---
library_name: transformers
tags:
- meta-math
- code
- instruct
- Zephyr-7B-Alpha
datasets:
- meta-math/MetaMathQA
base_model: HuggingFaceH4/zephyr-7b-alpha
license: apache-2.0
---

### Finetuning Overview:

**Model Used:** HuggingFaceH4/zephyr-7b-alpha  
**Dataset:** meta-math/MetaMathQA  

#### Dataset Insights:

The MetaMathQA dataset is a newly created dataset specifically designed for enhancing the mathematical reasoning capabilities of large language models (LLMs). It is built by bootstrapping mathematical questions and rewriting them from multiple perspectives, providing a comprehensive and challenging environment for LLMs to develop and refine their mathematical problem-solving skills.

#### Finetuning Details:

Using [MonsterAPI](https://monsterapi.ai)'s [LLM finetuner](https://docs.monsterapi.ai/fine-tune-a-large-language-model-llm), this finetuning:

- Was conducted with efficiency and cost-effectiveness in mind.
- Completed in a total duration of 10.9 hours for 0.5 epoch using an A6000 48GB GPU.
- Costed `$22.01` for the entire finetuning process.

#### Hyperparameters & Additional Details:

- **Epochs:** 0.5
- **Total Finetuning Cost:** $22.01
- **Model Path:** HuggingFaceH4/zephyr-7b-alpha
- **Learning Rate:** 0.0001
- **Data Split:** 95% train 5% validation
- **Gradient Accumulation Steps:** 4

---
Prompt Structure

```
Below is an instruction that describes a task. Write a response that appropriately completes the request.  


###Instruction:[query]


###Response:[response]
```

---
### Training loss:
![training loss](zephyr-mmqa-1.png "Training loss")



---
### Benchmark Results:

![GSM8K Accuracy ](benchmark.png "GSM8K Accuracy")

GSM8K is a dataset of 8.5K high quality linguistically diverse grade school math word problems, These problems take between 2 and 8 steps to solve, and solutions primarily involve performing a sequence of elementary calculations using basic arithmetic operations (+ − ×÷) to reach the final answer. A bright middle school student should be able to solve every problem. Its a industry wide used benchmark for testing an LLM for for multi-step mathematical reasoning.

 

---
license: apache-2.0