Chris Hay

Technologist, Futurist and Disruptor - Creator - YouTuber All tweets are my own and don't represent the views of my employer

[Gregor @ceonos3k]

I ran Gemma 3 4B without PyTorch, without CUDA — just flat binary files + 600 lines of Rust.

Paris #1. Jupiter #1. Ulm #1. Exact HF match.

CPU or Vulkan GPU. Any vendor. No framework.
Built on @chrishayuk's LarQL decomposition.

https://t.co/Gc9TPxehns
#DeepLearning #GPU #MLOps #AIEngineering #RustLang #HuggingFace #Gemma #WebGPU
#CrossPlatform #EdgeAI #OpenAI #LLMInference #FOSS #IndieResearch

[Gregor @ceonos3k]

@chrishayuk watched your LARQL video this morning, spent the day porting it to Windows/Linux + CUDA and wrapping it in a Gradio UI — live on HF Spaces 12 hours after your release.

👉 https://t.co/yowLwXztiP
👉 https://t.co/XWDBpJuDPJ
Already running in 3 of my projects. Great work.

[Chris Hay @chrishayuk]

@ceonos3k Ui looks great, would love the windows / Linux cuda stuff etc merged into main project. Awesome work

[Chris Hay @chrishayuk]

@ceonos3k Awesome, could you submit a PR please

[fynn @fynnso]

was messing with the OpenAI base URL in Cursor and caught this

accounts/anysphere/models/kimi-k2p5-rl-0317-s515-fast

so composer 2 is just Kimi K2.5 with RL
at least rename the model ID

[Chris Hay @chrishayuk]

@humancompressed yeah it’s ridiculously simple once you realize it.. chain of thought and attention is pretty much everything

[Chris Hay @chrishayuk]

Who says MoE experts need to be neural https://t.co/rr2TzotKev

[Chris Hay @chrishayuk]

@humancompressed Chain of Thought just needs to rewrite it to the correct syntax and then you’re good. This is exactly how gpt-5 gpt-oss handles math today. Type in 25 times 30 and look at the reasoning and notice it will rewrite it out to 25 * 30 =

[Chris Hay @chrishayuk]

@NeelNanda5 thought you might find this interesting LLMs Don’t Calculate - They Just Remember Everything
https://t.co/DCu0W9viuQ

[Chris Hay @chrishayuk]

so i figured out how gpt-oss does math... https://t.co/DCu0W9vQko

[Chris Hay @chrishayuk]

just dropped my mcp client oauth library, which i think makes it easier to work with oauth mcp servers.. totally standalone not tied to any agent
framework.. https://t.co/q30HW9LSav

[Chris Hay @chrishayuk]

so i updated mcp-cli... i think this version is a lot better, still got a bit to go, but i think it's much better

[Chris Hay @chrishayuk]

Super happy about this technique of using templates to bootstrap projects while vibe coding with Claude Code
https://t.co/t7nMXhKmBv

[Chris Hay @chrishayuk]

@xundecidability but yeah harmony just isn't just gpt-oss, it's gpt-5 also. its just easier to see what's going on with gpt-oss

[Chris Hay @chrishayuk]

yeah, i cover that about 30 mins of the video... but there is essentially a hierarchy of alignment... what the model has been trained on, the developer prompt, they system prompt, user prompt etc.. but if lower hierarchy prompt conflicts with higher hierarchy prompt.. higher wins

[Chris Hay @chrishayuk]

@biobootloader https://t.co/l3CPm3xIDE

[Chris Hay @chrishayuk]

updated mcp-cli to use gpt-oss as default model, works pretty sweetly.. also now supports gpt-5 and claude 4.1

[Chris Hay @chrishayuk]

recently spent some time with danny talking about pitching. was fun but different. https://t.co/GFF5yZoAOU

[Chris Hay @chrishayuk]

i've now moved mcp-cli onto chuk-llm with streaming, and really think it improves the performance now. i feel i'm getting close to a v1 release