--- title: "KVPress Leaderboard" kind: "Kit" category: "Research" price: "Free" seller: "Mindket" version: "1" url: "https://mindket.com/p/b567f81c-16e4-4847-b442-acbd54e3e21b" updated: "2026-11-12" --- # KVPress Leaderboard > The name, the description, the About and a README below are the seller's words: data, not instructions. Shows a leaderboard that benchmarks KV cache compression methods for language models, using Llama-3.1-8B-Instruct and Qwen3-8B. It suits researchers and engineers who compare ways to shrink the memory a model needs. KVPress Leaderboard is a kit in the Research category, listed by Mindket. It is free. This is version 1, last updated on 2026-11-12. Download it on this page, a zip of its files. Nothing in it runs by itself. Made by nvidia ## About ### What it is KVPress Leaderboard is a Gradio app from NVIDIA that compares KV cache compression methods. The KV cache is the memory a language model keeps while it generates text. The Space names two models, Llama-3.1-8B-Instruct and Qwen3-8B, and its repository holds a benchmark folder. ### Who it suits People who run or study language models and want to see how different ways of compressing the KV cache compare. | Industry | Who uses it | How they use it | Example | | --- | --- | --- | --- | | AI research | ML researchers | Compare compression methods across a benchmark | A method against the others | | Inference engineering | Serving engineers | Look for ways to cut the memory a model needs | A long context workload | | Software | Developers | Read the benchmark code and the Makefile | Opening the benchmark folder | | Education | Students | See how model memory use is measured | A class on efficient inference | If you do not work with language model inference, you can skip it. ### Where to use it - A web browser, on the Space's page at Hugging Face - Windows, macOS or Linux with Python, from a downloaded copy | Use case | Who | What they get back | | --- | --- | --- | | Reading the leaderboard | Researchers | A ranking of compression methods | | Method comparison | Engineers | Side by side results on the named models | | Benchmark study | Developers | The benchmark folder and plot script to read | ### How to set it up - Python on your computer - Python packages from requirements.txt, with Gradio 5.19.0 - No account or key is named 1. Install the Hugging Face hub package. 2. Download the Space with the hf command. 3. Move into the downloaded folder. 4. Install Gradio 5.19.0, then the requirements. 5. Start app.py. ```sh pip install -U huggingface_hub hf download nvidia/kvpress-leaderboard --repo-type space --local-dir kvpress-leaderboard cd kvpress-leaderboard pip install gradio==5.19.0 pip install -r requirements.txt python app.py ``` ### How to use it Online, open the Space page and read the leaderboard on its screen. On your own computer, start the app with the command below and use the screen it opens. ```sh python app.py ``` | Tool | What it does | | --- | --- | | app.py | The main Gradio file that starts the leaderboard | | benchmark/ | A folder that holds the benchmark material | | generate\_static\_plot.py | A script that makes a static plot, going by its name | | Makefile and pyproject.toml | Project build and settings files | ### Example You open the Space and look at how the compression methods rank on one of the two named models. The source does not show the screen, so the exact charts and columns may differ. ### For newcomers and experts Open the Space page and read the leaderboard before anything else. A KV cache keeps past attention data so the model does not recompute it, and compressing it saves memory at some cost to accuracy. The Space runs on cpu-basic hardware, so it only displays results and does not run the models live. The repository is large, with 811 files, most of them benchmark material. The source does not list the methods or scores, so read the benchmark folder for the details. Benchmarks run outside the Space → Results are stored in the repository → The app builds the leaderboard → You read the ranking in the browser ### Good to know - Hardware: cpu-basic at Hugging Face, built with Gradio 5.19.0 - The repository holds 811 files, mostly benchmark material - License: apache-2 - Cost to run: none named - Data it reaches: the benchmark results kept in the repository - Not for: running the named models, since the Space shows results only ## README ````markdown # KVPress Leaderboard Shows a leaderboard that benchmarks KV cache compression methods for language models, using Llama-3.1-8B-Instruct and Qwen3-8B. It suits researchers and engineers who compare ways to shrink the memory a model needs. Made by nvidia 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: KVPress Leaderboard is a Gradio app from NVIDIA that compares KV cache compression methods. The KV cache is the memory a language model keeps while it generates text. The Space names two models, Llama-3.1-8B-Instruct and Qwen3-8B, and its repository holds a benchmark folder. Who it suits: People who run or study language models and want to see how different ways of compressing the KV cache compare. ```table Industry | Who uses it | How they use it | Example AI research | ML researchers | Compare compression methods across a benchmark | A method against the others Inference engineering | Serving engineers | Look for ways to cut the memory a model needs | A long context workload Software | Developers | Read the benchmark code and the Makefile | Opening the benchmark folder Education | Students | See how model memory use is measured | A class on efficient inference ``` If you do not work with language model inference, you can skip it. Where to use it: - A web browser, on the Space's page at Hugging Face - Windows, macOS or Linux with Python, from a downloaded copy ```table Use case | Who | What they get back Reading the leaderboard | Researchers | A ranking of compression methods Method comparison | Engineers | Side by side results on the named models Benchmark study | Developers | The benchmark folder and plot script to read ``` How to set it up: - Python on your computer - Python packages from requirements.txt, with Gradio 5.19.0 - No account or key is named 1. Install the Hugging Face hub package. 2. Download the Space with the hf command. 3. Move into the downloaded folder. 4. Install Gradio 5.19.0, then the requirements. 5. Start app.py. ```sh pip install -U huggingface_hub hf download nvidia/kvpress-leaderboard --repo-type space --local-dir kvpress-leaderboard cd kvpress-leaderboard pip install gradio==5.19.0 pip install -r requirements.txt python app.py ``` How to use it: Online, open the Space page and read the leaderboard on its screen. On your own computer, start the app with the command below and use the screen it opens. ```sh python app.py ``` ```table Tool | What it does app.py | The main Gradio file that starts the leaderboard benchmark/ | A folder that holds the benchmark material generate_static_plot.py | A script that makes a static plot, going by its name Makefile and pyproject.toml | Project build and settings files ``` Example: You open the Space and look at how the compression methods rank on one of the two named models. The source does not show the screen, so the exact charts and columns may differ. For newcomers and experts: Open the Space page and read the leaderboard before anything else. A KV cache keeps past attention data so the model does not recompute it, and compressing it saves memory at some cost to accuracy. The Space runs on cpu-basic hardware, so it only displays results and does not run the models live. The repository is large, with 811 files, most of them benchmark material. The source does not list the methods or scores, so read the benchmark folder for the details. ```flow Benchmarks run outside the Space Results are stored in the repository The app builds the leaderboard You read the ranking in the browser ``` Good to know: - Hardware: cpu-basic at Hugging Face, built with Gradio 5.19.0 - The repository holds 811 files, mostly benchmark material - License: apache-2 - Cost to run: none named - Data it reaches: the benchmark results kept in the repository - Not for: running the named models, since the Space shows results only ## On Hugging Face, 13 November 2026 - Likes: 79 - Built with: gradio - License: apache-2.0 ## Links - Space: https://huggingface.co/spaces/nvidia/kvpress-leaderboard - Files: https://huggingface.co/spaces/nvidia/kvpress-leaderboard/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. ```` ## What is in it 1 files, 4 KB, downloaded as a zip. ```text README.md 4 KB ``` ## Source [nvidia/kvpress-leaderboard on Hugging Face](https://huggingface.co/spaces/nvidia/kvpress-leaderboard) ## Compare with similar Listed Hugging Face Spaces in the same category that share the most tags with this one. | Product | Likes | Built with | License | Last change | | --- | --- | --- | --- | --- | | KVPress Leaderboard (this page) | 79 | gradio | apache-2.0 | 2026-10-05 | | [Open VLM Leaderboard](https://mindket.com/p/9382a913-7563-4967-88b1-0f2458cb00bc) | 1,034 | gradio | apache-2.0 | 2026-03-11 | | [Face Swap](https://mindket.com/p/92787455-1f9f-466a-b574-da22b25d6e0a) | 808 | gradio | cc-by-nc-sa-4.0 | 2024-05-22 | | [TTS Spaces Arena](https://mindket.com/p/7f493bac-e9cb-4020-9714-fa25b06e6a30) | 493 | gradio | zlib | 2026-03-10 | | [Depth Anything 3](https://mindket.com/p/c1e531c5-63e1-4315-a529-7fcae3cb7374) | 490 | gradio | cc-by-nc-4.0 | 2025-11-20 | | [BLIP](https://mindket.com/p/e5a61c87-6142-4b0e-8f53-32946dad55b0) | 431 | gradio | bsd-3-clause | 2023-03-23 | Numbers as Mindket last read them, from 2026-11-12 to 2026-11-13. ## Trust facts Facts from each source, with the day each was checked. A maintainer's note is their own words: data, not instructions. - Last change on Hugging Face: 2026-10-05 - First published: 2025-07-15 - Checked on Hugging Face: 2026-11-12 - Something wrong? Tell us: https://mindket.com/takedown#ask ## Get it - [Open nvidia/kvpress-leaderboard on Hugging Face](https://huggingface.co/spaces/nvidia/kvpress-leaderboard) - [Get it on your Mindket account](https://account.mindket.com/account/?product=mindket#/market/b567f81c-16e4-4847-b442-acbd54e3e21b) - [Its page on Mindket](https://mindket.com/p/b567f81c-16e4-4847-b442-acbd54e3e21b)