krystv/neurolex-v4-creative-name-diffusion

🧠 NeuroLex v4 — Creative Name Diffusion Engine

A domain-specific AI architecture that generates truly creative, novel names for brands, YouTube channels, social media handles, and more — using Uniform Discrete Language Diffusion instead of autoregressive LLMs.

🚀 Quick Start (Colab)

# Clone and setup
!git clone https://huggingface.co/krystv/neurolex-v4-creative-name-diffusion
%cd neurolex-v4-creative-name-diffusion
!python setup.py  # ← IMPORTANT: fixes imports

# Train (~25 minutes on free T4)
!python train.py --size base --epochs 30 --batch_size 256

# Generate names
from neurolex_v4_model import *
checkpoint = torch.load('./checkpoints/neurolex_v4_best.pt')
config = NeuroLexConfig(**checkpoint['config'])
model = NeuroLexV4(config).cuda()
model.load_state_dict(checkpoint['state_dict'])
model.eval()

names = model.generate(
    domain_id=DOMAIN_TO_ID['tech'],
    style_id=STYLE_TO_ID['sharp'],
    lang_id=LANG_TO_ID['english'],
    target_length=8, batch_size=20,
    cfg_scale=2.5, temperature=0.9,
    n_steps=80, odd_alpha=8.0, device='cuda'
)
print(names)

🎯 The Problem We Solve

Why do LLMs and current AI name generators suck at creative naming?

Problem	Root Cause	Example
Repetition	AR probability feedback loops	Generates "Nexaflow" 50 times
Generic outputs	MLE training → common patterns	"TechFlow", "DataStream", "CloudSync"
Mode collapse	Small model memorizes modes	Only 47% uniqueness (v3)
Can't invent words	Subword tokenizers recombine known pieces	Just concatenation of morphemes
Sounds cringe	No phonotactic awareness	"Xyzptlk", "Blorpify"
No cultural sense	Ignores language-specific sound patterns	Same output for Japanese vs French vibe

✨ Our Solution: Discrete Diffusion (NOT Autoregressive)

LLM/GPT approach (BROKEN):
  [Start] → P(next|left) → P(next|left) → ... → same output every time

NeuroLex v4 (WORKS):
  [Random Noise] ← denoise ← denoise ← ... ← [Novel Name]
  (different noise each time = different output each time)

Key Innovations

Innovation	What It Does	Based On
UDLM	Uniform noise → iterative denoising	MDLM (NeurIPS 2024)
Classifier-Free Guidance	Control generation without mode collapse	Discrete CFG (2024)
ODD	Batch samples actively repel each other	ODD (2025)
adaLN	Condition modulates every layer	DiT (2023)
Cosine schedule	More refinement time at low noise	DDPM/MDLM
Character vocab	Generate truly novel sequences	ByT5 principles

📊 Specifications

Property	Value
Parameters	~12M (base)
Vocabulary	72 characters (a-z, A-Z, 0-9, specials)
Max name length	24 characters
Languages	25
Domains	20
Styles	10
Training time	~25 min on free Colab T4
GPU memory	<8 GB
Target diversity	90%+ uniqueness

📁 Repository Structure

├── neurolex_v4_model.py       # Core UDLM architecture (DiT + CFG + ODD)
├── neurolex_v4_dataset.py     # Built-in dataset (25 languages, 20 domains)
├── train.py                   # Training script (CLI)
├── generate.py                # Interactive generation script
├── test_model.py              # Validation tests
├── setup.py                   # Run first to fix imports
├── NeuroLex_v4_Training.ipynb # Complete Colab notebook
└── README.md                  # This file

🔬 Research Foundation

1. MDLM — NeurIPS 2024 (arxiv:2406.07524)
2. Discrete CFG — (arxiv:2412.10193)
3. ODD — (arxiv:2603.04893)
4. DiT — Peebles & Xie, 2023
5. GFlowNet — NeurIPS 2021 (arxiv:2106.04399)
6. Sound Symbolism — (arxiv:2310.16781)
7. ByT5 — (arxiv:2105.13626)
8. SimCTG — NeurIPS 2022 (arxiv:2202.06417)

📝 License

Apache 2.0

Generated by ML Intern

This model repository was generated by ML Intern, an agent for machine learning research and development on the Hugging Face Hub.

• Try ML Intern: https://smolagents-ml-intern.hf.space
• Source code: https://github.com/huggingface/ml-intern

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "krystv/neurolex-v4-creative-name-diffusion"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

For non-causal architectures, replace AutoModelForCausalLM with the appropriate AutoModel class.