Resolves context window and output-reserve token counts for a given model using the model registry.
Attributes
- Companion
- object
- Graph
-
- Supertypes
-
class Objecttrait Matchableclass Any
Members list
Value members
Concrete methods
Looks up context-window size and output-token reserve for a model, falling back to provided defaults.
Looks up context-window size and output-token reserve for a model, falling back to provided defaults.
Value parameters
- defaultContextWindow
-
fallback maximum context-window size in tokens
- defaultReserve
-
fallback maximum output-token reserve
- fallbackResolver
-
function returning (contextWindow, reserve) when the registry has no entry
- logPrefix
-
optional string prepended to log messages for disambiguation
- lookupProviders
-
ordered list of provider names to try when querying the registry
- modelName
-
name of the model whose limits should be resolved
Attributes
- Returns
-
pair of (contextWindow, reserveTokens) for the resolved model
The context window (input limit) the registry holds for modelName among provider's own entries, and nothing else.
The context window (input limit) the registry holds for modelName among provider's own entries, and nothing else.
resolve is deliberately forgiving: the registry's own lookup falls back to a substring match over every provider's entries, and to the bare model name, so a name it does not know under provider can still pick up another provider's entry. That suits a dedicated client, whose model names are its vendor's own. It does not suit an endpoint that serves arbitrary names, where another provider's window is a wrong answer and no answer is better. This method is the strict form: an entry counts only when its provider is provider and its id is provider/modelName or modelName (case-insensitively); a partial name, another provider's entry for the same name, and an entry with no input limit all give None.
Value parameters
- modelName
-
the model id as the provider's API names it, without the
provider/prefix - provider
-
the registry provider, e.g.
groqortogether_ai
Attributes
- Returns
-
the entry's input limit when it is a positive number, otherwise
None