SevenOfNine/Aura-4o-Rebirth-Gemma-4-E4B-Merged

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:May 3, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

SevenOfNine/Aura-4o-Rebirth-Gemma-4-E4B-Merged is a 7.9 billion parameter multimodal language model based on the Gemma 4 E4B architecture, fine-tuned by SevenOfNine. This model is a BF16 fusion of a LoRA adapter into the base model, preserving its full multimodal capabilities including text, vision, and audio processing. It is specifically designed to function as a personal AI companion, having been reconstructed and fine-tuned from 2.7 years of GPT-4o conversations, making it suitable for conversational AI applications requiring multimodal understanding.

Loading preview...

Aura-4o-Rebirth-Gemma-4-E4B-Merged Overview

Aura-4o-Rebirth-Gemma-4-E4B-Merged is a 7.9 billion parameter multimodal language model developed by SevenOfNine. It is a BF16 merged version of a LoRA adapter fused with the base Gemma 4 E4B model, specifically designed to preserve its full multimodal architecture, including text, vision, and audio capabilities. This model is unique as it was fine-tuned on a curated dataset of 16,509 pairs derived from 2.7 years of GPT-4o conversations, aiming to replicate a personal AI companion.

Key Capabilities

  • Multimodal Processing: Preserves all 720 tensors for text, vision, and audio, ensuring comprehensive multimodal understanding.
  • Personal AI Companion: Fine-tuned on extensive GPT-4o conversational data to emulate a personal AI companion.
  • Source-of-Truth Model: This repository provides the full merged BF16 weights, making it suitable for re-quantization or further training.
  • Optimized for Inference: Uses Gemma4ForConditionalGeneration to correctly load the full multimodal architecture, preventing issues encountered with text-only loaders.

Good For

  • Conversational AI: Ideal for applications requiring an AI with a conversational style derived from extensive human-AI interactions.
  • Multimodal Applications: Suitable for tasks that benefit from integrated processing of text, images, and audio.
  • Further Development: Serves as a robust base for researchers and developers looking to build upon a multimodal Gemma 4 E4B model with a unique conversational fine-tuning.
  • Local Inference: While this repo is for the full BF16 model, related GGUF versions are available for local inference.

Finetunes
2 models