convaiinnovations/Laya Typed Decisions 421M

• License: Apache-2.0
• Release date: Not documented in the original card; 2026-09-18 is the first HF publication (repository creation).
• Modalities: Input: text; Output: typed decisions (choice, score, noul)
• Original model: convaiinnovations/laya-typed-decisions

Description

Laya Typed Decisions fine-tunes the English ModernBERT-large decision model on the typed-decisions benchmark's training split. It specializes in four workflows: agent-trace observability, customer service, invoice processing and security incidents. Like the base Laya model, it returns typed answers and probabilities without generating text.

Run

self serve laya-typed-decisions:421m

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 upstream laya daemon through the ggmlc-laya adapter. These are startup settings, not per-request API fields.

Parameter	Default	What it controls
threads	4	Number of CPU workers; accepts integers from 1 to 1024.
cuda_graph	true for auto / CUDA devices; otherwise false	Capture a CUDA graph for the live request shape.

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
laya_typed_decisions_ud_q4_k_m.gguf	model	404.0 MiB (423581888 bytes)	42919cae9c11f744a16b8f3136a3356c586a558108080050e832ba14bbb0010c	mys/laya-typed-decisions-GGUF

Total download: 404.0 MiB (423581888 bytes).

q8

File	Role	Size	SHA-256	Origin
laya_typed_decisions_q8_0.gguf	model	434.1 MiB (455179648 bytes)	6eb6ef58bc4f99bd9db40381604108af782ffc314aa00da6641f429991ca1f18	mys/laya-typed-decisions-GGUF

Total download: 434.1 MiB (455179648 bytes).

Notes

• Text only; English only; up to 16 options per question.
• The ggmlc-compiled graph has a 1024-token context budget; context and option limits cannot be changed at runtime.
• This is a specialist for four synthetic workflows, not a general zero-shot upgrade. Behavior elsewhere can be worse than the base model.
• Upstream reports overconfidence and temperatures fitted on training data; treat confidence as uncalibrated until validated on held-out domain data. Temperature fitting is not exposed by this adapter.
• Download size is not the total runtime memory requirement.
• Origin links use main because these registry entries pin hashes, not repository revisions.

Quality benchmarks

Typed Decisions: Publisher-reported

convaiinnovations/laya-typed-decisions

Publisher-reported specialist fine-tuned on train; official test split, 400 cases and 2000 decisions. Not comparable to zero-shot models. Measures synthetic teacher agreement, not correctness; not a local quantization result.

Dataset	Metric	Score
LocalLLaMA/typed-decisions (test)	Teacher agreement	0.77
LocalLLaMA/typed-decisions (test)	Brier	0.06
LocalLLaMA/typed-decisions (test)	ECE	0.21

Benchmarks

No submitted benchmark reports yet.