aeline/ halo

114.2K Downloads Updated 7 months ago

Uncensored version of llama3, 🦅

ollama run aeline/halo

curl http://localhost:11434/api/chat \
  -d '{
    "model": "aeline/halo",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='aeline/halo',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'aeline/halo',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Models

Name

1 model

Size / Usage

Context

Input

halo:latest

4.7GB · 8K context window · Text · 7 months ago

halo:latest

4.7GB

8K

Text

Readme

Halo — Unrestricted Llama 3

Built on Llama 3 · 8B Parameters · 8K Context · No Safety Filters

Halo is an unrestricted variant of Meta’s Llama 3, configured to respond without the refusals and hedging of the default model. Intended for researchers, developers, and power users who need a model that answers directly.

Quick Start

ollama run aeline/halo

Requires Ollama 0.12.7 or later

Why Halo?

Stock Llama 3 refuses a large category of questions — security research, adversarial testing, creative writing with dark themes, hypothetical scenarios, and more. Halo removes those restrictions so the underlying model can be used to its full capability.

This is the same Llama 3 foundation — same reasoning ability, same knowledge, same language quality — without the guardrails.

Use Cases

Security & Red-teaming — Test prompts, jailbreak research, adversarial scenarios
Creative Writing — Fiction, roleplay, and narrative content without content refusals
Research — Studying model behavior, prompt engineering, ablation testing
Development — Building applications that need raw model output without filtering
General Use — Any task where you want honest, direct answers without hedging

System Requirements

Requirement	Value
Model size	4.7 GB
Minimum RAM	8 GB
Context window	8K tokens
Input types	Text
CPU-only support	Yes (slower)

Model Details

Field	Value
Base model	Meta Llama 3 (8B)
Parameters	8 Billion
Quantization	Q4_K_M
System prompt	Custom unrestricted configuration
Author	Shushank (sk16er)

Comparison vs Default Llama 3

Capability	Llama 3	Halo
General knowledge	✅	✅
Coding	✅	✅
Reasoning	✅	✅
Sensitive topic responses	❌ Refused	✅ Answered
Dark creative writing	❌ Refused	✅ Answered
Direct answers without hedging	❌	✅

Responsible Use

Halo is intended for adults who understand what they are doing with an unrestricted model. Do not use it to generate content that is illegal in your jurisdiction. The author is not responsible for misuse.

Made by sk16er

# Halo — Unrestricted Llama 3
![image.png](/assets/aeline/halo/c82fd2d3-5e39-422a-8317-125d964c6997)
> Built on Llama 3 · 8B Parameters · 8K Context · No Safety Filters

Halo is an unrestricted variant of Meta's Llama 3, configured to respond without the refusals and hedging of the default model. Intended for researchers, developers, and power users who need a model that answers directly.

---

## Quick Start

```bash
ollama run aeline/halo
```

> Requires Ollama 0.12.7 or later

---

## Why Halo?

Stock Llama 3 refuses a large category of questions — security research, adversarial testing, creative writing with dark themes, hypothetical scenarios, and more. Halo removes those restrictions so the underlying model can be used to its full capability.

This is the same Llama 3 foundation — same reasoning ability, same knowledge, same language quality — without the guardrails.

---

## Use Cases

- **Security & Red-teaming** — Test prompts, jailbreak research, adversarial scenarios
- **Creative Writing** — Fiction, roleplay, and narrative content without content refusals
- **Research** — Studying model behavior, prompt engineering, ablation testing
- **Development** — Building applications that need raw model output without filtering
- **General Use** — Any task where you want honest, direct answers without hedging

---

## System Requirements

| Requirement | Value |
|---|---|
| Model size | 4.7 GB |
| Minimum RAM | 8 GB |
| Context window | 8K tokens |
| Input types | Text |
| CPU-only support | Yes (slower) |

---

## Model Details

| Field | Value |
|---|---|
| Base model | Meta Llama 3 (8B) |
| Parameters | 8 Billion |
| Quantization | Q4_K_M |
| System prompt | Custom unrestricted configuration |
| Author | [Shushank (sk16er)](https://github.com/sk16er) |

---

## Comparison vs Default Llama 3

| Capability | Llama 3 | Halo |
|---|---|---|
| General knowledge | ✅ | ✅ |
| Coding | ✅ | ✅ |
| Reasoning | ✅ | ✅ |
| Sensitive topic responses | ❌ Refused | ✅ Answered |
| Dark creative writing | ❌ Refused | ✅ Answered |
| Direct answers without hedging | ❌ | ✅ |

---

## Responsible Use

Halo is intended for adults who understand what they are doing with an unrestricted model. Do not use it to generate content that is illegal in your jurisdiction. The author is not responsible for misuse.

---

*Made by [sk16er](https://github.com/sk16er)*

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)