93 lines
3.5 KiB
Markdown
93 lines
3.5 KiB
Markdown
---
|
||
library_name: transformers
|
||
license: apache-2.0
|
||
language:
|
||
- en
|
||
- zh
|
||
base_model:
|
||
- Qwen/Qwen2.5-1.5B-Instruct
|
||
pipeline_tag: text-generation
|
||
tags:
|
||
- text-generation-inference
|
||
- Code
|
||
- Math
|
||
- RL
|
||
- CoT
|
||
---
|
||
|
||
|
||
# **Monoceros-QwenM-1.5B**
|
||
|
||
> **Monoceros-QwenM-1.5B** is a **chain-of-thought reasoning model** fine-tuned from **Qwen-1.5B**, specifically designed for solving **mathematical problems** in both **English** and **Chinese**. It brings advanced reasoning and step-by-step problem-solving capabilities in a compact size, ideal for educational tools, tutoring systems, and math-focused assistants.
|
||
|
||
## **Key Features**
|
||
|
||
1. **Chain-of-Thought Math Reasoning**
|
||
Trained to produce intermediate steps for math problems, Monoceros-QwenM-1.5B enables interpretability and transparent logic in answers — critical for educational and verification purposes.
|
||
|
||
2. **Bilingual Proficiency (English + Chinese)**
|
||
Capable of understanding, reasoning, and explaining math problems fluently in **both English and Simplified Chinese**, making it suitable for multilingual learning environments.
|
||
|
||
3. **Compact yet Capable**
|
||
While only 1.5B parameters, this model delivers strong performance for arithmetic, algebra, geometry, word problems, and logical puzzles with minimal resource demands.
|
||
|
||
4. **Step-by-Step Computation**
|
||
Provides structured, multi-step answers that mirror human-like problem solving, making it easy to follow and learn from.
|
||
|
||
## **Quickstart with Transformers**
|
||
|
||
```python
|
||
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||
|
||
model_name = "prithivMLmods/Monoceros-QwenM-1.5B"
|
||
|
||
model = AutoModelForCausalLM.from_pretrained(
|
||
model_name,
|
||
torch_dtype="auto",
|
||
device_map="auto"
|
||
)
|
||
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
||
|
||
prompt = "Solve: A train travels 180 km in 3 hours. What is its average speed?"
|
||
messages = [
|
||
{"role": "system", "content": "You are a helpful tutor skilled in solving math problems with step-by-step explanations."},
|
||
{"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**
|
||
|
||
- **Math Tutoring Bots**: Step-by-step solvers for students across basic to intermediate levels.
|
||
- **Bilingual Educational Apps**: Teaching math in **English** and **Chinese**, improving accessibility.
|
||
- **STEM Reasoning Tools**: Reasoning for science, engineering, and logic-based problems.
|
||
- **Lightweight LLM Applications**: Embedded use cases in browsers, mobile apps, or low-resource environments.
|
||
|
||
## **Limitations**
|
||
|
||
1. **Limited Domain Generalization**:
|
||
Optimized for math; performance may drop in creative writing, casual conversation, or unrelated topics.
|
||
|
||
2. **Parameter Scale**:
|
||
Though efficient, it may underperform compared to larger models on highly complex or abstract math.
|
||
|
||
3. **Bias from Base Model**:
|
||
Inherits any biases from Qwen-1.5B’s pretraining. Outputs should be validated in sensitive settings.
|
||
|
||
4. **Prompt Sensitivity**:
|
||
Precise, structured input yields better stepwise results. |