File size: 2,982 Bytes
96642de
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
44
45
46
47
48
49
50
51
52
53
54
55
---
license: llama2
library_name: transformers
tags:
- code
model-index:
- name: Pandalyst_7B_v1.1
  results:
  - task:
      type: text-generation
    metrics:
    - name: exec@1
      type: exec@1
      value: 0.76
      verified: false
language:
- en
---


## Pandalyst: A large language model for mastering data analysis using pandas

<p align="center">
<img src="https://raw.githubusercontent.com/zhengyanzhao1997/Pandalyst/master/imgs/pandalyst.png" width="300"/>
</p>

<p align="center">
🐱 <a href="https://github.com/zhengyanzhao1997/Pandalyst" target="_blank">Github Repo</a> <br>
</p>

**What is Pandalyst**
- Pandalyst is a general large language model specifically trained to process and analyze data using the pandas library.

**How is Pandalyst**
- Pandalyst has strong generalization capabilities for data tables in different fields and different data analysis needs.

**Why is Pandalyst**
- Pandalyst is open source and free to use, and its small parameter size (7B/13B) allows us to easily deploy it on local PC. 
- Pandalyst can handle complex data tables (multiple columns and multiple rows), allowing us to enter enough context to describe our table in detail.
- Pandalyst has very competitive performance, significantly outperforming models of the same size and even outperforming some of the strongest closed-source models.


## News
- 🔥[2023/09/30] We released **Pandalyst-7B-V1.1** , which was trained on **CodeLlama-7b-Python** and achieves the **76.1 exec@1** in our **PandaTest_V1.0** and surpasses **Pandalyst-13B-V1.0**, **WizardCoder-Python-13B-V1.0** and **ChatGPT-3.5 (2023/06/13)**.
- 🔥[2023/09/28] We released **Pandalyst-13B-V1.0** , which was trained on **WizardCoder-Python-13B-V1.0** and achieves the **70.7 exec@1** in our **PandaTest_V1.0** and surpasses **WizardCoder-Python-13B-V1.0** and **ChatGPT-3.5 (2023/06/13)**.


| Model              | Checkpoint                                                                                  | Base Model | PandaTest_V1.0 | EASY | HARD | License |
|--------------------|---------------------------------------------------------------------------------------------|------------|----------------|---------------------|---------------------|  ----- | 
| Pandalyst-13B-V1.0 | 🤗 <a href="https://huggingface.co./pipizhao/Pandalyst_13B_V1.0" target="_blank">HF Link</a> |  WizardCoder-Python-13B-V1.0  | 70.7           | 75.6           | 65.9                |  <a href="https://ai.meta.com/resources/models-and-libraries/llama-downloads/" target="_blank">Llama2</a>  |
| Pandalyst-7B-V1.1  | 🤗 <a href="https://huggingface.co./pipizhao/Pandalyst-7B-V1.1" target="_blank">HF Link</a>  |  CodeLlama-7b-Python   | 76.1           | 85.2           | 67.0                   |  <a href="https://ai.meta.com/resources/models-and-libraries/llama-downloads/" target="_blank">Llama2</a>  |


## Usage
Please refer to <a href="https://github.com/zhengyanzhao1997/Pandalyst" target="_blank">Github</a>.