---
language:
- en
license: apache-2.0
library_name: transformers
tags:
- python
- java
- cpp
- sql
- function calling
- unit tests
- causalLM
- codeLLAMA modified archi
- document
- code
- code2doc
- instruction_tuned
- basemodel
- pytorch
- docstring
- documentation
- text-generation-inference
- plan
- planner
metrics:
- accuracy
pipeline_tag: text-generation
widget:
- text: '<example_response>--code:def function_divide2(x): return x / 2--question:Document
    the code--doc:Description:This function takes a number and divides it by 2.Parameters:-
    x (numeric): The input value to be divided by 2.Returns:- float: The result of
    x divided by 2.Example:To call the function, use the following code:function_divide2(1.0)</example_response><function_code>def
    _plot_bounding_polygon(polygons_coordinates, output_html_path=bounding_polygon_map.html):map_center
    = [sum([coord[0]for polygon_coords in polygons_coordinatesfor coord in polygon_coords])/
    sum([len(polygon_coords) for polygon_coords in polygons_coordinates]),sum([coord[1]for
    polygon_coords in polygons_coordinatesfor coord in polygon_coords])/ sum([len(polygon_coords)
    for polygon_coords in polygons_coordinates]),]my_map = folium.Map(location=map_center,
    zoom_start=12)for polygon_coords in polygons_coordinates:folium.Polygon(locations=polygon_coords,color=blue,fill=True,fill_color=blue,fill_opacity=0.2,).add_to(my_map)marker_cluster
    = MarkerCluster().add_to(my_map)for polygon_coords in polygons_coordinates:for
    coord in polygon_coords:folium.Marker(location=[coord[0], coord[1]], popup=fCoordinates:
    {coord}).add_to(marker_cluster)draw = Draw(export=True)draw.add_to(my_map)my_map.save(output_html_path)return
    output_html_path</function_code><question>Document the python code above giving
    function description ,parameters and return type and example how to call the function</question><doc>'
  example_title: example
---
# pip-code-bandit

[PipableAI](https://www.pipable.ai/)

[colab_notebook](https://colab.research.google.com/drive/10av3SxFf0Psx_IkmZbcUhiVznStV5pVS?usp=sharing)

[pipflow](https://github.com/PipableAI/pipflow)

[linkedin_post](https://www.linkedin.com/posts/pipable%2Eai_releasing-strategy-activity-7195750109886783489-tHrz?utm_source=share&utm_medium=member_desktop)

[reddit_post](https://www.reddit.com/r/LocalLLaMA/comments/1cqxdl9/unveiling_pipcodebandit_empowering_ai_in_agentic/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button)


## Objective


![image/png](https://cdn-uploads.huggingface.co/production/uploads/658d8095a2a6a6e0da8bb8a6/NuTFBTMAsPgFwMxCjdqFv.png)


Given a goal and tools, can AI intelligently use the tools to reach the goal?\
What if it has a meagre 1.3b params/neurons akin to that of an owl? Can it follow instructions and plan to reach a goal?\
It can!\
Releasing **pip-code-bandit** and **pipflow**\
A `model` and a `library` to manage and run goal-oriented agentic system.


## Model attributes

```javascript
-- number of params ~ 1.3b [2.9 Gb GPU memory footprint]
-- sequence length ~ 16.3k [Can go higher but will show performance degradation]
-- license - apache 2.0
-- instruction following , RL tuned.
-- tasks:
1. complex planning(plan) of sequential function calls | a list of callables and goal
2. corrected plan | feedback instructions with error
3. function calling | doc or code and goal
4. code generation | plan and goal
5. code generation | goal
6. doc generation | code
7. code generation | doc
8. file parsed to json | any raw data
9. sql generation | schema, question, instructions and examples

```


## How did we build it?

We used a simulator to simulate environments where the model could play games to achieve goals, given a set of actions available to it. 
All the model could do was find the right action and config to incur a positive reward.
The reward policy is around the concept of a model going to a stable state of zero net sum reward for both good and bad behaviour.
In this setup, the model, which was pre-trained on code, function documentation, and similar OS datasets, was RL-tuned for reliability and instruction-following. 

## License
```bash
complete open-sourced - apache 2.0. License
```

## Usage


### NOTE:


If you wish to try this model without utilizing your GPU, we have hosted the model on our end. To execute the library using the hosted model, initialize the generator as shown below:

```bash
pip3 install git+https://github.com/PipableAI/pipflow.git
```
```python
from pipflow import PipFlow

generator = PipFlow()
```

We have hosted the model at https://playground.pipable.ai/infer. Hence, one can also make a POST request to this endpoint with the following payload:

```json
{
    "model_name": "PipableAI/pip-code-bandit",
    "prompt": "prompt",
    "max_new_tokens": "400"
}
```

```bash
curl -X 'POST' \
  'https://playground.pipable.ai/infer' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/x-www-form-urlencoded' \
  -d 'model_name=PipableAI%2Fpip-code-bandit&prompt="YOUR PROMPT"&max_new_tokens=400'
```

Alternatively, you can directly access the UI endpoint at https://playground.pipable.ai/docs#/default/infer_infer_post.



### Library Usage

To directly use the model's capabilities without putting extra effort into schemas and prompts, try to use [pipflow](https://github.com/PipableAI/pipflow).

For detailed usage, refer to the [colab_notebook](https://colab.research.google.com/drive/10av3SxFf0Psx_IkmZbcUhiVznStV5pVS?usp=sharing)



### Model Usage

```bash
pip install transformers accelerate torch
```

```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from accelerate import Accelerator
model =AutoModelForCausalLM.from_pretrained("PipableAI/pip-code-bandit",torch_dtype=torch.bfloat16,device_map="auto")
tokenizer = tokenizer = AutoTokenizer.from_pretrained("PipableAI/pip-code-bandit")
new_tokens = 600
prompt = """
<question>
Generate a python function for adding two numbers.
</question>
<code>
"""
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=new_tokens)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
response = response.split("<code>")[1].split("</code>")[0]

print(response)
```


### Prompt

```python
prompt = f"""<example_response>{--question , --query}</example_response><function_code>{code}</function_code>
<question>Give one line description of the python code above in natural language.</question>
<doc>"""

prompt = f"""<example_response>{example of some  --question: , --query}</example_response><schema>{schema with cols described}</schema>
<question>Write a sql query to ....</question>
<sql>"""
```

### Team

```doc
Avi Kothari, Gyan Ranjan, Pratham Gupta, Ritvik Aryan Kalra, Soham Acharya
```