NIRVANA Cognitive Cartridges: Frozen 7B Models Get Removable Numerical Memory on Qwen and Mistral
Nearby_Indication474 · reddit · 2026-10-05
A Redditor demonstrates the NIRVANA "Cognitive Cartridge" architecture: a frozen 7B model processes text, its internal K/V states are compressed via a fixed source-independent codebook into an unreadable numerical memory module, and the original text is discarded. Questions can later be answered from the cartridge alone, with no source text in the prompt.
The mechanism has now been reproduced on Qwen2.5-7B-Instruct and Mistral-7B-Instruct-v0.3 — architecturally different families with different layer counts and KV structures. The Mistral run used 16 independently forged cartridges holding separate records, installed and queried without concatenation. It is not RAG, not fine-tuning, not a prompt template: weights stay frozen while information lives in numerical K/V-derived representations.
Zenodo DOI records are published for both models. Results remain a personal experiment; independent verification is still pending.
More from Research
- Margaret Mitchell: feeding models their own energy signals could cut LLM power use — mmitchell_ai · 2026-10-05
- LLM-Assisted Math: Mathieu Group M_23 Proven Galois over Q, Video Explains — _sathvikr · 2026-10-05
- LLM consciousness debate: if LLMs are functionally equivalent, burden of proof is on deniers — mathemagic1an · 2026-10-05
- Nobel Prizes 2026 start Oct 5 after AI-driven chemistry win for Hassabis and Jumper — chaitjo · 2026-10-05
- Google touts AI for Science push: AlphaGenome Atlas maps 9B genetic variants, WeatherNext 3 launches — demishassabis · 2026-10-05
- Fisher information gets a rare dedicated book showcasing versatile applications — FrnkNlsn · 2026-10-05