Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: llama2
|
| 3 |
+
library_name: transformers
|
| 4 |
+
tags:
|
| 5 |
+
- code
|
| 6 |
+
model-index:
|
| 7 |
+
- name: Pandalyst_7B_v1.1
|
| 8 |
+
results:
|
| 9 |
+
- task:
|
| 10 |
+
type: text-generation
|
| 11 |
+
metrics:
|
| 12 |
+
- name: exec@1
|
| 13 |
+
type: exec@1
|
| 14 |
+
value: 0.76
|
| 15 |
+
verified: false
|
| 16 |
+
language:
|
| 17 |
+
- en
|
| 18 |
+
---
|
| 19 |
+
|
| 20 |
+
|
| 21 |
+
## Pandalyst: A large language model for mastering data analysis using pandas
|
| 22 |
+
|
| 23 |
+
<p align="center">
|
| 24 |
+
<img src="https://raw.githubusercontent.com/zhengyanzhao1997/Pandalyst/master/imgs/pandalyst.png" width="300"/>
|
| 25 |
+
</p>
|
| 26 |
+
|
| 27 |
+
<p align="center">
|
| 28 |
+
🐱 <a href="https://github.com/zhengyanzhao1997/Pandalyst" target="_blank">Github Repo</a> <br>
|
| 29 |
+
</p>
|
| 30 |
+
|
| 31 |
+
**What is Pandalyst**
|
| 32 |
+
- Pandalyst is a general large language model specifically trained to process and analyze data using the pandas library.
|
| 33 |
+
|
| 34 |
+
**How is Pandalyst**
|
| 35 |
+
- Pandalyst has strong generalization capabilities for data tables in different fields and different data analysis needs.
|
| 36 |
+
|
| 37 |
+
**Why is Pandalyst**
|
| 38 |
+
- Pandalyst is open source and free to use, and its small parameter size (7B/13B) allows us to easily deploy it on local PC.
|
| 39 |
+
- Pandalyst can handle complex data tables (multiple columns and multiple rows), allowing us to enter enough context to describe our table in detail.
|
| 40 |
+
- Pandalyst has very competitive performance, significantly outperforming models of the same size and even outperforming some of the strongest closed-source models.
|
| 41 |
+
|
| 42 |
+
|
| 43 |
+
## News
|
| 44 |
+
- 🔥[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)**.
|
| 45 |
+
- 🔥[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)**.
|
| 46 |
+
|
| 47 |
+
|
| 48 |
+
| Model | Checkpoint | Base Model | PandaTest_V1.0 | EASY | HARD | License |
|
| 49 |
+
|--------------------|---------------------------------------------------------------------------------------------|------------|----------------|---------------------|---------------------| ----- |
|
| 50 |
+
| 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> |
|
| 51 |
+
| 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> |
|
| 52 |
+
|
| 53 |
+
|
| 54 |
+
## Usage
|
| 55 |
+
Please refer to <a href="https://github.com/zhengyanzhao1997/Pandalyst" target="_blank">Github</a>.
|