anywhere on Mac and Linux, trigger Ollama and optionally feed it your clipboard. I built it yesterday and it's already very useful to me. I'm pretty excited about it and wanted to share!
2024 · github.com
- 4FG
Built this as a guide for users of my hosting service. Everything here was run on a real instance. The Mesa/LLVM dependency is genuinely painful on a 1GB disk, and I couldn't find a way around it. WindowMaker won the window manager battle purely on dependency chain cleanliness; it pulls in no GTK, no Python, no audio stack. Happy to answer questions about any of the choices made. Screenshots at the end.
Mar 2026 · tierhive.com
- 5AC
It's quite limited but does the job, I invite anybody not aware of what cram tests are to give them a try, being on Unix with the original cram or Windows with this one ;)
Nov 2025 · github.com
- 6LH
I work on inference scheduling — KV cache-aware routing, load balancing across GPU workers, that kind of thing. I wanted something like k9s but for my inference stack. Nothing existed, so I built it. llmtop is a real-time terminal dashboard for LLM inference workers. It scrapes the Prometheus /metrics endpoints that vLLM, SGLang, and LMCache already expose and shows everything in one view: KV cache usage, queue depth, TTFT/ITL latencies (P50/P99 from histogram buckets), token throughput, prefix cache hit rates. Color-coded — red means go fix it. ``` brew install…
Mar 2026 · github.com
- 7IR
Democratisation of local AI is key. I've been working on pushing the limits of commercial hardware, squeezing any extra bit possible. My Scientific Agentic AI hareness helped me to reallocate every single bit of it. I rewrote the Kernel, I went down the CUDA rabbit hole until I have been able to explain any bit and any ms of computational power involved in the process pushing the Qwen 30B-A3B from 8 tok7s to 19 tok/s with llama.cpp up to 22.2 tok/s with my project and 109 tok/s on not novel content and speeding up the prefill by 5-9X
Jul 2026 · github.com
- 8IB
You know that old TI calculator you used in high school, then put in a box and forgot about? Have you ever wished you had an operating system for your calculator with preemptive multitasking, dynamic memory management, a tree filesystem conforming to the FHS, and all the comforts of Unix? Well, good news: that's totally a thing that exists. I've been working on my kernel for about three and a half years now and I'm looking for new contributors to help out. It's written entirely in z80 assembly, and it's both challenging and fun to work on. There's an IRC channel for contributors or people…
2014
- 9IB
Built a simple web app that tells you which open-source LLMs will work on your hardware. It auto-detects your specs, shows compatible models from Hugging Face, gives realistic performance estimates (tokens/sec), and recommends quantization settings. You can also manually input specs to see "what if I upgraded my RAM?" Made this after wasting time downloading giant models only to find they crawled on my hardware. Hope it saves you some frustration!
2025 · caniusellm.com
- 10UA
Hey HN! After using a combination of Unsloth and Axolotl a lot, and finding it generally painful to figure out the right performance tuning for things like batch sizing and multi-GPU sharding, I wrote a small Python lib that sets up known-good LoRA training configurations for Llama 3.1 8B and 70B Instruct, and includes helpers for distilling from larger models or training on serverless finetuning platforms, and includes a walkthrough for distilling DeepSeek-R1 into a Llama 3.1 8B LoRA... But you can use it for pretty much any finetuning task, not just distilling large models!
2025 · github.com
- 11AL
Hello all, a couple of moons ago I ported karpathy's llama2.c code to run inference on the TinyStories 260K & 15M checkpoints on the on the PS Vita with the ability to download/delete the models on device. Runs showed that the 260K model ran at ~120 tok/s and at 15M ran at 1.8 tok/s, which probably could be a bit higher if it weren't a single threaded application. Had fun working on it as a weekend project, check it out for yourselves: https://github.com/callbacked/psvita-llm
2025 · github.com
- 12NT
With the latest launch from Google I've added support for Gemma 3 270M, the speed for local LLM to TTS token time is incredible! This is an heavy obvious work in progress - any contributions or tips would be welcome. The idea is to have a fast moving edge model playground, and maybe have some utility (like the e reader) on the side.
2025 · github.com
- 13LV
Hello HN! We've been working on a new hypervisor https://kwarantine.xyz that can run strongly isolated containers. This is still a WIP, but we wanted to give the community an idea about our approach, its benefits, and various use cases it unlocks. Today, VMs are used to host containers, and make up for the lack of strong security as well as kernel isolation in containers. This work adds this missing security piece in containers. We plan on launching a free private beta soon. Meanwhile, we'd deeply appreciate any feedback, and happy to answer any questions here or on our slack…
2021
- 14PS
I was debating when to post this, but I've been using it locally in a variety of situations for several months and I think it's pretty stable (and usable!). There are more features I'd like to add, but I wanted to see if there was much interest before spending much more time on it right now.
2025 · github.com
- 15IT
Hi! This is a version 2.0 2x Context window 7-8x Faster Less hallucinations :)
2025 · shortcutbuilder.com
- 16IN
I wrote an article about how to get Nix on your Steam Deck! Big takeaways for me: * Steam Decks are very fun to hack on * Linux Consoles are good * There are more knobs in systemd units than I remember * Flatcar Linux is a thing that exists * Remoting into a console is fun * A/B boots are less interesting than one might expect when you're used to NixOS
2022 · determinate.systems
- 17IB
I had 14,000 photos sitting on a drive and wanted an excuse to play with local vision models and Elixir/Phoenix. I originally tried to get LLaVA to tell me if a photo was 'good' or matched my style, but quickly learned that LLMs have terrible taste. I ended up demoting the LLM to just extract metadata, and built a custom CLIP/Ridge Regression pipeline to actually learn my preferences based on how I rate things. The stack is Phoenix/Oban on the orchestrator side, and Python/FastAPI/Instructor for the AI workers. Happy to answer any questions about the architecture,…
Apr 2026 · qwelian.com
- 18AN
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
- 19WB
Over the past few months, as we scaled our internal AI Agents, we hit a dead end: Running LLM-generated arbitrary code in Docker is basically running naked on security due to container escape risks. But using full traditional VMs takes minutes to boot and eats too much memory to support high-density concurrency. We loved the developer experience of SaaS sandboxes on the market, but they are closed-source, expensive, and have too high a barrier to entry for self-hosting. So, our team decided to build our own. After months of grinding, using RustVMM and KVM, we built a blazing-fast,…
Apr 2026 · github.com
- 20IM
Hey HN! I made a completely open sourced alternative to Weights and Biases with (insert cringe) blazingly fast performance (yes we use rust and clickhouse) Weights and Biases is super unperformant, their logger blocks user code... logging should not be blocking, yet they got away with it. We do the right thing by being non blocking. Would love any thoughts / feedbacks / roasts etc
2025 · github.com
- 21RS
Hi HN! I've been hacking on this side project for the last month or two with the goal of making it dead simple to use cloud GPUs. I ran into this problem personally during the phd, and built my own tooling around it. I always thought it'd be fun to try to turn that tooling into a more general product... and bitbop.io is the result! All you have to do is run `ssh bitbop.io`, and you get your own personal dev GPU workstation in the cloud. Looking forward to hearing your thoughts!
2024 · twitter.com
- 225L
We've built InferX, a specialized runtime environment that fundamentally changes how LLMs are served. The core problem we solve is the latency bottleneck in AI inference, especially with large models. Current systems waste resources or suffer from painfully slow cold starts. InferX's AI-native architecture, with its "snapshot" technology, enables: * *Sub-2s cold starts:* Spin up models instantly. * *High density:* Serve more LLMs on the same GPUs. * *Optimal efficiency:* Maximize GPU utilization. This isn't just another API; it's a new execution layer designed from the ground up for the…
2025 · github.com
- 23LF
I submitted an earlier version of this a few months ago (as llama2.f90). At that time it had a lot of steps to run and was just a toy, now it's easy to run and is a competitive option for llm inference. See the motivation section for discussion and the `Performance` issue for an ongoing discussion about performance.
2023 · github.com
- 24WB
Here is a production-first Keras-inspired LM framework, built with the advice of François Chollet (ex-Google, creator of Keras and ARC-AGI), our technical advisor. This system have already been deployed in production with our clients (which is why we have already every LLMOps practice implemented). It is also compatible with Jupyter and Marimo to integrate seamlessly in you Data Scientists workflows. You can try the code examples online on HF space and you can find more information in the documentation and FAQ. If you have any feedback for us don't hesitate to join our discord! More releases…
2025 · github.com
Ranked by how close each launch is in meaning, then by votes. Refine with a description →