113 lines
4.0 KiB
Markdown
113 lines
4.0 KiB
Markdown
---
|
|
license: apache-2.0
|
|
language:
|
|
- en
|
|
base_model:
|
|
- prithivMLmods/Qwen3-0.6B-ft-bf16
|
|
pipeline_tag: text-generation
|
|
library_name: transformers
|
|
tags:
|
|
- text-generation-inference
|
|
- code
|
|
- moe
|
|
datasets:
|
|
- open-r1/Mixture-of-Thoughts
|
|
---
|
|
|
|
|
|
|
|
# **Theta-Crucis-0.6B-Turbo1**
|
|
|
|
> **Theta-Crucis-0.6B-Turbo1** is a compact, high-performance model designed for **code generation**, **technical reasoning**, and **structured output tasks**. Fine-tuned from **Qwen3-0.6B** using the **Mixture of Thoughts (MoT)** dataset with an emphasis on **code expert clusters**, this model delivers agile and accurate coding assistance in low-resource environments. At only **0.6B parameters**, it offers strong fluency in programming, structured syntax, and technical language generation.
|
|
|
|
> \[!note]
|
|
> GGUF: [https://huggingface.co/prithivMLmods/Theta-Crucis-0.6B-Turbo1-GGUF](https://huggingface.co/prithivMLmods/Theta-Crucis-0.6B-Turbo1-GGUF)
|
|
|
|
---
|
|
|
|
## **Key Features**
|
|
|
|
1. **MoT Fine-Tuning on Code Expert Clusters**
|
|
Leveraging the **Mixture of Thoughts (MoT)** dataset, this model is fine-tuned on high-quality programming data across languages, debugging patterns, and code reasoning structures.
|
|
|
|
2. **Turbo Code Generation & Debugging**
|
|
Excels at generating well-structured, clean code in Python, JavaScript, C++, and more. Capable of explaining logic, identifying bugs, and suggesting improvements.
|
|
|
|
3. **Structured Output Capabilities**
|
|
Supports outputs in **Markdown**, **JSON**, **YAML**, and **LaTeX**, making it ideal for auto-documentation, API formatting, and configuration file generation.
|
|
|
|
4. **Technical Fluency Across Languages**
|
|
Handles code queries and explanations in over **20 languages**, enabling global developer support and multilingual documentation.
|
|
|
|
5. **Lightweight, Inference-Optimized Design**
|
|
Suitable for deployment on **edge devices**, **laptops**, or **VRAM-limited GPUs**, with fast inference and strong accuracy in technical prompts.
|
|
|
|
---
|
|
|
|
## **Quickstart with Transformers**
|
|
|
|
```python
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
|
|
model_name = "prithivMLmods/Theta-Crucis-0.6B-Turbo1"
|
|
|
|
model = AutoModelForCausalLM.from_pretrained(
|
|
model_name,
|
|
torch_dtype="auto",
|
|
device_map="auto"
|
|
)
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
|
|
prompt = "Write a Python function that checks if a string is a palindrome. Explain each step."
|
|
|
|
messages = [
|
|
{"role": "system", "content": "You are an expert code assistant."},
|
|
{"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**
|
|
|
|
* Programming education, code synthesis, and debugging support
|
|
* Structured data and config file generation (e.g., JSON, YAML)
|
|
* Developer assistant roles in multilingual and technical environments
|
|
* Deployment on constrained devices with high code output needs
|
|
* Fast prototyping and script generation across languages
|
|
|
|
---
|
|
|
|
## **Limitations**
|
|
|
|
* May underperform in long conversational or abstract language tasks
|
|
* Context length limitations can restrict multi-file or large project reasoning
|
|
* Not designed for creative writing or open-ended dialogue
|
|
* Focuses on technical and structured domains—general fluency is limited
|
|
|
|
---
|
|
|
|
## **References**
|
|
|
|
1. [Qwen2.5 Technical Report (2024)](https://arxiv.org/pdf/2412.15115)
|
|
2. [YaRN: Efficient Context Window Extension of Large Language Models](https://arxiv.org/pdf/2309.00071)
|
|
3. [open-r1/Mixture-of-Thoughts](https://huggingface.co/datasets/open-r1/Mixture-of-Thoughts) |