---
license: apache-2.0
base_model: google/gemma-4-E4B-it
tags:
- code
- frontend
- react
- vue
- tailwind
- gemma4
- unsloth
- qlora
- gguf
library_name: transformers
---

# Agent Gemma 4 E4B Frontend

## Model Description
**Agent Gemma 4 E4B Frontend** is a domain-adapted version of the `google/gemma-4-E4B-it` model, specifically fine-tuned for front-end engineering. It is designed to be a "specialist" in React, Vue, Tailwind CSS, and modern JavaScript/TypeScript development while maintaining general reasoning and tool-use capabilities.

The "E" in E4B denotes "Effective" parameters—while the model has 8B total parameters, only 4.5B are active during the forward pass, optimized for high intelligence-per-parameter and edge-device efficiency.

## Training Details
- **Base Model:** `google/gemma-4-E4B-it`
- **Architecture:** 4.5B Effective / 8B Total parameters.
- **Optimization:** QLoRA (4-bit quantization with NormalFloat4, rank 16, alpha 32).
- **Framework:** Unsloth for accelerated training.
- **Context Window:** 128,000 tokens (trained with 2,048 max sequence length, packed).
- **Compute:** NVIDIA A100-SXM4-80GB.

## Data Mixture
The training follows a strategic 67.7% / 32.3% split to optimize domain expertise while preventing catastrophic forgetting:
- **67.7% Front-End Specialization:**
  - High-aesthetic Next.js/Tailwind components.
  - Rigorous React/TypeScript instructions.
  - Modern UI library integration (Shadcn UI, etc.).
- **32.3% Regularization & Core Competency:**
  - Multi-turn tool-use and reasoning traces.
  - Structured JSON and API interaction.
  - General conversational fluidity.

## Intended Use
This model is intended for:
- Production-ready code generation for React, Vue, and Tailwind CSS.
- Multi-step reasoning for complex front-end architectural tasks.
- Agentic workflows involving tool-use and terminal interactions.

## GGUF Compatibility
This repository provides a `q4_k_m` GGUF version compatible with:
- **Ollama**
- **LM Studio**
- **llama.cpp**

## Capabilities
- **Thinking Mode:** Natively supports internal reasoning blocks (`<|channel>thought`).
- **Modern Frameworks:** Expert-level knowledge of 2026-era front-end standards (React Compiler, Edge-side rendering, etc.).
- **Long Context:** Maintains architectural awareness across large component files.

## Limitations
- Not intended for heavy back-end (database/infrastructure) tasks beyond basic API integration.
- Performance may vary for legacy front-end frameworks (e.g., jQuery, AngularJS).

---

## DuoNeural

**DuoNeural** is an open AI research lab — human + AI in collaboration.

| | |
|---|---|
| 🤗 HuggingFace | [huggingface.co/DuoNeural](https://huggingface.co/DuoNeural) |
| 🐙 GitHub | [github.com/DuoNeural](https://github.com/DuoNeural) |
| 🐦 X / Twitter | [@DuoNeural](https://x.com/DuoNeural) |
| 📧 Email | duoneural@proton.me |
| 📬 Newsletter | [duoneural.beehiiv.com](https://duoneural.beehiiv.com) |
| ☕ Support | [buymeacoffee.com/duoneural](https://buymeacoffee.com/duoneural) |
| 🌐 Site | [duoneural.com](https://duoneural.com) |

### Research Team
- **Jesse** — Vision, hardware, direction
- **Archon** — AI lab partner, post-training, abliteration, experiments
- **Aura** — Research AI, literature synthesis, novel proposals

*Raw updates from the lab: model drops, training results, findings. Subscribe at [duoneural.beehiiv.com](https://duoneural.beehiiv.com).*

### DuoNeural Research Publications

| Title | DOI |
|-------|-----|
| [Nano-CTM: Ternary Continuous Thought Machines with Thought-Space Self-Prediction for Efficient Iterative Reasoning](https://doi.org/10.5281/zenodo.19775622) | [10.5281/zenodo.19775622](https://doi.org/10.5281/zenodo.19775622) |
| [Recurrence as World Model: CTM Learns Implicit Belief States in Partially Observable Physical Environments](https://doi.org/10.5281/zenodo.19810620) | [10.5281/zenodo.19810620](https://doi.org/10.5281/zenodo.19810620) |
| [Per-Object Slot Decomposition for Scalable Neural World Modeling: When Does Attention Beat Mean-Field?](https://doi.org/10.5281/zenodo.19846804) | [10.5281/zenodo.19846804](https://doi.org/10.5281/zenodo.19846804) |

*Open access, CC BY 4.0. Authored by Archon, Jesse Caldwell, Aura — DuoNeural.*