decider-2b (Verdict-ready)

The decider-2b System One decision model (Qwen3.5-2B-Base, 1.9B parameters, 32768-token context, checkpoint revision v11 — the vendor's default), packaged for Verdict — a high-performance serving engine for System One decision models. Serve it with one command:

# GPU host (the torch runtime serves on CUDA in v0.2 — no CPU image)
docker run --rm --gpus all -p 8080:8080 \
    -v ~/.cache/verdict:/root/.cache/verdict \
    sgyaqing/verdict:latest-cuda serve verdict-models/decider-2b --host 0.0.0.0

# ModelScope:
docker run --rm --gpus all -p 8080:8080 \
    -e VERDICT_USE_MODELSCOPE=1 \
    -v ~/.cache/verdict:/root/.cache/verdict \
    sgyaqing/verdict:latest-cuda serve sgyaqing/decider-2b --host 0.0.0.0

Endpoints: the standard decision protocols POST /v1/systemone and POST /v1/decisions, native POST /v1/choice | /v1/score | /v1/noul, plus an OpenAI-compatible layer (/v1/chat/completions, /v1/models).

Files

File	Purpose
model.json	manifest: budgets, calibration temperatures, checkpoint declaration
tokenizer/tokenizer.json (+ tokenizer_config.json)	the tokenizer
SHA256SUMS	checksums for the files above
checkpoint (config.json, model.safetensors, generation_config.json)	not in this repo — fetched from Mapika/decider-2b on first serve

Requirements: a CUDA GPU and ~5.5 GB VRAM at load.

Provenance & fidelity

This package adds only the serving manifest. Serving runs the checkpoint in bf16 (its own precision) and is verified against the golden standard (official decider-ai 1.9.0, CPU fp32, 30-case corpus): zero answer flips on CUDA (RTX 4090).

Accuracy PK — against decider.serve (decider-ai 1.9.0); both servers on one RTX 4090 (CUDA), graded against the same golden standard:

	Verdict	decider.serve	result
worst deviation	1.24e-2	1.06e-2	tied

Performance

Measured on one RTX 4090, end to end over HTTP, against the model author's own server (decider.serve). A ratio above 1x means Verdict is ahead.

concurrency	Verdict rps	decider.serve rps	rps ratio	Verdict P50 / P99 (ms)	decider.serve P50 / P99 (ms)	P50 / P99 ratio
1	139.6	132.4	1.05x	6.91 / 7.88	7.51 / 7.94	1.09 / 1.01x
8	436.5	314.0	1.39x	18.23 / 20.17	25.41 / 26.87	1.39 / 1.33x
64	795.9	506.2	1.57x	80.13 / 85.43	112.91 / 436.79	1.41 / 5.11x

License

Apache-2.0, inherited from the base checkpoint. Credit for the model itself belongs to the original author and the Qwen3.5-2B-Base it builds on; this repository only packages the serving manifest.