mogan

Checkpoints	Inputs
moganai/mogan-decision-31B-it	text

mogan is a LoRA adapter on google/gemma-4-31B-it. Flide reads the base checkpoint and runs the adapter beside its weights, as peft does, with no extra package.

Flide asks it with the release's own prompt. Each question is a chat prompt of its own. The system turn asks for one decision from the state, the question and the options, treats the state as data, tells the model to follow the question's evidence requirements, and asks for exactly one option letter, or one option code when the question has more than 26 options, with no explanation. The user turn holds the state written as JSON, the question's instructions, and one line per option, lettered A, B, C and so on. A choice shows each option's description, or its name when it has none. A noul shows the request's descriptions of its two outcomes, or "False" and then "True".

The answer is the softmax over the option letters at the first token of the reply. After Z the labels continue as two-letter codes. The temperature of each question type comes from temperatures.json in the checkpoint. The state comes first in every prompt, so Flide computes it once and forks it into the questions. Thinking is off.

Speed

On one H200, one request at a time, Flide answers 1.4 to 8.5 times faster than mogan's own di_gemma_engine on transformers 5.17 with peft. 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	68.0 ms	106.5 ms	1.6×
RACE	50.8 ms	70.3 ms	1.4×
MultiRC	396.0 ms	3378.1 ms	8.5×
QuALITY long	505.7 ms	902.0 ms	1.8×

Start a server

flide serve --model moganai/mogan-decision-31B-it

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. A choice takes up to 255 options. The longest prompt is 32768 tokens, and the state is not cut; a longer prompt is refused. A request with images is refused.

The template also refuses a question without instructions and a question with fewer than two options.

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