mini-infer

mini-infer

A browser studio that simulates GPU PagedAttention block allocation, prefix caching, speculative decoding checks and KV-cache sizing. For engineers and students who want to see how LLM serving manages memory.

Made by prajwalpujari16 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: mini-infer is a browser studio that simulates how an LLM serving engine handles the KV cache on a GPU. It covers PagedAttention block pool allocation, prefix caching built on a copy-on-write radix tree, speculative decoding verification and KV-cache sizing by analytical estimate. It is a static page, so nothing runs on a server.

Who it suits: It suits people who want to watch memory management inside an inference engine rather than read about it.

Title: Who might use this studio
Industry | Who uses it | How they use it | Example
ML infrastructure | Serving engineers | Build intuition for cache memory behavior | Step through block allocation for a few requests
Education | Students | Learn how paged attention works | Watch a block table fill up
AI research | Researchers | Reason about prefix sharing and speculation | Try the speculative decoding acceptance verifier

If you want to benchmark a real serving engine, you can skip it, since this is a simulation.

Where to use it: You can use it online or open its files on your own computer.

• Web browser: the Space's page on Hugging Face, with nothing to install
• Windows, macOS or Linux: the downloaded files, opened in a browser
• Python: a PyPI package of the same name, named in the readme

Title: What the studio is good for
Use case | Who | What they get back
Learn paged attention | Students | A visual feel for how blocks are handed out
Explore prefix reuse | Engineers | A simulation of requests that share a starting prefix
Size a KV cache | Planners | An analytical estimate of cache size

How to set it up: The hosted page needs nothing. To get the files you need Python and pip.

• Python, with pip
• Any web browser

1. Install the huggingface_hub package.
2. Download the Space's files into a folder.
3. Move into that folder.
4. Open index.html in a web browser.

pip install -U huggingface_hub
hf download prajwalpujari16/mini-infer --repo-type space --local-dir mini-infer
cd mini-infer

The readme also names a PyPI package, which installs with this command.

pip install mini-infer

How to use it: Open the Space online, or open index.html from your downloaded folder. Then try the interactive tools one at a time, starting with the block table allocator.

Title: The interactive tools
Tool | What it does
Virtual block table allocator | Lets you see how cache blocks are given out from a pool
Copy-on-write radix prefix caching simulator | Shows how requests can share a common prefix
Speculative decoding acceptance verifier | Checks how drafted tokens would be accepted
KV-cache sizing | Estimates cache size analytically

Example: You open the block table allocator and add a few requests, then watch blocks being assigned from the pool. Next you open the prefix caching simulator with two requests that start the same way, and see how that shared start is reused.

For newcomers and experts: If you are new to this, start with the block allocator. The KV cache is the memory a model keeps for the text it has already read, and paging splits it into blocks.

If you know the field, the studio models block tables, a copy-on-write radix tree for prefix reuse and the acceptance check in speculative decoding. The readme names Qwen2.5 0.5B Instruct and 1.5B Instruct, and a serving traces dataset.

Open the studio
Pick a tool to simulate
Adjust requests and settings
Watch blocks, prefixes or acceptance change

Good to know: This is a simulation and an analytical sizing aid, not a live serving engine.

• Static page: plain HTML and CSS, with no server code
• Models: the Space names Qwen2.5 0.5B Instruct and Qwen2.5 1.5B Instruct
• License: apache-2
• Cost to run: none named
• Data it reaches: nothing beyond its own files is named in the notes
• Not for: benchmarking or running a real serving engine

On Hugging Face, 13 November 2026

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

Links

• Space: https://huggingface.co/spaces/prajwalpujari16/mini-infer
• Files: https://huggingface.co/spaces/prajwalpujari16/mini-infer/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.