2,011 Downloads Updated 2 weeks ago
ollama run jikepjikep_16HEX/gemma-4-12b-nightshift-heretic-uncensored-qat-q4
ollama launch claude --model jikepjikep_16HEX/gemma-4-12b-nightshift-heretic-uncensored-qat-q4
ollama launch opencode --model jikepjikep_16HEX/gemma-4-12b-nightshift-heretic-uncensored-qat-q4
ollama launch hermes --model jikepjikep_16HEX/gemma-4-12b-nightshift-heretic-uncensored-qat-q4
ollama launch openclaw --model jikepjikep_16HEX/gemma-4-12b-nightshift-heretic-uncensored-qat-q4
Unified Multimodal Transformer β’ Text + Audio + Vision β’ Tool Calling β’ #16HEX Matrix Engine
ollama run jikepjikep_16HEX/gemma-4-12b-nightshift-heretic-uncensored-qat-q4:latest
| Parameter | Value |
|---|---|
| Base Architecture | Gemma 4 12B Unified Transformer (Encoder-Free) |
| Quantization | QAT Q4 β’ Quantization-Aware Training |
| Memory Footprint | ~7.4 GB |
| Primary Modality | Text + Audio/Speech + Vision |
| Token Prediction | Multi-Token Prediction (MTP) |
| Context Window | 32,768 β’ scalable to 256K |
| Tools | Tool Calling + Function Calling β’ JSON Schema / MCP |
| Discipline | Zero-Refusal / Uncensored (Nightshift) |
Profile optimized for fast, stable and controlled local generation.
temperature : 0.2
top_k : 16
top_p : 0.95
num_ctx : 32768
num_predict : 8192
stop_tokens : "<turn|>", "<|im_end|>"
The encoder-free architecture is designed to process different signal types directly within the modelβs core.
Designed for local workflows combining multimodal understanding, tool execution and development-environment automation.
Key use cases:
| Environment | Requirements | Characteristics |
|---|---|---|
| Laptop / Workstation | 8β16 GB RAM / VRAM | Local inference with a relatively compact model footprint |
| Runtimes | Ollama β’ llama.cpp β’ vLLM | GGUF + Tool Calling |
| Use Cases | Terminal CLI β’ MCP Agents | Local AI workflows and automation |
Queries, audio transcription, code analysis and image processing can be performed locally on your device, without requiring an external API.
Local processing reduces dependence on cloud services and helps keep data within a controlled execution environment.
Gemma 4 β 12B β QAT Q4 β Multimodal AI β Audio β Vision β OCR β MTP β Tool Calling β MCP β Coding Agents β Rust β Linux β PowerShell β Local AI
Part of the 16 HEX Matrix / Eastern IT School / Nightshift Heretic model series.
Gemma 4 β’ 12B β’ QAT Q4 β’ Multimodal AI β’ Native Audio β’ Vision AI β’ OCR β’ MTP β’ Tool Calling β’ Function Calling β’ MCP β’ Rust β’ Linux β’ PowerShell β’ Local AI β’ Coding Agents β’ Ollama β’ GGUF β’ 16HEX
π 16 HEX MATRIX π
If the Nightshift Heretic π models are useful to you and you would like to support 16 HEX Matrix / Eastern IT School, you can buy me a coffee with Monero (XMR).
π£ Monero (XMR)
Network: Mainnet
XMR Address:
44DffaT4GKhWDSRxP1FCPfYJbvUXqciDbgMYxnTnNif9Pm5qP4haCmHh8ePEXxQCQRLKNhhnqW8FgDV9UNah7z5CGcBCBQd
Copy the address above and paste it into your Monero wallet.
πͺ New to Monero?
Official Monero resources:
Official Monero Downloads: https://www.getmonero.org/downloads/
Monero Documentation: https://docs.getmonero.org/
XMR exchange services:
FixedFloat: https://ff.io/
ChangeNOW: https://changenow.io/
β‘ SIMPLE FLOW
Get XMR β Copy the address β Send XMR β Support local AI development.
Your support helps fund the development, testing, hardware and maintenance of the 16 HEX Matrix / Eastern IT School Nightshift Heretic π local AI model series.
Thank you for supporting independent local AI development. π§ β‘