DeepSmolLM3 3B Chat

DeepSmolLM3 3B Chat

A browser chat demo for DeepSmolLM3-3B, a 3B-parameter reasoning model, that shows the model's long thinking apart from its answer. For anyone who wants to try a small reasoning model without an API key.

Made by Oxygen-Low on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: DeepSmolLM3 3B Chat is a chat demo for DeepSmolLM3-3B from Oxygen-Low, a reasoning model with 3B parameters. The model was fine-tuned from HuggingFaceTB's SmolLM3-3B on the Jackrong/DeepSeek-V4-Pro-Distilled-200K dataset. You talk to it in the browser, and its long reasoning appears apart from the answer.

Who it suits: It suits people who want to watch a small reasoning model work through a problem.

Title: Who might use this demo
Industry | Who uses it | How they use it | Example
Machine learning | Researchers | Read the reasoning a 3B model produces | Ask a logic puzzle and read the thinking shown apart from the answer
Software | Developers | Test how sampling settings change replies | Raise the temperature and compare two answers
Education | Students | See a model work through a problem step by step | Ask for a maths check and follow the reasoning

If you need a hosted API for your own app, you can skip it, since this is a chat page.

Where to use it: You can use it online or run a copy yourself.

• Web browser: the Space's page on Hugging Face, with no API key needed
• Windows, macOS or Linux: a local copy that runs with Python
• Hugging Face Spaces: where the hosted chat runs, on ZeroGPU

Title: What the demo is good for
Use case | Who | What they get back
Chat with a reasoning model | Anyone curious | Replies with the thinking shown separately from the answer
Tune generation settings | Researchers | Replies shaped by system prompt, length, temperature and top-p
Compare thinking on and off | Developers | The same question answered with and without extended reasoning

How to set it up: The hosted chat needs nothing. A local copy needs Python and pip.

• Python, with pip
• The Python packages that requirements.txt lists

1. Install the huggingface_hub package.
2. Download the Space's files into a folder.
3. Move into that folder.
4. Install Gradio 6.5.1.
5. Install the packages from requirements.txt.
6. Start the app with Python.

pip install -U huggingface_hub
hf download Oxygen-Low/DeepSmolLM3-3B --repo-type space --local-dir DeepSmolLM3-3B
cd DeepSmolLM3-3B
pip install gradio==6.5.1
pip install -r requirements.txt
python app.py

How to use it: Open the Space online, or start your copy with the command below. Then type a message, watch the reply stream in, and adjust the settings when you want a different result.

python app.py

Title: The controls you will use
Tool | What it does
Thinking mode toggle | Turns the model's reasoning on or off
/think | Asks for extended thinking, and the long reasoning is displayed in its own place
System prompt | Replaces SmolLM3's built-in reasoning instructions when you fill it in
Output length | Sets how long a reply can be, up to 8192 tokens
Temperature and top-p | Control how varied the replies are

Example: You ask, "Is 91 a prime number? Explain your reasoning." The reasoning model works through the question, its long thinking appears apart from the answer, and then the final reply streams in.

For newcomers and experts: If you are new to this, leave the system message empty and ask one question. That way the model uses its own reasoning instructions, and you see how it thinks.

If you know the field, this is a 3B model fine-tuned from SmolLM3-3B on a distilled dataset, and the app streams tokens with transformers on ZeroGPU. Filling in a system prompt swaps out SmolLM3's default reasoning instructions, and the model may then answer without thinking first.

You type a message
The model reasons at length
The thinking shows apart from the answer
The final reply streams in

Good to know: Thinking is optional, and settings change what you get.

• Thinking: you can switch reasoning on or off
• System prompt: a custom one can make the model skip its thinking, so leave it empty to keep the reasoning
• Length: replies can run up to 8192 tokens
• License: apache-2
• Cost to run: no API key is required, and no other paid service is named
• Data it reaches: the messages you send, plus the Python packages pip installs from requirements.txt
• Not for: keeping the model's reasoning when you use a custom system prompt

On Hugging Face, 13 November 2026

• Likes: 1
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/Oxygen-Low/DeepSmolLM3-3B
• Files: https://huggingface.co/spaces/Oxygen-Low/DeepSmolLM3-3B/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.