File size: 1,447 Bytes
a5e4a17
 
 
 
 
 
b5abee3
a5e4a17
 
 
 
 
 
 
b5abee3
 
 
 
0b3d0a0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
---
datasets:
- AI-MO/NuminaMath-CoT
language:
- en
library_name: transformers
license: cc-by-4.0
tags:
- text-generation-inference
- chat
- qwen2
- conversational
- math
- maths
- unsloth
- trl
- sft
---

# xsanskarx/qwen2-0.5b_numina_math-instruct

This repository contains a fine-tuned version of the Qwen-2 0.5B model specifically optimized for mathematical instruction understanding and reasoning. It builds upon the Numina dataset, which provides a rich source of mathematical problems and solutions designed to enhance reasoning capabilities even in smaller language models.

## Motivation

My primary motivation is the hypothesis that high-quality datasets focused on mathematical reasoning can significantly improve the performance of smaller models on tasks that require logical deduction and problem-solving. Uploading benchmarks is the next step in evaluating this claim.

## Model Details

* **Base Model:** Qwen-2 0.5B
* **Fine-tuning Dataset:** Numina COT
* **Key Improvements:** Enhanced ability to parse mathematical instructions, solve problems, and provide step-by-step explanations.

## Usage

You can easily load and use this model with the Hugging Face Transformers library:

```python
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("xsanskarx/qwen2-0.5b_numina_math-instruct")
model = AutoModelForCausalLM.from_pretrained("xsanskarx/qwen2-0.5b_numina_math-instruct")