npm package discovery and stats viewer.

Discover Tips

  • General search

    [free text search, go nuts!]

  • Package details

    pkg:[package-name]

  • User packages

    @[username]

Sponsor

Optimize Toolset

I’ve always been into building performant and accessible sites, but lately I’ve been taking it extremely seriously. So much so that I’ve been building a tool to help me optimize and monitor the sites that I build to make sure that I’m making an attempt to offer the best experience to those who visit them. If you’re into performant, accessible and SEO friendly sites, you might like it too! You can check it out at Optimize Toolset.

About

Hi, 👋, I’m Ryan Hefner  and I built this site for me, and you! The goal of this site was to provide an easy way for me to check the stats on my npm packages, both for prioritizing issues and updates, and to give me a little kick in the pants to keep up on stuff.

As I was building it, I realized that I was actually using the tool to build the tool, and figured I might as well put this out there and hopefully others will find it to be a fast and useful way to search and browse npm packages as I have.

If you’re interested in other things I’m working on, follow me on Twitter or check out the open source projects I’ve been publishing on GitHub.

I am also working on a Twitter bot for this site to tweet the most popular, newest, random packages from npm. Please follow that account now and it will start sending out packages soon–ish.

Open Software & Tools

This site wouldn’t be possible without the immense generosity and tireless efforts from the people who make contributions to the world and share their work via open source initiatives. Thank you 🙏

© 2026 – Pkg Stats / Ryan Hefner

@altronis/tokenmark-cli

v0.4.1

Published

TokenMark CLI: recommend the best local-LLM model + config for your hardware from real benchmarks (Strix Halo, DGX Spark, RTX, and more).

Readme

@altronis/tokenmark-cli

Recommend the best local-LLM model + config for your hardware, from real community benchmarks: TokenMark in your terminal.

It pulls the live benchmark snapshot from tokenmark.app and ranks measured configs (Strix Halo, DGX Spark, RTX, Mac). It never invents numbers: every pick is a real config with a source link.

Use

npx @altronis/tokenmark-cli recommend --hardware "Strix Halo" --tasks coding,reasoning --prefer balanced

Or install it:

npm i -g @altronis/tokenmark-cli
tokenmark recommend --hardware "DGX Spark" --tasks vision --prefer quality

Commands

tokenmark recommend --hardware <hw> [--tasks a,b] [--prefer speed|quality|balanced] [--limit N] [--cards]
tokenmark configs   [--model <name>] [--hardware <hw>] [--cards]
tokenmark search    <term>
tokenmark models
tokenmark hardware  [<name>]
tokenmark submit    <github-or-hf-repo> [--note "..."] [--source hf]
tokenmark status    <submission-id>
tokenmark login     [--no-browser]
tokenmark whoami
tokenmark logout    [--env] [--force]
tokenmark submit-result [--from <file>] [--hardware <hw>] [--pick N] [--dry-run] [...]
tokenmark version
  • --tasks: any of coding, reasoning, long-context, vision, agentic, general
  • --prefer: speed (fastest), quality (most capable), balanced (default)
  • Flags take --flag value or --flag=value. A misspelt flag is an error, so --dryrun never sends for real.

Point it at a different source with TOKENMARK_URL (default https://tokenmark.app). A sign-in is only ever sent over https, or plain http to this machine.

Hardware catalogue

tokenmark hardware              # every platform: memory, bandwidth, boxes, tracked configs
tokenmark hardware gorgon       # one platform: chip specs with sources, the boxes that ship it, SG prices
tokenmark hardware "Mac Ultra"

Covers AMD Strix Halo, AMD Gorgon Halo, NVIDIA DGX Spark and Apple M Max / M Ultra. Names are forgiving: an id (gorgon-halo), a chip (Ryzen AI Max+ 395) or a box (GMKtec EVO-X5 Pro) all work.

GPU cards

GPU cards (RTX 5090, RTX PRO 6000) are ranked in a lane of their own, never against the AI boxes, so a 32 GB card at 1.8 TB/s does not bury a 128 GB box. Name a card as the hardware, or add --cards:

tokenmark recommend --hardware "RTX 5090"   # card picks only
tokenmark configs --cards                    # every card config, fastest first
tokenmark hardware "RTX PRO 6000"           # VRAM and bandwidth with sources, tracked configs

Box results say how many more matches sit in the card lane, and the other way round.

Adding numbers

Two ways in, and a person reviews both before anything goes live:

  • tokenmark submit <repo> sends a GitHub or Hugging Face repo whose README has benchmark numbers.
  • tokenmark submit-result sends a speed you measured yourself.

tokenmark status <id> shows where either one is.

Sending a speed you measured

Sign in once:

tokenmark login

It prints a short code and opens https://tokenmark.app/cli. Sign in on that page if you aren't already. The page shows which program asked and how long ago: check the code matches your terminal, then approve. If the request came from a different network than your browser, as it does over SSH, the page asks you to type the code yourself. The terminal then holds a sign-in that can only send results for review, and it lasts 90 days. It is saved in ~/.config/tokenmark/credentials.json, readable only by you. On a box with no browser, run tokenmark login --no-browser and open the printed link on another device.

tokenmark whoami shows who the terminal is signed in as. tokenmark logout revokes the saved sign-in on the server and deletes the local copy. If the server can't be reached it keeps the sign-in and says so: run it again, or add --force to delete the local copy anyway (it then stays valid until it expires). --env also revokes a token set in TOKENMARK_TOKEN.

https://tokenmark.app/cli lists every terminal signed in to your account, with a Revoke button on each, so you can end one without the machine it is on.

Then pipe llama-bench straight in:

llama-bench -m Qwen3-8B-Q4_K_M.gguf -o json | tokenmark submit-result --hardware "Strix Halo"

Or send a saved file, or type the numbers in:

tokenmark submit-result --from bench.json --hardware dgx-spark
tokenmark submit-result --model Qwen3.8-27B --quant Q4_K_M --hardware strix-halo --backend llama.cpp --decode 72.4

It reads llama-bench output (json, jsonl, csv or the default table) and vllm bench serve output. Model, quant, backend and speeds come from the output, and any flag you type wins over what it read. If the output holds more than one setup, it lists them and you choose one with --pick N. --dry-run prints the request it would send and sends nothing.

More fields: --prefill --ctx --concurrency --variant --backend-version --mode --device --ram --proof <url> --note --credit <handle> --command "<how you ran it>".

In CI, set TOKENMARK_TOKEN to the token that tokenmark login saved, instead of keeping the credentials file.

No terminal? The signed-in form at https://tokenmark.app/submit takes the same result.

What it shows

Each pick is a real measured config: decode tok/s, quant, backend, the hardware it ran on, why it was picked, who benchmarked it, and a link to the source. Aggregate/multi-GPU numbers are labeled so they aren't read as single-stream speed, and runs that used speculative decoding (MTP, DFlash, a draft model) say so, so they aren't read as a plain run of the same quant.

MIT licensed. Data aggregated from public community benchmarks with attribution.