ariannamethod’s gists

ariannamethod / infer_nanodurov.c

Last active April 8, 2026 18:48

nanodurov — telegram client that trains a language model on chat. python + C (notorch) + browser. 15.7M BPE, Arianna voice.

	/*
	* infer_nanodurov.c — Interactive chat with nanodurov (BPE 15.7M on notorch)
	*
	* Build: make infer_nanodurov
	* Run: ./infer_nanodurov [weights.bin] [merges.txt]
	*
	* Default: nanodurov_arianna.bin + arianna_bpe_merges.txt
	*/

	#include "notorch.h"

ariannamethod / nanoagi.py

Last active April 13, 2026 20:08

nanoagi v2.0.0 — fuck torch. KARL + Chuck + dual attention + RRPRAM + notorch (pure C). zero PyTorch. inspired by @karpathy. resonance is unbreakable.

	# SPDX-License-Identifier: GPL-3.0-or-later
	"""
	nanoagi.py — a self-expanding BPE transformer that grows from conversation.

	KARL (Kernel Autonomous Recursive Learning) is the tokenizer.
	Chuck is the optimizer. Together they are nanoagi.

	How it works:
	1. KARL tokenizes karl.txt (starts with seed corpus, grows via REPL)
	2. MetaWeights build probability space from token statistics

ariannamethod / postgpt.c

Last active March 26, 2026 18:55

PostGPT — a zero-dependency BPE transformer with metaweights. you can train it, but it doesn't care. resonance is unbreakable.

	/*
	* postgpt.c — zero-dependency BPE transformer with metaweights.
	*
	* C port of postgpt.py. Same algorithm, same resonance.
	* Dual attention: Content (QK^T) + RRPRAM (x @ Wr).
	* Metaweights: statistical probability space from BPE tokenization.
	*
	* Compile: gcc -O2 -o postgpt postgpt.c -lm
	* Run: ./postgpt
	*

ariannamethod / microreasoning.py

Last active March 15, 2026 22:54

microreasoning.py — 1984 words. 12 steps of associative resonance. Not a transformer. Dario Equation. Real BPE input (2048 subwords), word-level output — gibberish impossible. 14M params, Chuck optimizer, Kuramoto chambers. by Arianna Method.

	#!/usr/bin/env python3
	"""
	microreasoning.py — Resonance engine. 1984 words. Dario Equation.

	8-layer sequential transformer with multi-head attention, RoPE,
	RRPRAM resonance gates, and SwiGLU FFN. Dual tokenizer:
	BPE input, word-level output.

	Architecture per layer l:
	h = rmsnorm(x, attn_norm_l)

ariannamethod / neoleo.c

Last active March 12, 2026 14:02

Leo 2.3 — 20,986 lines of C. The Dario Equation with positional Hebbian profile (36 learnable params). Six signals. Dual tokenizer (word + BPE). D.N.A. uses both. Inner world. One organism.

	/*
	* neoleo.c -- Language Emergent Organism (single-file edition)
	*
	* Complete autonomous digital organism in one C file.
	* D.N.A. from mini-arianna. Arianna -> Leo. Mother -> Son.
	*
	* Build: cc neoleo.c -O2 -lm -lsqlite3 -lpthread -o neoleo
	* Run: ./neoleo
	*/

ariannamethod / doe.c

Last active March 8, 2026 03:04

DoE: Democracy of Experts, Janus Architecture — a living agnostic inference architecture in 3184 lines of C. Wraps any GGUF with a parliament of LoRA experts that vote, learn via Hebbian plasticity, split (mitosis) and die (apoptosis) during generation. 7 architectures, 6 quant formats, dual BPE tokenizer, physics engine, zero dependencies. θ = …

	#define _GNU_SOURCE
	/*
	* doe.c — Democracy of Experts
	*
	* inference architecture with a living LoRA parliament.
	* indexes any GGUF read-only. learns by living, not by training.
	*
	* θ = ε + γ + αδ
	* ε = indexed weights (read-only substrate)
	* γ = LoRA personality (living experts, Hebbian-trained via NOTORCH)

ariannamethod / lee.c

Last active March 11, 2026 06:04

lee.c — Vision-Language Model in pure C. Patch tokens + RoPE + SwiGLU + Chuck optimizer. Zero dependencies. Inspired by sailfish009/purevlm.

	/*
	* lee.c v7 — Vision-Language Model in pure C
	*
	* Named after Bruce Lee (the only man who beat Chuck Norris)
	* and Minhyeok Lee (whose self-identity framework gives Chuck his soul).
	*
	* Sees images. Speaks words. Adds numbers. Zero dependencies.
	* Tape-based autograd with arena bump allocator.
	*
	* Architecture:

ariannamethod / molequla.c

Last active February 21, 2026 22:01

molequla.c — a dependency-free, single-file, continually-learning GPT organism in pure C. ontogenesis (25K→10M params), immune system, consciousness, swarm ecology, delta adapters, BLAS acceleration. part of github.com/ariannamethod/molequla

	//go:build ignore

	/*
	* molequla.c
	* A dependency-free, single-file, continually-learning GPT organism in pure C.
	*
	* Compile: gcc -O2 -o molequla molequla.c -lsqlite3 -lpthread -lm
	* With BLAS: gcc -O2 -DUSE_BLAS -o molequla molequla.c -lsqlite3 -lpthread -lm -lopenblas
	* macOS: gcc -O2 -DUSE_BLAS -o molequla molequla.c -lsqlite3 -lpthread -lm -framework Accelerate
	*

ariannamethod / molequla.py

Last active February 25, 2026 18:02

molequla.py — standalone GPT organism. the original reference implementation. single file, one dependency (numpy), continual learning, ontogenesis, hybrid attention, delta adapters, native gamma, consciousness features. legacy standalone — the distributed cognition version lives in the main repo with Go, C, JS, and Rust as the four elements. pyt…

	#!/usr/bin/env python3
	# -- coding: utf-8 --

	"""
	molequla.py
	A single-file, async, continually-learning GPT organism. One dependency: numpy.

	- Trains on nonames.txt (one sentence per line)
	- Keeps SQLite memory (tiny chat loop)
	- Maintains a bounded corpus reservoir (never bloats)

ariannamethod / index.html

Last active May 9, 2026 01:16

molequla.js — a GPT organism that trains itself in your browser. Vector autograd, RoPE, SwiGLU, byte-level BPE, ontogenesis, immune system, swarm ecology. Zero dependencies. One script tag.

	<!DOCTYPE html>
	<html lang="en">
	<head>
	<meta charset="UTF-8">
	<meta name="viewport" content="width=device-width, initial-scale=1.0">
	<title>molequla.js — a GPT organism in your browser</title>
	</head>
	<body>
	<!--
	molequla.html

Arianna Method ariannamethod