capacity-profile.schema.json
artificial-intelligence/large-language-models/models/capacity-profile.schema.json · 47 line(s) · json
{
"$schema": "http://json-schema.org/draft-07/schema#",
"$id": "https://sleela.dev/ai/llm/models/capacity-profile-1.0.json",
"title": "SLeeLa LLM model-capacity profile (version 1.0)",
"description": "A declarative, USER-OWNED description of the model the caller intends to run, within a supported envelope of ~2 GB to 16 GB of weights. SLeeLa ships no weights and selects no model; this records the user's choice so the reasoning layer can scale its budget. The envelope bounds are advisory.",
"type": "object",
"additionalProperties": false,
"required": ["schema_version", "model_id", "weight_bytes"],
"properties": {
"schema_version": { "const": "1.0" },
"model_id": {
"type": "string",
"minLength": 1,
"description": "The caller's identifier for THEIR model. Not a model shipped by SLeeLa."
},
"weight_format": {
"type": "string",
"description": "Informational, e.g. 'gguf-q4_k_m', 'safetensors-fp16'."
},
"weight_bytes": {
"type": "integer",
"minimum": 1,
"description": "Declared weight size in bytes. The supported envelope is 2 GiB (2147483648) to 16 GiB (17179869184); values outside are accepted with an advisory warning."
},
"tier_hint": {
"type": "string",
"enum": ["compact", "standard", "capable", "large"],
"description": "Optional; the native layer derives the tier from weight_bytes. If present it should agree with that derivation."
},
"context_tokens": {
"type": "integer",
"minimum": 0,
"description": "0 => derive from tier. A nonzero value overrides the tier default."
},
"ram_budget_bytes": {
"type": "integer",
"minimum": 0,
"description": "Host memory the user allots; 0 => unspecified."
},
"accelerated": {
"type": "boolean",
"description": "True if a GPU/accelerator backend is used."
},
"note": { "type": "string" }
}
}