|
|
--- |
|
|
license: apache-2.0 |
|
|
language: |
|
|
- en |
|
|
base_model: |
|
|
- Qwen/Qwen3-0.6B |
|
|
pipeline_tag: text-generation |
|
|
library_name: transformers |
|
|
tags: |
|
|
- text-generation-inference |
|
|
- code |
|
|
- general-reasoning |
|
|
- moe |
|
|
- math |
|
|
--- |
|
|
|
|
|
 |
|
|
|
|
|
# **Lynx-TinySync-0.6B** |
|
|
|
|
|
> **Lynx-TinySync-0.6B** is a lightweight, high-performance model designed for **mathematical reasoning**, **code generation**, and **general-purpose inference**. Built on a custom modular dataset and powered by an efficient architecture, it excels in delivering structured, accurate outputs even in mid-resource environments. Despite its compact **0.6B parameter size**, it demonstrates remarkable proficiency in math, code, and technical language understanding. |
|
|
|
|
|
> \[!note] |
|
|
> GGUF: [https://huggingface.co/prithivMLmods/Lynx-TinySync-0.6B-GGUF](https://huggingface.co/prithivMLmods/Lynx-TinySync-0.6B-GGUF) |
|
|
|
|
|
--- |
|
|
|
|
|
## **Key Features** |
|
|
|
|
|
1. **Custom Modular Dataset Training** |
|
|
Fine-tuned using a handcrafted blend of math, code, and reasoning datasets, ensuring high performance in symbolic tasks and general queries. |
|
|
|
|
|
2. **Mathematical Reasoning** |
|
|
Handles algebra, calculus, geometry, and symbolic logic with clarity—ideal for tutoring, educational support, and math competitions. |
|
|
|
|
|
3. **Compact Code Assistant** |
|
|
Generates clean, efficient code in Python, JavaScript, and more—complete with explanations and bug-fix breakdowns. |
|
|
|
|
|
4. **Structured Output Generation** |
|
|
Outputs in JSON, Markdown, LaTeX, and tabular formats—well-suited for documentation, structured data templates, and technical content. |
|
|
|
|
|
5. **Multilingual Technical Reasoning** |
|
|
Supports math and code queries in 20+ languages with consistent output—enabling multilingual academic and professional use cases. |
|
|
|
|
|
6. **Optimized for Low-Resource Deployment** |
|
|
With only 0.6B parameters, it's ideal for inference on edge devices, local machines, and GPU-constrained environments. |
|
|
|
|
|
--- |
|
|
|
|
|
## **Quickstart with Transformers** |
|
|
|
|
|
```python |
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer |
|
|
|
|
|
model_name = "prithivMLmods/Lynx-TinySync-0.6B" |
|
|
|
|
|
model = AutoModelForCausalLM.from_pretrained( |
|
|
model_name, |
|
|
torch_dtype="auto", |
|
|
device_map="auto" |
|
|
) |
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name) |
|
|
|
|
|
prompt = "Solve the equation: 2(x - 4) + 3 = 11. Show all steps." |
|
|
|
|
|
messages = [ |
|
|
{"role": "system", "content": "You are a step-by-step math tutor."}, |
|
|
{"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** |
|
|
|
|
|
* Mathematical problem solving and symbolic logic |
|
|
* Lightweight code generation and debugging |
|
|
* Generation of structured content (e.g., JSON, LaTeX, Markdown) |
|
|
* Educational support across languages and domains |
|
|
* Low-resource deployment in academic or field settings |
|
|
|
|
|
--- |
|
|
|
|
|
## **Limitations** |
|
|
|
|
|
* May underperform on long-form creative generation tasks |
|
|
* Smaller context window may limit deep multi-turn reasoning |
|
|
* Less capable in adversarial or abstract reasoning queries |
|
|
* Technical multilingual use focused—general dialogue fluency limited |
|
|
|
|
|
--- |
|
|
|
|
|
## **References** |
|
|
|
|
|
1. [Qwen2.5 Technical Report](https://arxiv.org/pdf/2412.15115) |
|
|
2. [YaRN: Efficient Context Window Extension of Large Language Models](https://arxiv.org/pdf/2309.00071) |
