mys/laya-typed-decisions-GGUF

Laya Typed-Decisions GGUF (ggmlc)

Specialist System 1 decision model compiled from convaiinnovations/laya-typed-decisions (ModernBERT-large, 421M, context 1024). Fine-tuned for invoice match, SOC alerts, customer-service next action, and agent-trace harnesses.

These files are not llama.cpp / llama-cli GGUFs. They are produced by ggmlc, a neural network compiler that lowers PyTorch, JAX, Flax, and Keras models to high-performance GGML execution. Loading them in llama.cpp will fail.

Typed questions (choice / score / noul) are scored in one encoder pass. There is no autoregressive token generation.

Source, CLI, and binaries: examples/laya

Other families: laya-GGUF (English) · laya-multilingual-GGUF

Files

File	Quant	Size	Notes
laya_typed_decisions_f16.gguf	F16	~811 MB	Full precision.
laya_typed_decisions_q8_0.gguf	Q8_0	~434 MB	Smaller, still accurate.
laya_typed_decisions_ud_q4_k_m.gguf	UD_Q4_K_M	~404 MB	Smallest. 1D norms/biases stay F32.

English ModernBERT tokenizer ([CLS]/[SEP]/[PAD]/[MASK]). max_len=1024.

huggingface-cli download mys/laya-typed-decisions-GGUF laya_typed_decisions_f16.gguf --local-dir .

Run with laya

Download a binary from ggmlc releases (laya.exe / laya). --device defaults to auto (CUDA or Metal if present, else CPU).

--models-dir routing does not pick this family automatically unless --family typed-decisions or the question ids match a specialist workflow (invoice, security, customer_service, harness).

laya help
laya list-presets
laya info laya_typed_decisions_f16.gguf

laya decide laya_typed_decisions_f16.gguf --preset invoice --device auto --cuda-graph
laya decide laya_typed_decisions_f16.gguf --preset security --json
laya decide laya_typed_decisions_f16.gguf --preset customer_service
laya decide laya_typed_decisions_f16.gguf --preset harness
laya serve laya_typed_decisions_f16.gguf --port 8080 --device auto --cuda-graph
laya bench laya_typed_decisions_f16.gguf --preset invoice --device auto --cuda-graph

Force it from a mixed --models-dir:

laya decide --models-dir . --family typed-decisions --preset invoice

serve starts Decision Studio (GET /) and POST /api/decide. daemon is newline JSON-RPC on stdin/stdout.

What this is

Laya is the open reproduction of TypeSafe Jev: given a state and typed questions, it returns calibrated probabilities instead of generating tokens. This checkpoint is the specialist sibling of the general English model — same tokenizer family, longer context, trained on typed-decision workflows.

License

Apache 2.0, same as the upstream Laya weights. Compiler: ggmlc (MIT).