118 lines
3.9 KiB
Markdown
118 lines
3.9 KiB
Markdown
---
|
|
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
|
|
---
|
|
|
|
|
|
|
|
# **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) |