---
base_model: Qwen/Qwen2.5-1.5B-Instruct
library_name: peft
pipeline_tag: text-generation
language:
- id
- en
tags:
- base_model:adapter:Qwen/Qwen2.5-1.5B-Instruct
- lora
- peft
- sft
- transformers
- trl
- unsloth
- indonesian
- chatbot
- reasoning
license: apache-2.0
---

# Hazel-1o v1

**Hazel-1o** is an Indonesian chat model fine-tuned from **Qwen2.5-1.5B-Instruct** using **LoRA (Low-Rank Adaptation)**. It is designed as a friendly AI assistant named Hazel-1o with structured reasoning via a custom `<thinking>` tag.

---

## Model Details

### Model Description

- **Developed by:** amnottdevv
- **Model type:** Causal Language Model (LoRA adapter)
- **Base model:** [Qwen/Qwen2.5-1.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct)
- **Language(s) (NLP):** Bahasa Indonesia (primary), English (secondary)
- **License:** Apache 2.0
- **Finetuned from:** Qwen2.5-1.5B-Instruct

### Model Sources

- **Repository:** https://huggingface.co/amnottdevv/Hazel-1o-v1
- **Base model:** https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct

---

## Uses

### Direct Use

This model is suitable for:

- **Indonesian persona chatbot** — casual conversation as the "Hazel-1o" persona
- **Structured reasoning** — basic math problems using the `<thinking>` tag
- **Basic C++ & Python coding** — simple code generation with Indonesian explanations
- **Indonesian FAQ** — answering general questions in a friendly tone

### Out-of-Scope Use

This model is **NOT** suitable for:

- ❌ **Production use** — outputs require human verification
- ❌ **Illegal use** — 

---

## Bias, Risks, and Limitations

### Known Limitations

1. **Factual hallucination** — The model can produce incorrect facts with a convincing format. Always verify factual outputs.

2. **Persona inconsistency** — Occasionally generates random names (e.g., Xaeros, Aida) for meta questions. Common issue in 1.5B models.

3. **Context contamination** — Topics from previous turns may leak into subsequent answers. Reset history between distinct topics.

4. **Math (4+ digits)** — Low accuracy for large-number arithmetic.

5. **Code debugging** — Unable to analyze code errors accurately.

### Recommendations

- Always **verify outputs** for factual topics
- Use **temperature 0.4–0.5** for the most stable results
- **Reset history** before switching to unrelated topics
- **Do not use in production** without human supervision

---

## How to Get Started

### Installation

```bash
pip install transformers peft accelerate torch