Pinned Loading

1. memory-bench memory-bench Public

   Controlled benchmark of memory mechanisms for transformers, built on nanochat

   Python

2. logprobe logprobe Public

   Detect normalization errors and entropy bias in LLM logprob data

   Rust

3. openai/parameter-golf openai/parameter-golf Public

   Train the smallest LM you can that fits in 16MB. Best model wins!

   Python 5.1k 3.3k

4. logprobe-calibration logprobe-calibration Public

   Calibration Under API Logprob Truncation — evidence from 16 model-provider conditions across OpenAI, Together AI, Groq, and self-hosted vLLM

   Python

5. FormalSLT FormalSLT Public

   Zero-sorry Lean 4 library: finite-sample SLT bounds, sharp McDiarmid, PAC-Bayes Bernstein margin shell, and Dudley chaining. Standard Lean/Mathlib axioms only.

   Lean 33 1

6. spectral-pde-lab spectral-pde-lab Public

   Source snapshot of browser-native spectral PDE demos, notes, and Python comparison code

   TypeScript 1