---
license: apache-2.0
datasets:
- open-r1/Mixture-of-Thoughts
language:
- en
base_model:
- prithivMLmods/Qwen3-0.6B-ft-bf16
pipeline_tag: text-generation
library_name: transformers
tags:
- text-generation-inference
- code
- science
- math
- moe
---

![12.png](/static-proxy?url=https%3A%2F%2Fcdn-uploads.huggingface.co%2Fproduction%2Fuploads%2F65bb837dbfb878f46c77de4c%2FFAnKrTFpUHZbu9J0fLKCL.png)

# **Eta-Aurigae-0.6B-Echelon1**

> **Eta-Aurigae-0.6B-Echelon1** is a compact, efficient model specialized in **science, factual accuracy**, and **structured reasoning**. Fine-tuned on **Qwen3-0.6B** using the **MoT (Mixture of Thoughts)** dataset—focused on scientific understanding and expert factual domains—it delivers high-precision outputs for STEM education, tutoring, and analytical thinking in resource-constrained environments.

> \[!note]
> GGUF: [https://huggingface.co/prithivMLmods/Eta-Aurigae-0.6B-Echelon1-GGUF](https://huggingface.co/prithivMLmods/Eta-Aurigae-0.6B-Echelon1-GGUF)

---

## **Key Features**

1. **MoT Fine-Tuning for Science & Facts**
   Trained on a **Mixture of Thoughts** dataset emphasizing scientific accuracy, explanatory depth, and structured reasoning across biology, physics, chemistry, and factual domains.

2. **Scientific Precision in a Small Footprint**
   Delivers clear, step-by-step reasoning in scientific problems—ideal for students, educators, and lightweight educational tools.

3. **Factually Consistent Output Generation**
   Optimized for **high factual alignment** and structured explanations—reliable for knowledge recall, concept breakdowns, and factual analysis.

4. **Supports Markdown, LaTeX, and JSON**
   Outputs clean, structured formats like **Markdown**, **LaTeX**, and **JSON**, useful for technical documentation and educational content.

5. **Multilingual Science-Aware Responses**
   Handles factual content in 20+ languages, especially in academic and technical contexts.

6. **Lightweight and Inference-Ready**
   Efficient on **CPUs**, **low-VRAM GPUs**, and **offline edge deployments** without sacrificing factual clarity.

---

## **Quickstart with Transformers**

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "prithivMLmods/Eta-Aurigae-0.6B-Echelon1"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

prompt = "What causes the northern lights (Aurora Borealis)? Explain in simple terms."

messages = [
    {"role": "system", "content": "You are a science tutor that explains complex concepts clearly."},
    {"role": "user", "content": prompt}
]

text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)

model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=512
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)
```

---

## **Intended Use**

* Science education and fact-based tutoring
* Concept explanations in physics, biology, and chemistry
* Structured technical content generation (e.g., LaTeX, Markdown)
* Deployment in low-resource, educational, or mobile scenarios
* Lightweight inference with high factual fidelity

---

## **Limitations**

* Not optimized for general conversation or creative writing
* Short context limits multi-document scientific reasoning
* Performance dips in abstract reasoning outside scientific scope
* Not tuned for code or free-form generation

---

## **References**

1. [Qwen2.5 Technical Report (2024)](https://arxiv.org/pdf/2412.15115)
2. [Mixture of Thoughts Dataset](https://huggingface.co/datasets/open-r1/Mixture-of-Thoughts)
3. [YaRN: Efficient Context Extension for LLMs](https://arxiv.org/pdf/2309.00071)