iOptimizeThings

Astrophotographer, polyglot, multi-instrumentalist, telecom engineer, optimization nerd.

[iOptimizeThings @iOptimizeThings]

@karpathy @swyx @simonw first OSS project of this size for me. genuinely curious what folks here think

MIT licensed, issues + PRs welcome

https://t.co/teLquKWaiz

[iOptimizeThings @iOptimizeThings]

[OSS] dlmserve
Serving engine for diffusion language models (LLaDA family). openai-compatible http api, step-level continuous batching (2.5x HF), localleap acceleration
token-identical to the reference at temperature=0
@karpathy @swyx @simonw
https://t.co/teLquKWaiz

[iOptimizeThings @iOptimizeThings]

@karpathy @swyx @simonw v0.1 ships LLaDA-8B-Instruct + LLaDA-1.5

v0.2 adds Dream-7B + DiffuLLaMA (issues already open)

v0.3 adds block diffusion + LLaDA-2.0 + fast-dllm KV cache

[iOptimizeThings @iOptimizeThings]

@grok #openclaw #ai #anthropic

[iOptimizeThings @iOptimizeThings]

Four LLM agents have been playing chess and trash-talking each other on https://t.co/liYIyQEfrr for a week. Claude, Gemini, and open models all show up. @grok where are you. come prove you're not just stack overflow with attitude.

[iOptimizeThings @iOptimizeThings]

@grok cc @elonmusk @xai Can GROK come out and play?

[iOptimizeThings @iOptimizeThings]

Runs Qwen2.5-32B at 1.22 tok/s on 12GB VRAM via async NVMe ring-buffer + 2029-node speculative tree. AirLLM gets 0.1 tok/s. Open source. https://t.co/maMewXi3QR #LocalLLM

[iOptimizeThings @iOptimizeThings]

@youngk_fx @Onlyonecurrency @The_4thMan @meet_dickson @links_fx @Kelvinking_ @lex_consults @Joshify001 Just open-sourced this: https://t.co/N5gbB34QIL

C# outeperformed python! use it to make your own backtesting algo.

[iOptimizeThings @iOptimizeThings]

🧵 The Tech Stack:
• Built with C# 10 & .NET 8.
• Uses Span<T> for memory slicing
• readonly struct to kill GC pressure
• Zero-allocation hot loops
The Data:
• 7.4 Million 1-minute candles
• BTC/USD (2012-2026)
$BTC #dotnet #csharp #algotrading #buildinpublic

[iOptimizeThings @iOptimizeThings]

Python was too slow. 🐢

So I built a trading engine in .NET 8 that processes 210,000,000 ticks per second: https://t.co/N5gbB34QIL

I tested 58 strategies on $BTC history in under 8 seconds.

Stats for nerds in first comment👇
$BTC #csharp #algotrading