Hacker News: Yukonv

New comment by Yukonv in "GLM-5.1: Towards Long-Horizon Tasks"

Yukonv — Tue, 07 Apr 2026 17:06:49 +0000

Unsloth quantizations are available on release as well. [0] The IQ4_XS is a massive 361 GB with the 754B parameters. This is definitely a model your average local LLM enthusiast is not going to be able to run even with high end hardware.

[0] https://huggingface.co/unsloth/GLM-5.1-GGUF

New comment by Yukonv in "Running Gemma 4 locally with LM Studio's new headless CLI and Claude Code"

Yukonv — Sun, 05 Apr 2026 21:38:57 +0000

With that you are taking a significant performance penalty and become severely I/O bottlenecked. I've been able to stream Qwen3.5-397B-A17B from my M5 Max (12 GB/s SSD Read) using the Flash MoE technique at the brisk pace of 10 tokens per second. As tokens are generated different experts need to be consulted resulting in a lot of I/O churn. So while feasible it's only great for batch jobs not interactive usage.

New comment by Yukonv in "April 2026 TLDR Setup for Ollama and Gemma 4 26B on a Mac mini"

Yukonv — Fri, 03 Apr 2026 15:02:35 +0000

The latest release v0.3.2 has partial support, generation is supported but not all special tokens are handled. I've done some personal testing to add tool calling and <|channel> thinking support. https://github.com/Yukon/omlx

New comment by Yukonv in "Google releases Gemma 4 open models"

Yukonv — Thu, 02 Apr 2026 23:46:49 +0000

The model does have the format specified but there is no _one_ standard. For this model it’s defined in the [ tokenizer_config.json [0]. As for llama.cpp they seem to be using a more type safe approach to reading the arguments.

[0] https://huggingface.co/google/gemma-4-31B-it/blob/main/token...

New comment by Yukonv in "Ollama is now powered by MLX on Apple Silicon in preview"

Yukonv — Tue, 31 Mar 2026 07:03:48 +0000

Good to see Ollama is catching up with the times for inference on Mac. MLX powered inference makes a big difference, especially on M5 as their graphs point out. What really has been a game changer for my workflow is using https://omlx.ai/ that has SSD KV cold caching. No longer have to worry about a session falling out of memory and needing to prefill again. Combine that with the M5 Max prefill speed means more time is spend on generation than waiting for 50k+ content window to process.

New comment by Yukonv in "iPhone 17 Pro Demonstrated Running a 400B LLM"

Yukonv — Mon, 23 Mar 2026 19:12:39 +0000

That’s exactly what I thought about. Getting my hands on an M5 Max this week and going to see hows Dan’s experiment performs with faster I/O. Also going to experiment with running active parameters at Q6 or Q8 since output is I/O bottlenecked there should room for higher accuracy compute.

Is AI capable of Intelligent Disobedience? [video]

Yukonv — Sat, 21 Mar 2026 19:18:59 +0000

Article URL: https://www.youtube.com/watch?v=Qu-00j9XuF0

Comments URL: https://news.ycombinator.com/item?id=47470321

Points: 2

# Comments: 0

New comment by Yukonv in "Rust error handling"

Yukonv — Sun, 15 Sep 2024 01:39:59 +0000

Another good option I’ve personally used if you want a smaller API surface with just Result and Maybe concepts is True Myth. https://true-myth.github.io/true-myth-csharp/

New comment by Yukonv in "High-speed 10Gbps full-mesh network based on USB4 for just $47.98"

Yukonv — Mon, 15 Jan 2024 19:38:44 +0000

Related, Intel was showing off Thunderbolt Share at CES[1]. Allows Thunderbolt 4/5 device-to-device transfer of files. Theoretical speeds in the 20Gbps and 40Gbps for Thunderbolt four and five respective.

One idea for why they were only able to reach 11Gbps is having only one Thunderbolt/USB4 controller[2], meaning the two USB4 ports split the 40Gbps PCIe lane. Throw in a full-duplex connection and you get 10Gbps in one direction.

[1] https://youtu.be/GqCwLjhb4YY?t=81 [2] Just a theory but seems like a sane assumption.

New comment by Yukonv in "Super Mario 64 on the Web"

Yukonv — Thu, 11 Jan 2024 03:15:34 +0000

Great find! Seems to be a common issue with games, found the same issue with trying to auto play PICO-8 cartrages on web.

New comment by Yukonv in "Super Mario 64 on the Web"

Yukonv — Thu, 11 Jan 2024 03:00:38 +0000

Edit: Keyboard input does not work :/ it was a good effort.

Found a workaround throw it in an iFrame and have the frame load with a user interaction. Here is a jsFiddle link, just click "Run" after the page loads.

https://jsfiddle.net/sg1r3h60/

New comment by Yukonv in "Super Mario 64 on the Web"

Yukonv — Thu, 11 Jan 2024 02:30:25 +0000

Firefox prevents audio from playing with no initial user interaction like a play button. Can see the warning if you pop open the dev console. Don't know of workaround besides sites not creating an AudioContext on page load.

Starlink Maritime

Yukonv — Thu, 07 Jul 2022 19:26:04 +0000

Article URL: https://www.starlink.com/maritime

Comments URL: https://news.ycombinator.com/item?id=32018218

Points: 606

# Comments: 508

New comment by Yukonv in "Ask HN: Is your company sticking to on-premise servers? Why?"

Yukonv — Wed, 13 May 2020 04:14:48 +0000

Little late but though I would say hi. I too got started programming thanks to Minecraft. My first real job was working at Overcast Network (oc.tc). I remember having to scale out our infrastructure to seven dedicated servers after a popular YouTuber featured us. At the time that felt crazy for a Minecraft server and here you are now with hundreds of servers. Huge congrats on scaling to where you are today.

Have lots of fond memories of those early years, especially Minecon 2013.