ModelAufgelistetExtern

Cozet

Architecture: SYNAXIM (Symbiotic Native Axim Inference Machine) Author: GRRN Research License: Proprietary Status: Phase 1 --- Architecture Validation (130M parameters)

Quellehuggingface.co/GRRNMAKER/Cozet/tree/badcd4ae730588b83e29274675973dd46a025677

ArtOtherVersionRevision badcd4aCBewertung

Modellquelle

Art: Other
Dateigröße: 129 MB–1,61 GB
Version: vRevision badcd4a
Aktualisiert: 22. Okt. 2026
Downloads: 0
Quelle: Hugging Face

Quellenbeschreibung

Cozet --- Native SYNAXIM Base Model

Architecture: SYNAXIM (Symbiotic Native Axim Inference Machine) Author: GRRN Research License: Proprietary Status: Phase 1 --- Architecture Validation (130M parameters)

What is Cozet?

Cozet is a language model built from scratch on the SYNAXIM architecture --- a non-transformer design that replaces self-attention with a persistent associative memory matrix (the Symbiotic Gate).

Cozet is not a converted transformer. It is born native. Every weight is trained from random initialization through the M-matrix paradigm. No KV cache. O(1) memory. Infinite context.

Architecture: Symbiotic Gate

Standard transformers compute attention as:

output = softmax(Q @ K^T / sqrt(d)) @ V    # O(n^2) compute, O(n) KV cache

SYNAXIM computes attention as:

gate   = sigmoid(mean(Q * K))               # scalar routing
M_new  = gate * M + (1-gate) * outer(k, v)  # O(1) persistent memory update
output = q @ M_new                           # associative retrieval

The M-matrix accumulates context through gated outer product updates and never grows. Memory is O(D^2) fixed regardless of sequence length.

Model Configurations

Config	Parameters	D	Layers	Heads	Intermediate	Status
Cozet-Small	130M	1024	12	16/4 GQA	4096	Phase 1
Cozet-Medium	1.3B	2048	24	16/4 GQA	8192	Phase 2
Cozet-Large	7.2B	4096	32	32/8 GQA	14336	Phase 3

Training

Native pretraining from randomly initialized weights on FineWeb-Edu / RedPajama using truncated BPTT through the M-matrix chain.

Download files

Modellartefakte

64 Artefakte

checkpoints/cozet_2.5m_final.pt

pt · 892 MB · SHA-256 89e988bb1745…aea5 · Hugging Face

Download file

checkpoints/cozet_50k_synthetic.pt

pt · 821 MB · SHA-256 0df9a8dcf1dd…005a · Hugging Face

Download file

checkpoints/cozet_5m_final.pt

pt · 892 MB · SHA-256 89af14d67d85…4c6c · Hugging Face

Download file

checkpoints/cozet_grown_24L.pt

pt · 1,61 GB · SHA-256 08cc4f63715b…4ecc · Hugging Face

Download file

checkpoints/cozet_lora_adapters.pt

pt · 129 MB · SHA-256 c8a4501ff4b1…4271 · Hugging Face

Download file

checkpoints/cozet_merged.pt

pt · 952 MB · SHA-256 0d36ff19337f…95b6 · Hugging Face

Download file

checkpoints/cozet_nemotron_decay.pt

pt · 1,61 GB · SHA-256 92044cbcdb9e…311d · Hugging Face

Download file

checkpoints/cozet_p2_step10000.pt

pt · 1,61 GB · SHA-256 67c5e9d14f47…3c2f · Hugging Face

Download file

checkpoints/cozet_p2_step100000.pt

pt · 892 MB · SHA-256 236ed2a683cf…129c · Hugging Face

Download file

checkpoints/cozet_p2_step101000.pt

pt · 892 MB · SHA-256 e97c14f646f9…d16d · Hugging Face

Download file

checkpoints/cozet_p2_step102000.pt

pt · 892 MB · SHA-256 249311eaea68…1e0e · Hugging Face

Download file

checkpoints/cozet_p2_step103000.pt

pt · 892 MB · SHA-256 3b9e0d76a8f4…c5ac · Hugging Face

Download file

checkpoints/cozet_p2_step104000.pt

pt · 892 MB · SHA-256 681ab8e8eaee…4b0e · Hugging Face

Download file

checkpoints/cozet_p2_step105000.pt

pt · 892 MB · SHA-256 c9d975eab757…0b79 · Hugging Face

Download file

checkpoints/cozet_p2_step106000.pt

pt · 892 MB · SHA-256 e185858e2c6d…c77a · Hugging Face

Download file

checkpoints/cozet_p2_step107000.pt

pt · 892 MB · SHA-256 3b6aea0e95a9…5b84 · Hugging Face

Download file

checkpoints/cozet_p2_step108000.pt

pt · 892 MB · SHA-256 6b09e7409200…f50d · Hugging Face

Download file

checkpoints/cozet_p2_step109000.pt

pt · 892 MB · SHA-256 a77c233b2541…e54c · Hugging Face

Download file

checkpoints/cozet_p2_step11000.pt

pt · 1,61 GB · SHA-256 1a171e2edc33…ffc7 · Hugging Face

Download file

checkpoints/cozet_p2_step110000.pt

pt · 892 MB · SHA-256 97077b991ef6…3b08 · Hugging Face

Download file

checkpoints/cozet_p2_step111000.pt

pt · 892 MB · SHA-256 b9d8c4033646…d78c · Hugging Face

Download file

checkpoints/cozet_p2_step112000.pt

pt · 892 MB · SHA-256 aa3ba9665383…e7a6 · Hugging Face

Download file

checkpoints/cozet_p2_step113000.pt

pt · 892 MB · SHA-256 c6e721a2d387…5bce · Hugging Face

Download file

checkpoints/cozet_p2_step114000.pt

pt · 892 MB · SHA-256 39a9e46a7e22…0888 · Hugging Face

Download file

checkpoints/cozet_p2_step115000.pt

pt · 892 MB · SHA-256 d3dfae143f01…f5c3 · Hugging Face

Download file

checkpoints/cozet_p2_step116000.pt

pt · 892 MB · SHA-256 93bea3f84c07…6b0a · Hugging Face

Download file

checkpoints/cozet_p2_step117000.pt

pt · 892 MB · SHA-256 cbf0926603e5…6fe4 · Hugging Face

Download file

checkpoints/cozet_p2_step118000.pt

pt · 892 MB · SHA-256 7401d4d338ff…d1c4 · Hugging Face

Download file

checkpoints/cozet_p2_step119000.pt

pt · 892 MB · SHA-256 fdbfe964fd7b…4ee2 · Hugging Face

Download file

checkpoints/cozet_p2_step12000.pt

pt · 1,61 GB · SHA-256 7e124cb432d1…4a9d · Hugging Face

Download file

checkpoints/cozet_p2_step120000.pt

pt · 892 MB · SHA-256 9edbd975ae19…f174 · Hugging Face

Download file

checkpoints/cozet_p2_step121000.pt

pt · 892 MB · SHA-256 7ffcac8129db…f3c7 · Hugging Face

Download file

checkpoints/cozet_p2_step122000.pt

pt · 892 MB · SHA-256 72d47d8b0198…269c · Hugging Face

Download file

checkpoints/cozet_p2_step123000.pt

pt · 892 MB · SHA-256 48f9864f2f03…a187 · Hugging Face

Download file

checkpoints/cozet_p2_step124000.pt

pt · 892 MB · SHA-256 8bd9b9b796d5…321e · Hugging Face

Download file

checkpoints/cozet_p2_step125000.pt

pt · 892 MB · SHA-256 d539d219b52c…70fd · Hugging Face

Download file

checkpoints/cozet_p2_step126000.pt

pt · 892 MB · SHA-256 fbef5e26ed38…5f06 · Hugging Face

Download file

checkpoints/cozet_p2_step127000.pt

pt · 892 MB · SHA-256 94025e271ca7…e593 · Hugging Face

Download file

checkpoints/cozet_p2_step128000.pt

pt · 892 MB · SHA-256 128f5761a844…0055 · Hugging Face

Download file

checkpoints/cozet_p2_step129000.pt

pt · 892 MB · SHA-256 9b55cad75aea…768f · Hugging Face

Download file

checkpoints/cozet_p2_step13000.pt

pt · 1,61 GB · SHA-256 1c1d8145807a…482e · Hugging Face

Download file

checkpoints/cozet_p2_step130000.pt

pt · 892 MB · SHA-256 fb47d5c6297f…a4a1 · Hugging Face

Download file

checkpoints/cozet_p2_step131000.pt

pt · 892 MB · SHA-256 abe147768b90…c499 · Hugging Face

Download file

checkpoints/cozet_p2_step132000.pt

pt · 892 MB · SHA-256 bb8e964b2b9e…7a1f · Hugging Face

Download file

checkpoints/cozet_p2_step133000.pt

pt · 892 MB · SHA-256 6c037c1b467c…cb86 · Hugging Face

Download file

checkpoints/cozet_p2_step134000.pt

pt · 892 MB · SHA-256 dd0ec37a3111…a0e6 · Hugging Face

Download file

checkpoints/cozet_p2_step135000.pt

pt · 892 MB · SHA-256 e60435805b34…4274 · Hugging Face

Download file

checkpoints/cozet_p2_step136000.pt

pt · 892 MB · SHA-256 bc3bb552ddb5…20f4 · Hugging Face

Download file

checkpoints/cozet_p2_step137000.pt

pt · 892 MB · SHA-256 fae981864bf1…cc5c · Hugging Face

Download file

checkpoints/cozet_p2_step138000.pt

pt · 892 MB · SHA-256 5bf1c1079925…9296 · Hugging Face

Download file

checkpoints/cozet_p2_step139000.pt

pt · 892 MB · SHA-256 5c99557a85dc…7c28 · Hugging Face

Download file

checkpoints/cozet_p2_step14000.pt

pt · 1,61 GB · SHA-256 4fb555ed372d…df84 · Hugging Face

Download file

checkpoints/cozet_p2_step140000.pt

pt · 892 MB · SHA-256 b426d20c2431…422e · Hugging Face

Download file

checkpoints/cozet_p2_step141000.pt

pt · 892 MB · SHA-256 1f6ef44943a7…1bae · Hugging Face

Download file

checkpoints/cozet_p2_step142000.pt

pt · 892 MB · SHA-256 9201923460c0…6b12 · Hugging Face

Download file

checkpoints/cozet_p2_step143000.pt

pt · 892 MB · SHA-256 c9df8c73574d…69c5 · Hugging Face

Download file

checkpoints/cozet_p2_step144000.pt

pt · 892 MB · SHA-256 5fdbc308882d…7323 · Hugging Face

Download file

checkpoints/cozet_p2_step145000.pt

pt · 892 MB · SHA-256 9f203857138c…c39e · Hugging Face

Download file

checkpoints/cozet_p2_step146000.pt

pt · 892 MB · SHA-256 020e9ab82461…d947 · Hugging Face

Download file

checkpoints/cozet_p2_step147000.pt

pt · 892 MB · SHA-256 716575264188…5940 · Hugging Face

Download file

checkpoints/cozet_p2_step148000.pt

pt · 892 MB · SHA-256 1f59e36dd06c…4756 · Hugging Face

Download file

checkpoints/cozet_p2_step149000.pt

pt · 892 MB · SHA-256 2e9f7a599a76…0deb · Hugging Face

Download file

checkpoints/cozet_p2_step15000.pt

pt · 1,61 GB · SHA-256 f2a4f28d44b9…4af3 · Hugging Face

Download file

checkpoints/cozet_p2_step150000.pt

pt · 892 MB · SHA-256 a6057b72fca5…1773 · Hugging Face

Download file

Quellenauszüge

2 Auszüge

Readme17. Julihuggingface.co/GRRNMAKER/Cozet

# Cozet --- Native SYNAXIM Base Model **Architecture**: SYNAXIM (Symbiotic Native Axim Inference Machine) **Author**: GRRN Research **License**: Proprietary **Status**: Phase 1 --- Architecture Validation (130M parameters) --- ## What is Cozet? Cozet is a language model built from scratch on the **SYNAXIM architecture** --- a non-transformer design that replaces self-attention with a persistent associative memory matrix (the Symbiotic Gate). Cozet is **not** a converted transformer. It is born native. Every weight is trained from random initialization through the M-matrix paradigm. No KV cache. O(1) memory. Infinite context. ## Architecture: Symbiotic Gate Standard transformers compute attention as: ``` output = softmax(Q @ K^T / sqrt(d)) @ V # O(n^2) compute, O(n) KV cache ``` SYNAXIM computes attention as: ``` gate = sigmoid(mean(Q * K)) # scalar routing M_new = gate * M + (1-gate) * outer(k, v) # O(1) persistent memory update output = q @ M_new # associative retrieval ``` The M-matrix accumulates context through gated outer product updates and never grows. Memory is O(D^2) fixed regardless of sequence length. ## Model Configurations | Config | Parameters | D | Layers | Heads | Intermediate | Status | |--------|-----------|------|--------|-------|-------------|--------| | Cozet-Small | 130M | 1024 | 12 | 16/4 GQA | 4096 | Phase 1 | | Cozet-Medium | 1.3B | 2048 | 24 | 16/4 GQA | 8192 | Phase 2 | | Cozet-Large | 7.2B | 4096 | 32 | 32/8 GQA | 14336 | Phase 3 | ## Training Native pretraining from randomly initialized weights on FineWeb-Edu / RedPajama using truncated BPTT through the M-matrix chain. ## Links - **Architecture**: SYNAXIM --- https://github.com/GRRN-MAKER/SYNAXIM - **Theory**: Odyssey Meta-Cognitive Framework - **Inference Engine**: SYNAXIM `.symb` format (pure NumPy, zero dependencies) --- (c) 2026 GRRN Research. All rights reserved.

Note17. Julihuggingface.co/GRRNMAKER/Cozet

Source context: 0 downloads · 1 likes · Repo GRRNMAKER/Cozet