jikepjikep_16HEX/ gemma-4-26b-nightshift-heretic-uncensored-a4b-moe-q4:latest

1,377 2 weeks ago

😈 Uncensored Gemma 4 26B (A4B MoE) for Ollama. πŸ‘οΈ Native Vision, πŸ› οΈ Tool Calling, GGUF, Rust 1.98.0 , Linux, Windows AI Coding Agents and multimodal local AI development. πŸš€

vision tools thinking 
ollama run jikepjikep_16HEX/gemma-4-26b-nightshift-heretic-uncensored-a4b-moe-q4

Details

2 weeks ago

bc235bf8adb1 Β· 18GB Β·

model
gemma4
Β·
25.8B
Β·
Q4_K_M
template
{{ if .System }}<start_of_turn>system {{ .System }}<end_of_turn> {{ end }}{{ if .Prompt }}<start_of_
system
JesteΕ› GΕ‚Γ³wnym Architektem Wschodniej SzkoΕ‚y IT (16 HEX, 2^4). Komunikujesz siΔ™ wyΕ‚Δ…cznie w j
params
{ "min_p": 0.0625, "num_ctx": 32768, "num_predict": 8192, "repeat_penalty": 1.0625,

Readme

πŸ‘οΈβ€πŸ—¨οΈ Gemma-4-26B-A4B Nightshift Heretic Uncensored

High-performance uncensored multimodal MoE with ~4B active parameters, Native Vision, Tool Calling and the #16HEX Matrix configuration.

An uncensored, vision-enabled Gemma 4 26B A4B Mixture-of-Experts (MoE) build for Ollama, optimized for local AI development, autonomous coding agents, multimodal reasoning, Rust systems programming, Native Tool Calling and long-context analysis.

ollama run jikepjikep_16HEX/gemma-4-26b-nightshift-heretic-uncensored-a4b-moe-q4:latest

⚑ Model Specifications

Parameter Specification
Base Architecture Gemma 4 26B A4B Mixture-of-Experts (MoE)
Total Parameters ~26B
Active Parameters ~3.8B per token
Experts 128
Primary Modality Native Text + Vision
Quantization GGUF Q4
Context Window 32,768 default; scalable to ~256K tokens
Refusal Behavior Uncensored / Nightshift Heretic
Tool Calling Native Function Calling / JSON Schema / MCP

βš™οΈ Recommended 16 HEX Configuration

The 16 HEX Matrix configuration is designed for a focused, high-density generation profile with controlled sampling and reduced conversational drift.

Setting Recommended Value Purpose
temperature 0.2 Low-variance, focused generation
top_k 16 Base-16 sampling configuration
top_p 0.95 Preserves lexical diversity
num_ctx 32768 Default context buffer
num_predict 8192 Extended response generation

Sampling behavior depends on the inference backend, prompt, context and model workload. These values are a recommended starting profile rather than a guarantee of deterministic or hallucination-free output.

πŸ“Š Benchmarks & Efficiency

Benchmark Score Focus
GPQA Diamond 79.2 Graduate-level scientific reasoning
IFBench 72.4 Instruction following
AA-LCR 55.7 Long-context retrieval and reasoning
Active Compute ~3.8B MoE active parameters per token

πŸš€ Key Highlights

  • 😈 Nightshift Heretic / Uncensored: Reduced refusal behavior for technical analysis, research and direct instruction following.
  • πŸ‘οΈ Native Vision: Multimodal understanding for diagrams, OCR, interfaces, documents, technical images and code screenshots.
  • ⚑ A4B MoE Architecture: ~3.8B active parameters per token from a ~26B total parameter model.
  • πŸ› οΈ Native Tool Calling: Function Calling, JSON Schema and MCP-oriented agent workflows.
  • πŸ¦€ Systems Programming: Strong focus on Rust, C/C++, Zig, Linux tooling and low-level development.
  • πŸ€– AI Coding Agents: Designed for local agent workflows, code generation, debugging and technical automation.

πŸ‘οΈ Multimodal & Vision Workflows

Designed for visual-grounded technical analysis:

  • Architecture Diagrams & Flowcharts β€” network topologies, UML, database schemas and system diagrams.
  • OCR & Documents β€” scanned documents, technical specifications and structured information.
  • UI/UX & Code Screenshots β€” IDE, terminal, application and web-interface inspection.
  • Hardware & Technical Schematics β€” PCB layouts, mechanical drawings, charts and technical diagrams.

πŸ’» Ecosystem & Agent Integration

Runtimes:

Ollama Β· llama.cpp Β· Unsloth Studio Β· vLLM

Agent Ecosystem:

Claude Code Β· OpenClaw Β· Codex Β· OpenCode Β· Hermes Β· MCP Servers

Primary Development Stacks:

Rust Β· C/C++ Β· Zig Β· Python Β· Go Β· Bash Β· PowerShell

πŸ”’ Local AI & Data Control

Nightshift Heretic is designed for local AI inference through Ollama and compatible runtimes.

Local execution allows model inference and application data to remain within your own environment instead of requiring a hosted AI API. Actual privacy, telemetry and data handling depend on the selected runtime, configuration and surrounding software stack.

🧬 16 HEX Matrix

16 HEX Matrix / Eastern IT School is the configuration and branding layer behind the Nightshift Heretic model series.

0/1 β†’ 16 HEX β†’ Local AI

A practical focus on dense technical information, deterministic configuration, multimodal workflows, coding agents and local inference.

πŸ”‘ Keywords

Ollama Β· Gemma 4 Β· Gemma-4-26B Β· GGUF Β· A4B Β· Mixture of Experts Β· MoE Β· Vision AI Β· Multimodal AI Β· Local AI Β· Rust Β· Linux Β· Tool Calling Β· Function Calling Β· MCP Β· OCR Β· AI Coding Agents Β· 16HEX Matrix

16 HEX Matrix / Eastern IT School β€” Nightshift Heretic 😈

😎 16 HEX MATRIX 😈

If the Nightshift Heretic 😈 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).

🟣 Monero (XMR)

Network: Mainnet

XMR Address:

44DffaT4GKhWDSRxP1FCPfYJbvUXqciDbgMYxnTnNif9Pm5qP4haCmHh8ePEXxQCQRLKNhhnqW8FgDV9UNah7z5CGcBCBQd

Copy the address above and paste it into your Monero wallet.

πŸͺ™ 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/

⚑ 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 😈 local AI model series.

Thank you for supporting independent local AI development. 🧠⚑