convaiinnovations/Laya Multilingual 322M

• License: Apache-2.0
• Release date: Not documented in the original card; 2026-09-19 is the first HF publication (repository creation).
• Modalities: Input: text; Output: typed decisions (choice, score, noul)
• Original model: convaiinnovations/laya-multilingual

Description

Laya Multilingual is a non-autoregressive decision model built on mmBERT-base with a trained decision head. It reads a state and typed questions and returns option probabilities in one forward pass, without text generation. The checkpoint covers more than 100 languages and is the family's alternative to the English ModernBERT checkpoint for non-English inputs.

Run

self serve laya-multilingual:322m

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": "Me cobraron dos veces por el mismo pedido.",
		"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: q8. Sizes and SHA-256 hashes are pinned in the model registry.

q4

File	Role	Size	SHA-256	Origin
laya_multilingual_ud_q4_k_m.gguf	model	499.6 MiB (523916512 bytes)	7c0e47f3dfaf1242929d69e91afb8868fcc5ee78092e7784e1aa2ee85adc9840	mys/laya-multilingual-GGUF

Total download: 499.6 MiB (523916512 bytes).

q8 (default)

File	Role	Size	SHA-256	Origin
laya_multilingual_q8_0.gguf	model	345.0 MiB (361712736 bytes)	757c1a4b1f0f41824113dde76d6cd06b0881d37b796103a338de77f9f9b935c3	mys/laya-multilingual-GGUF

Total download: 345.0 MiB (361712736 bytes).

Notes

• Text only; up to 16 options per question. Language coverage does not imply equal accuracy, especially for low-resource languages.
• The ggmlc-compiled graph has a 1024-token context budget; context and option limits cannot be changed at runtime. Upstream Python's 8192-token override is not an adapter setting here.
• Upstream reports overconfidence, weak zero-shot typed-workflow accuracy, position bias in ordinal scores and under-reported noul positives. Validate outputs on your own data.
• This server loads one checkpoint; it does not automatically route English inputs to the English model.
• The registered q4 file is larger than q8; sizes reflect the actual compiled artifacts, not a conventional quantization size ordering.
• Download size is not the total runtime memory requirement.
• Origin links use main because these registry entries pin hashes, not repository revisions.

Benchmarks

No submitted benchmark reports yet.