File size: 1,591 Bytes
d9faf63
 
7be0622
 
 
d9faf63
 
 
7be0622
 
 
 
 
 
d9faf63
 
7be0622
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
title: Smol Private Spreadsheet
emoji: 💻
colorFrom: blue
colorTo: yellow
sdk: docker
pinned: false
short_description: A spreadsheet powered by smolLM2 using WebGPU
tags:
  - custom-components-for-agents-track   # 🧩 Track 2
  - track2
  - gradio
  - smolllm
  - spreadsheet
---

# Smol-Private-Spreadsheet ✨

A tiny proof-of-concept that embeds **SmolLM 2** directly inside a sheet.  
Type a formula, get an answer—no extra UI required.

## Why it matters

| Feature | What it does |
|---------|--------------|
| **`=SMOLLM("prompt")`** | Asks the model and returns the reply inside the cell. |
| **Context support** | `=SMOLLM("prompt", A1)` or a range `A1:C3` to pass extra context. |
| **GPU / WebGPU ready** | Runs locally inside the Space; zero external API calls. |
| **First Step Towards Creating a WebGPU custom Gradio Component** |I'm hoping that this space can showcase the power of UI abstraction when combined with SmolLM2 and allow me to reach mentors that are willing to teach me how to create a custom gradio component that allows the use of small models using WebGPU.  |
| **Next Steps** | I'm still trying to create one Gradio app that allows me to do both, run a model locally using WebGPU and use MCP tools and other models using inference to bring the best of both worlds. It's a very challenging journey for me, especially given the limitation that all my ideas are competing for the same port at the same time: 7860. |

## Quick start

1. **Launch** the Space.  
2. In any empty cell, enter:

   ```text
   =SMOLLM("Summarise A1:C3 in one sentence", A1:C3)