---
license: llama3.2
datasets:
- glaiveai/reflection-v1
base_model:
- meta-llama/Llama-3.2-3B-Instruct
language:
- en
pipeline_tag: text2text-generation
---

# Llama-3.2-3B-Instruct Fine-tuned on glaiveai/reflection-v1

- **Developed by:** Meshwa
- **License:** apache-2.0
- **Finetuned from model :** unsloth/Llama-3.2-3B-Instruct

This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.

[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)



## Overview

* Contains **Llama-3.2-3B-Instruct**,
* Fine-tuned on the **glaiveai/reflection-v1** dataset using the **Unsloth** library.
* Model has been quantized into several formats (`q4`, `q5`, `q6`, `q8` and `f16`)
* Modelfile for use with Ollama is included, The default quantization is set to **Q8_0**, edit if you want to.


## Model Description

### Objective
Tried to finetune **Llama-3.2-3B-Instruct** leveraging the **glaiveai/reflection-v1** dataset. I thought it would be fun to see how smaller models perform on this task.

### Dataset: glaiveai/reflection-v1
The **glaiveai/reflection-v1** dataset is tailored for reflective, introspective tasks, including open-ended conversation, abstract reasoning, and context-aware response generation. This dataset includes tasks such as:
- Thoughtful question answering
- Summarization of complex ideas
- Reflective problem solving

### Fine-tuning Methodology: Unsloth Library
**Unsloth** was used for 2x faster finetuing of the base Llama-3.2 model.

## Usage

### Inference with gguf Quantized Models
To use the model in gguf format, load your preferred quantized version with a compatible inference framework such as `llama.cpp` or any gguf-supported libraries:

```python
from llama_cpp import Llama

llama_model = Llama(model_path="path_to_model/Llama-3.2-3B-Instruct-q8_0.gguf")
result = llama_model("Your instruction prompt here")
print(result)
```

### Using with Ollama
The included Modelfile supports direct loading in Ollama. To use the default model, simply run:

```bash
ollama create "model_name_here" -f "Modelfile_path"
```

Directly importing from HF 🤗
```bash
ollama pull hf.co/Meshwa/llama3.2-3b-Reflection-v1:{quant_type}
```

make sure to replace `{quant_type}` with one of these:
  - `Q4_K_M`
  - `Q4_0`
  - `Q4_1`
  - `Q6_K`
  - `Q8_0` (default in my modelfile)
  - `Q5_K_M`
  - `F16`


For Better results use the below system prompt:
```bash
You are a world-class AI system capable of complex reasoning and reflection. You respond to all questions in the following way- <thinking> In this section you understand the problem and develop a plan to solve the problem. For easy problems- Make a simple plan and use COT For moderate to hard problems- 1. Devise a step-by-step plan to solve the problem. (don't actually start solving yet, just make a plan) 2. Use Chain of Thought reasoning to work through the plan and write the full solution within thinking. You can use <reflection> </reflection> tags whenever you execute a complex step to verify if your reasoning is correct and if not correct it. </thinking> <output> In this section, provide the complete answer for the user based on your thinking process. Do not refer to the thinking tag. Include all relevant information and keep the response somewhat verbose, the user will not see what is in the thinking tag. </output>
```

## License

This model is released under the **Apache 2.0**.

## Citation

If you use this model, please cite the following:

```
@article{Llama-3.2-3B-Instruct-Reflection-v1,
  author = {Meshwa},
  title = {Llama-3.2-3B-Instruct Fine-tuned on glaiveai/reflection-v1},
  year = {2024},
  published = {https://huggingface.co./Meshwa/llama3.2-3b-Reflection-v1}
}
```