Linne

Checkpoints	Inputs
PNPM244/linne-fern-1-0	text

linne-fern-1-0 is a decision model on Qwen3.8-27B (it moved from MISA-core/linne-fern-1-0, which Flide also recognizes). Flide asks it with the release's own prompt. Each question is a chat prompt of its own with one user turn and thinking off. The turn holds the state as it is, or as compact JSON when it is not a string. A choice follows with "Question:" and the question, one line per option as "A: name - description" (the name alone when it has no description), and "Answer with the letter of one option only.". A noul reads "Is the following true?" with the question, lists the request's descriptions as "yes:" and "no:" lines, and ends with "Answer with yes or no only.".

The answer is the softmax at the first token of the reply over the option letters, A to Z, or two-letter labels when a choice has more than 26 options, or over yes and no, at temperature 1. The state comes first in every prompt, so Flide computes it once and forks it into the questions.

Speed

On one H200, one request at a time, Flide answers 1.2 to 8.6 times faster than the checkpoint's own linne_engine.py on transformers. Each workload is 32 requests, each run once on both sides before timing, and the table gives the median latency of a request. MultiRC requests ask about 49 questions each.

Workload	Flide	Release	Speedup
Decision Index	46.4 ms	108.1 ms	2.3×
RACE	39.4 ms	101.3 ms	2.6×
MultiRC	245.8 ms	2101.8 ms	8.6×
QuALITY long	343.3 ms	418.9 ms	1.2×

Start a server

flide serve --model PNPM244/linne-fern-1-0

Send a request

curl http://localhost:8000/v1/systemone -H "Content-Type: application/json" -d '{
  "state": "Order #1182 arrived with a cracked screen. The customer asks for a refund and attached two photos.",
  "questions": {
    "action": {"type": "choice", "instructions": "What should support do next?",
               "criteria": {"refund": "refund the order", "replace": "send a replacement", "escalate": "ask a human"}},
    "urgent": {"type": "noul", "instructions": "Does this need a reply within the hour?"}
  }
}'

Limits and options

A request takes choice and noul questions, and a score is refused. The longest prompt is 65535 tokens. A request with images is refused.

Option	What it does
--format template --format-arg spec=misa-core__linne-fern	reads a copy of the checkpoint kept in a directory Flide does not recognize