A model that
knows your labels.

uli fine-tunes a small model on your own labelled data, inside a running inference engine on a Mac, in the idle time between requests.

Free. Sign-ups close 20 November. How it works for testers

uli watch banking77-20261029-184325
$ uli submit train.chat.jsonl --base qwen3.5-9b --name banking77-20261029-184325
dataset 2,000 examples · 77 labels · 0 bad rows · 724 KB
uploaded sha256 27c2…76f6
assigned node e2e-node · qwen3.5-9b resident · idle
job banking77-20261029-184325 (j-bc722f26) queued → training

$ uli watch banking77-20261029-184325

  banking77-20261029-184325 · qwen3.5-9b · rank 8 · 3 epochs   training

  progress   █████████████████████████████░   100%   update 1,499 / 1,500
  epoch      3 of 3
  loss       2.17 █▃▃▃▂▂▂▂▁▁▁▁ 0.00
  elapsed    8h 29m        eta  ~0h
  checkpoint update 1475 · 7 min ago

  node       e2e-node · serving 0.0 req/s alongside training
  gaps       90% idle → training · longest piece 0.09 s

  ✓ banking77-20261029-184325 succeeded in 8h 29m

  adapter    qwen3.5-9b:banking77-20261029-184325 · 62 MB · stored on hub

Trained in the gaps of a 16 GB Mac.

The real output of the overnight run of 29 to 30 October 2026. This overnight run served no chat traffic, so it trained the whole time.

Read the paper

Bring your labelled data

A labelled dataset of a few hundred rows or more, one example per line.

Getting started

One example

When a customer writes My card has not arrived, the answer is card_not_arriving.

Train

uli checks every row before uploading and lists any line that breaks the format.

Minutes to hours, depending on the size of your data.

Grade

Grade your model, then the base model given every label, on the same held-out rows.

Held-out rows never go in the training file.

It trains between requests

It trains a LoRA adapter on the base model the engine already holds in memory, using the idle time between requests, and then serves that adapter from the same machine.

Read the paper

Held-out questions, 77-intent banking task

Trained adapter, from the question alone90 of 100
Same base model, given every label69 of 100

A model that knows your labels

On a 77-intent banking task, the trained adapter answered 90 of 100 held-out questions from the question alone. The same base model, given every label, answered 69.

See the results

What it unlocks

Work for smaller Macs.

The whole loop, from training to serving, runs on a 16 GB M1 Pro.

Training that stays out of the way.

In the serving test, training added at most 0.134 s to a waiting request (p99, at 0.2 requests per second).

A job that survives a crash.

Killed mid-job, it resumed from its last checkpoint and finished with an identical adapter. The training data is deleted from the Mac when the job ends.

Work beyond inference

Serving a generic model is a race on price and speed, and large data centers win it. A model trained on one customer's labels is a different kind of work. Its value comes from that customer's data, not from scale, so it suits a Mac that already holds the base model and has time to spare.

Built on Darkbloom's engine

uli runs inside EngineV2, the open-source inference engine for Apple Silicon that Darkbloom runs on its providers' Macs, published under the MIT license in Layr-Labs/mlx-swift-lm. uli adds an idle hook and the training loop. It is an independent research project, not an official Darkbloom product, and it uses no code from Darkbloom's provider app.

Where it stands

An ongoing research project, measured so far on one Mac and one base model, Qwen3.5-9B in 4-bit. The paper gives every number with its source and its limits.

Read the paper

uli is in private alpha

The first cohort is free and starts with the Darkbloom community. Testers train a model on their own data and use it, on a Mac uli runs. Getting started shows every step, and how your data is handled says what is stored and when it is deleted.

Sign-ups close 20 November.