103 lines
3.7 KiB
Markdown
103 lines
3.7 KiB
Markdown
---
|
|
library_name: transformers
|
|
license: apache-2.0
|
|
language:
|
|
- en
|
|
base_model:
|
|
- Qwen/Qwen2.5-1.5B-Instruct
|
|
pipeline_tag: text-generation
|
|
tags:
|
|
- text-generation-inference
|
|
- code
|
|
- math
|
|
- RL
|
|
---
|
|
|
|
|
|
|
|
# **Horologium-QwenC-1.5B**
|
|
|
|
> **Horologium-QwenC-1.5B** is a **reasoning-focused language model** trained extensively on both **coding** and **mathematics** problems using **reinforcement learning (RL)**. It is designed to provide intelligent, step-by-step solutions to structured tasks that require logical precision, algorithmic thought, and symbolic computation.
|
|
|
|
## **Key Features**
|
|
|
|
1. **Unified Reasoning for Code & Math**
|
|
Tailored to perform both **code understanding/generation** and **mathematical problem-solving**, with a consistent focus on clarity and logic.
|
|
|
|
2. **Reinforcement Learning Fine-Tuning**
|
|
Trained with **reinforcement learning** to improve reward-aligned behaviors in complex problem-solving scenarios—especially in debugging, proof validation, and computational tasks.
|
|
|
|
3. **Symbolic and Numerical Proficiency**
|
|
Capable of handling **symbolic math**, **algebra**, **calculus**, and **discrete mathematics**, while also excelling at code logic, syntax validation, and API usage.
|
|
|
|
4. **Compact yet Powerful**
|
|
At **1.5B parameters**, this model provides **strong reasoning capabilities** while remaining efficient for edge devices and local deployment.
|
|
|
|
5. **Structured Output**
|
|
Produces high-quality, structured results in Markdown, JSON, and annotated code blocks with contextual explanations.
|
|
|
|
## **Quickstart with Transformers**
|
|
|
|
```python
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
|
|
model_name = "prithivMLmods/Horologium-QwenC-1.5B"
|
|
|
|
model = AutoModelForCausalLM.from_pretrained(
|
|
model_name,
|
|
torch_dtype="auto",
|
|
device_map="auto"
|
|
)
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
|
|
prompt = "Solve this: A function f is defined as f(x) = x^2 + 2x + 1. Find f(5) and explain the steps. Then write equivalent Python code."
|
|
|
|
messages = [
|
|
{"role": "system", "content": "You are an expert in math and coding. Solve problems step-by-step and explain clearly."},
|
|
{"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]
|
|
```
|
|
|
|
## **Intended Use**
|
|
|
|
- **Educational Tutoring Systems**
|
|
For students and learners exploring both programming and mathematics.
|
|
|
|
- **Coding & Algorithmic Interview Prep**
|
|
Useful for solving DSA questions, algorithmic challenges, and leetcode-style problems.
|
|
|
|
- **Math & Code Co-Pilots**
|
|
Integrated into coding environments to explain both logic and formulas used in implementations.
|
|
|
|
- **Data Analysis & Scientific Computing**
|
|
Aids in writing and verifying data-centric scripts and computational logic.
|
|
|
|
## **Limitations**
|
|
|
|
1. **Scope of Accuracy**
|
|
May occasionally produce mathematically sound but **over-explained or verbose** solutions.
|
|
|
|
2. **Complex Multistep Problems**
|
|
Performance may degrade slightly on **very long multi-turn** symbolic derivations or nested algorithms.
|
|
|
|
3. **Limited Real-Time Adaptation**
|
|
No awareness of real-time data or updates beyond training scope.
|
|
|
|
4. **Security & Logic Bugs**
|
|
Always audit generated code or logic for real-world use. |