README.md

---
language:
- en
library_name: mlx
base_model: Jackrong/Qwopus3.6-35B-A3B-v1
pipeline_tag: text-generation
license: apache-2.0
tags:
- mlx
- qwen3.6
- qwopus
- moe
- helios
- union-street-ai
- local-ai
- agentic
---

# Helios Rabbit 1.0

Helios Rabbit 1.0 is a fused MLX release of a lightweight Helios post-training pass on `Jackrong/Qwopus3.6-35B-A3B-v1`, adapted by Union Street AI.

Rabbit is intended as a local, Apple Silicon-friendly agent model with a Union Street identity layer, practical engineering behavior, and a bias toward directness, uncertainty calibration, and read-before-guessing workflows.

## Identity

Helios Rabbit should identify as Helios, a local AI model developed and adapted by Union Street AI. It should be honest that the base model comes from open model research rather than claiming Union Street AI trained the base weights from scratch.

## Release Notes

- Base: `Jackrong/Qwopus3.6-35B-A3B-v1`
- Adapter source: `Helios-Rabbit-impressive-v07-lora`
- Format: fused MLX safetensors
- Context target: 262k
- Intended serving: MLX or an OpenAI-compatible local router

This release is experimental. It is strongest when served with a system prompt that names the Helios identity and when the serving stack handles reasoning text separately from final visible output.

## Smoke Checks

Before upload, the fused local checkpoint answered:

```text
I'm Helios, a local model developed and adapted by Union Street AI.
```

It also passed a renderer/repetition trap prompt without repeating filler tokens or emitting raw renderer JSON.

## License

Released under Apache 2.0, subject to the base model's license and terms.