---
license: apache-2.0
language:
- en
base_model:
- Qwen/Qwen2.5-1.5B-Instruct
pipeline_tag: text-generation
library_name: transformers
tags:
- text-generation-inference
- math
- trl
- SFT
---

![89.png](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/vxacZQLTe3BR8F7Np_CQU.png)

# **Ross-640-BMath-1.5B**

> **Ross-640-BMath-1.5B** is an **experimental, high-precision math explanation model** fine-tuned on **Qwen2-1.5B**, designed to provide **step-by-step mathematical derivations** and **detailed concept explanations** across a wide range of mathematical domains. It is **not optimized for general reasoning or conversation**, and focuses primarily on **structured, non-reasoning math workflows** including algebra, calculus, number theory, and combinatorics.

> \[!note]
> GGUF: [https://huggingface.co/prithivMLmods/Ross-640-BMath-1.5B-GGUF](https://huggingface.co/prithivMLmods/Ross-640-BMath-1.5B-GGUF)

---

## **Key Features**

1. **Hard Math Concept Focus**
   Specializes in **algebra**, **calculus**, **combinatorics**, **linear algebra**, **number theory**, and more—delivering fine-tuned, low-latency outputs ideal for **math-intensive applications**.

2. **Step-by-Step Explanations**
   Emphasizes **procedural clarity** over abstract reasoning, offering structured, educational breakdowns of mathematical problems and derivations.

3. **Symbolic Computation & Annotation**
   Outputs include LaTeX-compatible syntax, inline math symbols, and clear annotation to support academic and technical workflows.

4. **Educational Utility**
   Optimized for **learning and teaching**, providing clear responses to mathematical queries with minimal noise or conversational drift.

5. **Lightweight Architecture**
   Built on Qwen2-1.5B and fine-tuned for **efficiency and precision**, making it suitable for deployment in **resource-constrained environments**, educational tools, or math-centric chat interfaces.

---

## **Quickstart with Transformers**

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "prithivMLmods/Ross-640-BMath-1.5B"

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

prompt = "Explain step-by-step how to integrate (x^2 + 1)/(x^3 + 3x) dx."

messages = [
    {"role": "system", "content": "You are a helpful assistant skilled in solving complex math problems with clear and structured steps."},
    {"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**

* Detailed mathematical explanations and problem-solving
* Education-focused tutoring and math derivation tools
* Math-focused applications and formula documentation
* Symbolic derivations and LaTeX generation
* Integration with learning platforms and academic software

---

## **Limitations**

* Not suitable for general-purpose conversation or reasoning tasks
* Context length constraints may limit effectiveness on large proofs
* May struggle with non-mathematical or open-ended creative tasks
* Experimental: Fine-tuned primarily for **explanation clarity**, not deep symbolic reasoning or formal proof validation