FerrellSyntheticIntelligence/FSI_FELON

FSI FELON

Ferrell Synthetic Intelligence — Foundational Emergent Linguistic Observation Network

A production-grade hybrid neural-symbolic code generation platform combining:

• GatedConv + GQA neural architecture (4.5M / 10M params)
• BPE tokenizer (vocab 4096, trained on Python)
• Q-NFRE cognitive state engine (priority-driven inference)
• 59 code domain templates with multi-language generation

Quick Start

pip install -r requirements.txt
python fsi_terminal.py

Models

File	Params	Status
felon_codegen_best.pt	4.5M	Train complete, syntax-verified
felon_10m_codegen_best.pt	10M	Training in progress

Architecture

• quantum/gated_conv_engine.py — GatedConv1D + Grouped-Query Attention
• quantum/bpe_tokenizer.py — Byte-Pair Encoding tokenizer
• quantum/felon_inference.py — Unified load/generate entry point
• quantum/engine.py — Q-NFRE priority-scored cognitive state
• agent/agent.py — Priority 0 neural → template fallback agent
• domain_templates.py — 59 code generation templates
• chimera/engine.py — Multi-model routing engine

Training

# 4.5M model
python train_fast.py

# 10M model
python train_10m_full.py

License

MIT