|
--- |
|
license: apache-2.0 |
|
language: |
|
- en |
|
tags: |
|
- Mathematical Reasoning |
|
--- |
|
# Model Card for Model ID |
|
|
|
This model is instruction-tuned [Mistral-7B](https://huggingface.co/mistralai/Mistral-7B-v0.1) model using QLoRA on single RTX 4090 GPU. Model is tuned to reason and answer mathematical problems. Model is also capable of writing a Python program that upon compilation prints answer to the question. To generate python program, you can ask model to write a Python program (as part of prompt) along with question. Refer to **Results** section for examples. |
|
|
|
## Model Details |
|
|
|
It is a Instruction-tuned Mistral-7B and performs mathematical reasoning and optionally write a Python program. Expect updates to model files for improved output quality. |
|
|
|
### Model Description |
|
|
|
- **Developed by:** [Ashvini Kumar Jindal](https://www.linkedin.com/in/ashvini-jindal-26653262/) |
|
- **Funded by:** self-work |
|
- **Model type:** Instruction-tuned |
|
- **Language(s) (NLP):** English |
|
- **Finetuned from model:** mistralai/Mistral-7B-v0.1 |
|
|
|
|
|
## Uses |
|
|
|
Arithmo-Mistral-7B is trained with the following format: |
|
|
|
``` |
|
Question: <question> |
|
|
|
Answer: |
|
``` |
|
|
|
It will perform best if queried in this way. |
|
|
|
## How to Get Started with the Model |
|
|
|
Model is compatibale with Huggingface. I will publish a generation/inference script soon. Model inference on CPU also works; I have tested it on Macbook M1 Pro. GPU inference is much faster than CPU inference. |
|
|
|
|
|
### Results |
|
|
|
Here are sample screenshots of model output for few questions :) |
|
|
|
|
|
 |
|
|
|
|
|
 |
|
|
|
|
|
 |
