Models

A lightweight, hardware-aware router built on Gemma 4 E2B. It acts as a dispatcher for local AI setups, automatically deciding whether a prompt should run on edge hardware (like a Jetson Nano), a local GPU, or the cloud based on task complexity.

Uncensored Gemma 4 E4B & E2B (0/465 refusals) with MTP drafter. 32k context, up to 80 TPS on RTX 3050 8GB.

Gemma 4 E4B & E2B (Q4_K_M) with MTP drafter. 32k context, up to 80 TPS on RTX 3050 8GB.

Fast, lightweight Gemma 4 E2B coding agent for Claude Code, 64K context, native tool-calling, 100% GPU on 16GB Apple Silicon.

Gemma 4 models are designed to deliver frontier-level performance at each size. They are well-suited for reasoning, agentic workflows, coding, and multimodal understanding.

Gemma 4 E2B-IT uncensored by HauhauCS. 0/465 Refusals

Gemma 4 E2B (Google DeepMind) with thinking mode disabled. Compact multimodal model — 2.3B effective / 5.1B total parameters. Supports text, image and audio input. Designed for edge devices and local deployment. Knowledge cutoff: January 2025.

Based in gemma-4-E2B-it-UD-Q2_K_XL.gguf. Context 4096 and no thinking enabled in template.

Gemma 4 E2B (Google DeepMind) with thinking mode enabled. Compact multimodal model — 2.3B effective / 5.1B total parameters. Supports text, image and audio input. Designed for edge devices and local deployment. Knowledge cutoff: January 2025.

QLoRA fine-tune of Gemma 4 E2B for scam detection. F1 86.1% / FPR 1.1% on a 300-sample real test set. 12-tool function calling for SMS, email, voice transcripts, and OCR'd MMS images.

Labadain E2B-it Tetun is a Gemma 4 E2B model fine-tuned on 138,033 Tetun multi-turn conversations (roughly 102 million tokens) covering news, government, health, education, politics, financial, general knowledge and everyday chat.

THOX Open minimal-refusal vision build on Gemma 4 E2B (abliterated) — THOX-configured, not THOX-trained. 120.44 tok/s, 1.86 GB loaded. Minimal-refusal build: can produce content other models refuse. Apache-2.0.

Google Gemma 4 E2B QAT checkpoint converted to GGUF by TurinTech. Text only; size tags map to the Hugging Face files. Measured Q3_K_L: Mean KLD 0.016, Same top p 95.8% vs BF16 on 20 fixed documents (the selection tier); protocol ryzen9950x-2026-10-12.

A local-first psychology assistant fine-tuned from Gemma 4 on real therapy dialogues. Two model series: E2B (2.3B, 3.3–5.0 GB) and 12B (11.96B, 6.2–11.8 GB). 128K context. Legal Notice: Not a substitute for professional care.

A Q4_K_M GGUF of Gemma 4 E2B, fine-tuned on Apple Silicon (M3 Pro, 18 GB) using MLX LoRA to act as an autonomous AIOps Orchestrator Agent.

Fine-tuned Gemma 4 E2B Q8 model for generating dense, production-ready `SKILL.md` files for autonomous AI agents

Gemma 4 E2B fine-tune for biological microscopy · 95 genera · 89% genus accuracy · runs offline · Apache-2.0 + CC-BY 4.0

Gemma 4 models are designed to deliver frontier-level performance at each size. They are well-suited for reasoning, agentic workflows, coding, and multimodal understanding.