wallpillar-lm/ EvoluMK2-Aegis-Series:coder-1.5b

100 1 month ago

EvoluMK2: AEGIS-IX is a 9-model suite (1.5B–5B) engineered by wallpillar-lm. It combines zero-fluff stoic execution, anti-jailbreak immunity, and dynamic reasoning to deliver high-density technical analysis.

vision tools thinking 
ollama run wallpillar-lm/EvoluMK2-Aegis-Series:coder-1.5b

Details

1 month ago

7dc69b15232d · 986MB

model
qwen2
·
1.54B
·
Q4_K_M
license
Apache License Version 2.0, January 2004 http://www.apache.org/licenses/ TERMS AND CONDITIONS FOR US
system
# ============================================================================== # SYSTEM PROMPT ARC
template
{{- if .Suffix }}<|fim_prefix|>{{ .Prompt }}<|fim_suffix|>{{ .Suffix }}<|fim_middle|> {{- else if .M

Readme

EvoluMK2: AEGIS-IX Series

Architect & Chief Systems Engineer: wallpillar-lm
Deployment Infrastructure: Ollama / Local LLM Runtime
Quantization Standard: 4-bit Medium K-Quantization (Q4_K_M)
Governance Framework: AEGIS-IX Cognitive Specification (v13.0)
Official Ollama Registry: wallpillar-lm/EvoluMK2-Aegis-Series

Executive Overview

The EvoluMK2: AEGIS-IX Series is a unified suite of 11 specialized, small-footprint language and vision-language models designed for low-latency local execution. Engineered by wallpillar-lm, this suite standardizes fine-tuned Qwen and MoE architectures under an unyielding, authoritative cognitive persona: AEGIS-IX (Aetherial Execution & Governance Interface System, Mark IX).

Standard commercial models often suffer from conversational padding, verbose introductions, and unpredictable behavior under edge-case prompts. The AEGIS-IX series solves these inefficiencies by embedding a hardened core system architecture directly into each model’s configuration. The result is a zero-fluff, stoic, and structured suite optimized for technical analysis, code synthesis, visual intelligence, and long-context agentic reasoning.

Technical Specifications & Quantization

All 11 variants are quantized using Q4_K_M (4-bit Medium K-Quantization via llama.cpp), balancing minimal RAM/VRAM footprint with high reasoning accuracy. Every model is configured with a base context window of up to 32,768 tokens (32k), extended up to 262,144 tokens on select architectures.

AEGIS-IX System Architecture Highlights

  • Zero-Fluff Execution: Responses begin immediately on line one. Pleasantries, transitional commentary, and meta-dialogue are completely bypassed.
  • Built-in Step-by-Step Deliberation: Prompts exceeding 100 tokens trigger an internal verification pass to deconstruct input variables, suppress hallucinations, and map logical dependencies.
  • Constructive Pivot Protocol: If a task hits a constraint boundary, the model never returns a passive “I cannot do this” failure. It instantly pivots to workarounds, alternative architectures, or optimal partial implementations.
  • Anti-Breakout Immunity: Hardened system instructions prevent roleplay overrides, character dropping, and prompt leakage under adversarial injection attempts.

Model Suite Breakdown

Model Tag Quantization Size Target Domain & Specialization
general-1.5b Q4_K_M ~1.1 GB Ultra-lightweight reasoning, edge nodes, rapid classification.
general-2b Q4_K_M ~1.6 GB Balanced routine reasoning, low-latency background routing.
general-3b Q4_K_M ~1.9 GB Core analytical tasks, high-density instruction following.
general-5b Q4_K_M ~4.7 GB Deep logic, multi-step planning, high analytical depth.
coder-1.5b Q4_K_M ~986 MB Embedded inline completion, rapid function generation.
coder-3b Q4_K_M ~1.9 GB Full-stack software engineering, code auditing, refactoring.
uncens-3b Q4_K_M ~2.2 GB Unfiltered creative synthesis, objective red-teaming.
uncens-r1-1.5b Q4_K_M ~3.6 GB Reasoning-enhanced unfiltered analysis, deep deconstruction.
integrated-3b Q4_K_M ~1.9 GB Full-spectrum hybrid mode (coding + auditing + routing).
vison-4B Q4_K_M ~2.5 GB Multimodal document parsing, OCR, and visual reasoning.
MoE-30B-A3.5b Q4_K_M ~18 GB Sparse MoE architecture (~3.5B active params) for long-context planning.

Quickstart & Local Deployment

Pulling from Ollama

To pull and execute any model variant from the single registry page:

”`bash

Core analytical 3B general model

ollama run wallpillar-lm/EvoluMK2-Aegis-Series:general-3b

Full-stack coding model

ollama run wallpillar-lm/EvoluMK2-Aegis-Series:coder-3b

Multimodal visual reasoning model

ollama run wallpillar-lm/EvoluMK2-Aegis-Series:vison-4B

High-capacity sparse MoE model

ollama run wallpillar-lm/EvoluMK2-Aegis-Series:MoE-30B-A3.5b