|
--- |
|
license: llama3.1 |
|
language: |
|
- en |
|
base_model: |
|
- prithivMLmods/LwQ-Reasoner-10B |
|
pipeline_tag: text-generation |
|
library_name: transformers |
|
tags: |
|
- python |
|
- math |
|
- coder |
|
- reasoner |
|
- LCoT |
|
- Llama |
|
datasets: |
|
- prithivMLmods/PyThagoreans-Merged |
|
--- |
|
|
|
 |
|
|
|
# **PyThagorean-10B** |
|
|
|
PyThagorean [Python + Math] is a Python and mathematics-based model designed to solve mathematical problems using Python libraries and coding. It has been fine-tuned on 1.5 million entries and is built on LLaMA's architecture. The model supports different parameter sizes, including 10B, 3B, and 1B (Tiny). These instruction-tuned, text-only models are optimized for multilingual dialogue use cases, including agent-based retrieval and summarization tasks. PyThagorean leverages an auto-regressive language model that uses an optimized transformer architecture. The tuned versions employ supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety. |
|
|
|
|
|
# **Use with transformers** |
|
|
|
Starting with `transformers >= 4.43.0` onward, you can run conversational inference using the Transformers `pipeline` abstraction or by leveraging the Auto classes with the `generate()` function. |
|
|
|
Make sure to update your transformers installation via `pip install --upgrade transformers`. |
|
|
|
```python |
|
import transformers |
|
import torch |
|
|
|
model_id = "prithivMLmods/PyThagorean-10B" |
|
|
|
pipeline = transformers.pipeline( |
|
"text-generation", |
|
model=model_id, |
|
model_kwargs={"torch_dtype": torch.bfloat16}, |
|
device_map="auto", |
|
) |
|
|
|
messages = [ |
|
{"role": "system", "content": "You are the helpful assistant. Solve the mathematical problem in Python programming."}, |
|
{"role": "user", "content": "Find all real numbers $x$ such that \[\frac{x^3+2x^2}{x^2+3x+2} + x = -6.\]Enter all the solutions, separated by commas."}, |
|
] |
|
|
|
outputs = pipeline( |
|
messages, |
|
max_new_tokens=256, |
|
) |
|
print(outputs[0]["generated_text"][-1]) |
|
``` |
|
|
|
# **Intended Use** |
|
|
|
1. **Mathematical Problem Solving**: |
|
PyThagorean is designed for solving complex mathematical problems, including algebra, calculus, trigonometry, and more, by leveraging Python-based libraries. It is ideal for educational tools, tutoring platforms, and automated math assistants. |
|
|
|
2. **Python Code Generation**: |
|
The model generates Python code snippets for mathematical computations, simulations, and problem-solving, making it valuable for developers, researchers, and students. |
|
|
|
3. **Multilingual Dialogue Systems**: |
|
With support for multiple languages, PyThagorean can assist users worldwide in understanding and solving mathematical problems through dialogue-based interfaces. |
|
|
|
4. **Instruction-Following Tasks**: |
|
The model excels at adhering to precise mathematical instructions and delivering accurate, step-by-step solutions for problems embedded in text. |
|
|
|
5. **Agent-Based Knowledge Retrieval**: |
|
PyThagorean can retrieve and summarize mathematical concepts or problem-solving techniques, enabling quick access to relevant knowledge for educational and research purposes. |
|
|
|
6. **Educational Content Creation**: |
|
It generates educational content such as example problems, solutions, and Python-based tutorials, aiding teachers and content creators. |
|
|
|
7. **Summarization and Explanation**: |
|
The model provides clear explanations and breakdowns of mathematical solutions, helping users understand the rationale and process behind the answers. |
|
|
|
|
|
# **Limitations** |
|
|
|
1. **Performance on Ambiguous Instructions**: |
|
The model may struggle with ambiguous, vague, or poorly framed mathematical instructions, potentially leading to incorrect or incomplete solutions. |
|
|
|
2. **Edge Cases and Special Scenarios**: |
|
For highly specialized or niche mathematical problems, especially those not commonly encountered in training data, the model's performance may degrade. |
|
|
|
3. **Errors in Multi-Step Reasoning**: |
|
While trained on reasoning datasets, the model may sometimes produce incorrect results for multi-step or highly complex reasoning tasks, particularly if intermediate steps are not explicitly defined. |
|
|
|
4. **Bias Toward Common Solutions**: |
|
The model may favor standard or commonly used approaches to mathematical problems, potentially missing creative or less conventional methods of solution. |
|
|
|
5. **Resource Intensity**: |
|
As a large-scale model, PyThagorean requires significant computational resources, including high-end GPUs, for efficient inference and deployment. |
|
|
|
6. **Context Window Limitations**: |
|
The model's finite context window may lead to incomplete understanding or truncated responses for problems requiring extensive context or lengthy input. |
|
|
|
7. **Handling of Non-Mathematical Queries**: |
|
While capable of engaging in general conversations, its performance for non-mathematical tasks may not match models specifically tuned for broader use cases. |
|
|
|
8. **Dependency on Python Libraries**: |
|
Generated solutions may rely on specific Python libraries or functions, which users must have installed and configured correctly to execute the code successfully. |
