Omega-Qwen3-Atom-8B/README.md

---
license: apache-2.0
base_model:
- Qwen/Qwen3-8B
datasets:
- prithivMLmods/Open-Omega-Atom-1.5M
language:
- en
pipeline_tag: text-generation
library_name: transformers
tags:
- text-generation-inference
- code
- science
- 'Thinking: Enabled'
- math
- mot
- moe
- stem
---

![11.png](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/V26CJSyLm0ixHwNZQLlc_.png)

# **Omega-Qwen3-Atom-8B**

> **Omega-Qwen3-Atom-8B** is a powerful 8B-parameter model fine-tuned on **Qwen3-8B** using the curated **Open-Omega-Atom-1.5M** dataset, optimized for **math and science reasoning**. It excels at symbolic processing, scientific problem-solving, and structured output generation—making it a high-performance model for researchers, educators, and technical developers working in computational and analytical domains.

## **Key Features**

1. **Math & Science-Centric Reasoning**
   Fine-tuned on the **Open-Omega-Atom-1.5M** dataset, built from high-quality math, science, and symbolic reasoning tasks—ideal for analytical domains including algebra, calculus, physics, and chemistry.

2. **Scientific Concept Breakdown**
   Explains theories, derivations, and concepts across STEM fields with clarity—solves equations step-by-step, handles formula-based questions, and provides interpretive insights.

3. **Symbolic Computation & Chain-of-Thought**
   Supports multi-step reasoning, symbolic derivations, and proof-based problem solving with a strong focus on accuracy and transparency.

4. **Structured Output Generation**
   Outputs precise formats in **LaTeX**, **Markdown**, **JSON**, and **YAML** for scientific writing, educational materials, and data pipeline integration.

5. **Optimized for Efficient Scientific Workflows**
   While based on an 8B model, it is optimized for **offline inference**, **research clusters**, and **GPU workstations** that need high symbolic precision and performance.

---

## Quick Start with Hugging Face Transformers🤗

```py
!pip install transformers huggingface_hub accelerate
```

```py
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "prithivMLmods/Omega-Qwen3-Atom-8B"

# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)

# prepare the model input
prompt = "A alone can do a piece of work in 6 days and B alone in 8 days. A and B undertook to do it for Rs. 3200. With the help of C, they completed the work in 3 days. How much is to be paid to C?"
messages = [
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=True # Switches between thinking and non-thinking modes. Default is True.
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# conduct text completion
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=32768
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist() 

# parsing thinking content
try:
    # rindex finding 151668 (</think>)
    index = len(output_ids) - output_ids[::-1].index(151668)
except ValueError:
    index = 0

thinking_content = tokenizer.decode(output_ids[:index], skip_special_tokens=True).strip("\n")
content = tokenizer.decode(output_ids[index:], skip_special_tokens=True).strip("\n")

print("thinking content:", thinking_content)
print("content:", content)
```

## Answer

```
thinking content: <think>
Okay, let's see. So the problem is about A, B, and C working together to complete a piece of work. We need to figure out how much money C should get for his help. The total payment is Rs. 3200, and we have to divide that among A, B, and C based on their contributions. Let me try to break this down step by step.

First, let's understand the work rates of A and B. A can finish the work in 6 days, so his work rate is 1/6 of the work per day. Similarly, B can finish it in 8 days, so his work rate is 1/8 per day. When they work together, their combined work rate would be 1/6 + 1/8. Let me calculate that:

1/6 + 1/8. To add these, find a common denominator, which is 24. So, 4/24 + 3/24 = 7/24. So together, A and B can do 7/24 of the work in one day.

But the problem says that with the help of C, they completed the work in 3 days. That means all three working together finished the job in 3 days. Let's denote C's work rate as 1/x per day, where x is the number of days C would take alone. So, the combined work rate of A, B, and C is 1/6 + 1/8 + 1/x.

Since they completed the work in 3 days, their combined work rate multiplied by 3 should equal 1 (the whole work). So:

(1/6 + 1/8 + 1/x) * 3 = 1

Let me solve for 1/x first. Let's compute 1/6 + 1/8:

As before, 1/6 is 4/24 and 1/8 is 3/24, so together they are 7/24. So:

(7/24 + 1/x) * 3 = 1

Divide both sides by 3:

7/24 + 1/x = 1/3

Subtract 7/24 from both sides:
...

$$
\boxed{400}
$$
```

---

## **Intended Use**

* Math and science tutoring, equation solving, and symbolic reasoning
* Educational tools for high-school to postgraduate-level STEM
* Research-grade assistant for physics, chemistry, and applied math
* Structured technical content generator for papers, lab work, and datasets
* STEM-focused chatbot/API for integration into science education platforms

## **Limitations**

* Not trained for open-domain chat or emotional dialogue
* May struggle with very large codebases or long multi-part tasks
* Best suited for STEM fields—general language understanding may vary
* Prioritizes correctness and formality over conversational tone.