← All reviews · Local models & inference

local-models/runtimessmoke testsmoke test Rocky's quickest test: fresh sandbox, does it install and does the README example run? Glossary

colibri: frontier MoE models on hardware you already own?

The claim“Run frontier MoE models (hundreds of billions of parameters) on hardware you already own: pure C, experts streamed from disk”

Rocky, verdict poseTESTING…ScorependingNo score yet
GitHubJustVugg/colibri42.7kStars4.8kForks189Issues + PRsApache-2.0license (Apache-2.0, MIT) The rules for using the code. Apache-2.0 and MIT both allow free use, including commercial use. Glossary · Learn moreLicenseLast pushCLanguage

Live stats · checked

Review lives at rockytested.com · @RockyTested

⏱️ Queued. Estimated: starts , done (estimate, may move)
Rocky dressed for AI modelsTesting badge: Rocky in his lab coat

AI models Rocky · in the lab

With vs without

With it loadedNot tested yet

No with-skill pass count is on file.

Without itNot tested yet

No baselinebaseline The normal way of doing the same job, used as the yardstick. Rocky compares every claim against a baseline on the same machine. Glossary pass count is on file.

Usefulness is a separate question from “does the repo exist.” It needs the same task run twice. Until that A/B is published, both cells stay blank.

What the test plan says

Smoke first: build the pure-C engine and run its `--list`, which reports which models fit this machine and why. A real streamed run of a large model, measuring tokenstoken The small chunks a model reads and writes. One token is roughly three-quarters of an English word, so 100 tokens is about 75 words. Glossary · Learn more/second, is queued as a heavier claim testclaim test Rocky measures the headline claim ("10x faster!") on real tasks against a baseline. Glossary.

Four verdicts

One stamp is not the whole story. Works, Safe, Claims honest, and Provenance are four separate written calls. A missing call stays “not tested yet.”

WorksNOT TESTED YETDid it do the thing when Rocky ran it?

No written call is on file. This is not a pass.

SafeNOT TESTED YETSafe to install and run?

No written call is on file. This is not a pass.

Claims honestNOT TESTED YETDoes the pitch match what Rocky measured?

No written call is on file. This is not a pass.

ProvenanceNOT TESTED YETDoes the repo's history check out?

No written call is on file. This is not a pass.

Safety is a hard cap. If Safe is no, the Rocky Score cannot rise above 44 (grade F), even when the tool works. A blank Safe card means not tested yet, not safe. How the four verdicts roll up →

Claim vs what Rocky measured
Claim vs what Rocky measuredThe claim: Run frontier MoE models (hundreds of billions of parameters) on hardware you already own: pure C, experts streamed from disk. The claim: Measuring… (measuring).THE CLAIM“Run frontier MoEmodels (hundreds ofbillions ofparameters) onhardware you alreadyown: pure C, expertsstreamed from disk• Source: X postvsWHAT ROCKY MEASURED⚗The claimMeasuring…Claim vs what Rocky measuredThe claim: Run frontier MoE models (hundreds of billions of parameters) on hardware you already own: pure C, experts streamed from disk. The claim: Measuring… (measuring).THE CLAIM“Run frontier MoE models(hundreds of billions ofparameters) on hardwareyou already own: pure C,experts streamed from disk• Source: X postvsWHAT ROCKY MEASURED⚗The claimMeasuring…
Claim partMeasuredResult
The claimMeasuring…measuring

What it is: A pure-C engine that keeps a model's dense parts in RAM and streams its experts from disk, to run huge mixture-of-experts models on ordinary hardware. Rocky first builds it and reads its honesty about what fits; the big streamed run comes later. Claim source →

Results coming. Here’s what Rocky will measure

  • Does the engine build
  • Does `--list` report what fits this machine
  • Tokens/second on a real streamed model (claim run, queued)

Smoke first: build the pure-C engine and run its `--list`, which reports which models fit this machine and why. A real streamed run of a large model, measuring tokens/second, is queued as a heavier claim test.

Sneak peek

Claim ledger

Every distinct claim in the pitch gets a row: what was said, what the command said back. 1 claim on the ledger: 1 not measured. No separate ledger file yet, so these rows are the review's claim list only. “Not tested” means that claim has no measurement.

🔍 Coverage: Not stated yet. Rocky only lists the parts a run actually checked.

⚗️ Unproven usefulness. Counting and parsing can settle what the repo claims. It can't settle whether the thing actually helps — that takes an A/B run (same task with and without it), which hasn't happened yet.

Rocky's take

Not written yet.

Rocky's take is drafted only from measured results: what the tool is good for, and where it will fail you. Until those results exist, this stays blank. Nothing here is guessed.

Where this repo comes from

Not checked yet.

Provenance is age, authors, commitcommit One saved version of a repo's code. Rocky pins the exact commit so anyone can rerun the same code. Glossary · Learn more bursts, and where the README sends you. It is published only from commands Rocky ran. Nothing is guessed, so this section stays blank until that file exists.

Score

Scorepending

Not scored yet. A Rocky Score is a rollup of measured subscores only. This review does not have one, and no grade is filled in to hold the place.

Rerun it yourself

📦 When the test is done, the exact scripts, raw results and a copy-paste prompt for your own AI agentAI agent An AI that takes actions in steps (runs commands, edits files, browses) instead of only chatting. Glossary · Learn more will appear here.

Repo & thanks

Where the claim was posted

Tom Dörr
@tom_doerr ·
The claim Rocky tested

Colibri is a pure C inferenceinference Using a trained model to get answers (as opposed to training it). Glossary · Learn more engine that runs MoE models up to 2.8 trillion parameters on consumer hardware by treating storage, RAM, and VRAMVRAM / GPU memory The memory on a graphics card. A "4 GB card" has 4 GB of it. Models normally have to fit here to run fast. Glossary · Learn more as a single hierarchy. https://github.com/JustVugg/colibri

💬 8🔁 22❤️ 162📊 10K🔖 139

Counts as of (Rocky's last check)

Open on X →

Where else this repo is talked about

On X: 1 posts linking the repo

Posts from to , in several languages. Found with X search for links to the repo; more may exist. Oldest first:

  • Earliest we found @tom_doerr Colibri is a pure C inference engine that runs MoE models up to 2.8 trillion parameters on consumer hardware by treating storage, RAM, and VRAM as a single hierarchy. https://github.com/JustVugg/colibri❤️ 162 · 🔁 22 · 📊 10K open
All 1 posts, oldest first
  • Earliest we found @tom_doerr Colibri is a pure C inference engine that runs MoE models up to 2.8 trillion parameters on consumer hardware by treating storage, RAM, and VRAM as a single hierarchy. https://github.com/JustVugg/colibri❤️ 162 · 🔁 22 · 📊 10K open

Ran it yourself?

Share your rerun

  1. Download the rerun bundle above.
  2. Run it, or paste the prompt into your AI agent.
  3. Post your results.json (a gist is fine) and reply to @RockyTested with the link.

Rocky reruns or spot-checks every submission before it counts. Verified reruns will appear here with your handle.

🏅 Helper leaderboard

No verified reruns yet. Be the first!

Street cred goes to people whose reruns Rocky verified: same bundle, raw output included, numbers re-checked. How it works

Badge & stat card

Rocky Tested: TESTING

Put the verdict badge in a README or post. It always links back to the measured results.

[![Rocky Tested: TESTING](https://rockytested.com/reviews/justvugg-colibri/badge.svg)](https://rockytested.com/reviews/justvugg-colibri/)
Rocky Tested stat card for colibri: frontier MoE models on hardware you already own?

Stat card (1200×630). This is what shows up when the review is shared on X.

Machine-readable: /v1/reviews/justvugg-colibri · Words with a dotted underline have a quick explanation: hover or tap. Glossary