README.md

---
license: apache-2.0
language:
- en
base_model:
- Qwen/Qwen3-1.7B
pipeline_tag: text-generation
library_name: transformers
tags:
- text-generation-inference
- code
- trl
---

![1.png](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/4LKArEzZk53evXdv_no2u.png)

# **Pyxidis-Manim-CodeGen-1.7B (Experimental)**

> **Pyxidis-Manim-CodeGen-1.7B** is an **experimental math animation coding model** fine-tuned on **Qwen/Qwen3-1.7B** using **Manim-CodeGen code traces**.
> It is specialized for **Python-based mathematical animations with Manim**, making it ideal for educators, researchers, and developers working on math visualization and animation pipelines.

> \[!note]
> GGUF: [https://huggingface.co/prithivMLmods/Pyxidis-Manim-CodeGen-1.7B-GGUF](https://huggingface.co/prithivMLmods/Pyxidis-Manim-CodeGen-1.7B-GGUF)

---

## **Key Features**

1. **Manim-Specific Code Generation**
   Trained on **Manim-CodeGen traces**, optimized for **Python-based animation scripting** of mathematical concepts and visual proofs.

2. **Math + Code Synergy**
   Generates step-by-step **math derivations with corresponding animation code**, bridging symbolic reasoning with visualization.

3. **Animation Workflow Optimization**
   Provides structured code for **scenes, transformations, graphs, and equations** in Manim, reducing boilerplate and debugging effort.

4. **Python-Centric Reasoning**
   Produces **clean, modular, and reusable Python code**, supporting educational and research-driven animation pipelines.

5. **Structured Output Mastery**
   Capable of outputting in **Python**, **Markdown**, and **LaTeX**, ideal for tutorials, educational notebooks, and automated video generation workflows.

6. **Lightweight but Specialized**
   Focused on **Manim coding efficiency** while maintaining a deployable footprint for **GPU clusters** and **research labs**.

---

## **Quickstart with Transformers**

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "prithivMLmods/Pyxidis-Manim-CodeGen-1.7B"

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

prompt = "Write a Manim script to animate the Pythagorean theorem using squares on the triangle's sides."

messages = [
    {"role": "system", "content": "You are a Python coding assistant specialized in Manim-based math animations."},
    {"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**

* **Manim-based math animation coding** for research, teaching, and content creation
* **Educational visualization assistant** to convert math problems into animations
* **Python tutoring tool** for math-heavy animation workflows
* **Prototype generator** for interactive STEM video content

## **Limitations**

* Experimental model – may generate code requiring manual debugging
* Limited to **Manim coding workflows**, not general-purpose code assistant
* May not handle **complex multi-scene projects** without iterative refinement
* Prioritizes structured math + animation reasoning, less optimized for general dialogue