Mapika/Decider 4B

• License: Apache-2.0
• Release date: 2026-09-24 (v2.1, the registered weights; v1 was released 2026-09-22).
• Modalities: Input: text; Output: typed decisions (choice, score, noul)
• Original model: Mapika/decider-4b

Description

Decider 4B reads a state and typed questions and returns distributions over explicit options without decoding text. The Qwen3.5-4B-Base model is trained for knowledge questions and harder judgments in two supervised stages; the registered v2.1 weights merge a LoRA trained on harder decisions and replay. There is no reinforcement-learning stage.

Run

self serve decider:4b

Here is a link to the API docs: Decision API.

Example Request

With the server running, open another terminal and send a request:

curl http://localhost:8080/v1/systemone \
	-H 'Content-Type: application/json' \
	-d '{
		"state": "I was charged twice for the same order.",
		"questions": {
			"department": {
				"type": "choice",
				"instructions": "Which department should handle this support ticket?",
				"criteria": {
					"billing": "Payments and refunds",
					"technical": "Technical problems"
				}
			}
		}
	}'

The server returns a JSON decision for the department question.

Engine Parameters

Served by the llama.cpp-based ggmlc-custom-decider engine and adapter. These are startup settings, not per-request API fields.

Parameter	Default	What it controls
context_size	8192	Context tokens including the prompt; accepts integers from 512 to 262144.
threads	llama.cpp default	CPU workers; accepts integers from 1 to 1024.
temperature	1.0	Softmax temperature on option-letter logits, not text sampling; accepts 0.01 to 100.
gpu_layers	-1 (all); 0 on CPU devices	Layers offloaded to the GPU; accepts integers from -1 to 10000.
flash_attn	auto	Flash attention mode: auto, on, or off.
image_min_tokens	Unset (projector metadata)	Minimum image tokens, 1 to 65536; no effect on this text-only model.
image_max_tokens	Unset (projector metadata)	Maximum image tokens, 1 to 65536; no effect on this text-only model.

Override startup settings with --set <parameter>=<value>.

Quantizations and Files

Default variant: q4. Sizes and SHA-256 hashes are pinned in the model registry.

q4 (default)

File	Role	Size	SHA-256	Origin
decider-4b-v2.1-Q4_K_M.gguf	model	2.52 GiB (2708804640 bytes)	c7083fcfc93f650cd66caeade4a3840ef9eeeeb1e80920907c554f70619d7c56	Mapika/decider-4b-GGUF

Total download: 2.52 GiB (2708804640 bytes).

q8

File	Role	Size	SHA-256	Origin
decider-4b-v2.1-Q8_0.gguf	model	4.17 GiB (4482403360 bytes)	74614659ab849bc0ec4f0431e1f84ae8bc11fe30b42c3a235a8b0049d36cbc3d	Mapika/decider-4b-GGUF

Total download: 4.17 GiB (4482403360 bytes).

Notes

• Text only; up to 10 options per question in this engine.
• The 4B name is the upstream size class; the original card reports approximately 4.2B parameters.
• The engine uses one temperature (default 1.0), not upstream v2.1's per-type calibration map. Upstream probabilities and benchmark results are not guaranteed to match.
• Upstream reports overconfidence on hard multi-step decisions and regressions in form filling and some games; validate on your own data before trusting confidence.
• English is the main training language; multilingual coverage is limited.
• Download size is not the total runtime memory requirement.
• Origin links use main because these registry entries pin hashes, not repository revisions.

Quality benchmarks

Decision Index: #80 (40.33)

Mapika/decider-4b

Published result for the named upstream model and evaluated configuration; checkpoint revision and local quantization equivalence are not guaranteed.

Evaluated: decider-4b

Edition: Decision Index 0.3.2

Checked 2026-10-13

Benchmarks

No submitted benchmark reports yet.