A wrapper around a compiled graph object that adds a modifiable context that BNNS requires to support dynamically sized models and set execute-time options.
Synchronously sets the batch size for a graph.
This is a special case of `setDynamicShapes(_:forFunction:)` where the only dynamic sizes that occur are the first index
of their tensor (that is, the batch dimension) and are all equal. This allows just passing a single `batchSize` constant.
This function has no effect if you pass a `batchSize` value that's less than zero.
- Parameter batchSize: The batch size.
- Parameter function: The specific function to set shapes for. You may set this to `nil` if there is only one function.
- SeeAlso: `BNNSGraphContextSetBatchSize`
Sets the batch size for a graph.
This is a special case of `setDynamicShapes(_:forFunction:)` where the only dynamic sizes that occur are the first index
of their tensor (that is, the batch dimension) and are all equal. This allows just passing a single `batchSize` constant.
This function has no effect if you pass a `batchSize` value that's less than zero.
- Parameter batchSize: The batch size.
- Parameter function: The specific function to set shapes for. You may set this to `nil` if there is only one function.
- SeeAlso: `BNNSGraphContextSetBatchSize`
method · iOS 18.0+
argumentCount
TypeScript
argumentCount(function_?: (string) | null): number
Swift declaration
func argumentCount(forFunction function: String? = nil) -> Int
The number of arguments for a graph function.
- Parameter function: The specific function to query. You may set this to `nil` if there is only one function.
- SeeAlso: `BNNSGraphGetArgumentCount`
The argument names of outputs and inputs to a graph function.
- Parameter function: The specific function to query. You may set this to `nil` if there is only one function.
- SeeAlso: `BNNSGraphGetArgumentNames`
get · iOS 18.0+
functionCount
TypeScript
get functionCount(): number
Swift declaration
var functionCount: Int { get }
The number of callable functions in this graph.
- Parameter function: The specific function to query. You may set this to `nil` if there is only one function.
- SeeAlso: `BNNSGraphGetFunctionCount`
get · iOS 18.0+
functionNames
TypeScript
get functionNames(): Array<string>
Swift declaration
var functionNames: [String] { get }
The names of callable functions in the graph.
- SeeAlso: `BNNSGraphGetFunctionNames`
Returns an allocated tensor for a given function argument.
- Parameter function: The specific function to create the tensor for. You may set this to `nil` if there is only one function.
- Parameter argument: The argument to create the tensor for.
- Parameter fillKnownDynamicShapes: if `true`, the function replaces any dynamic shapes
with shapes that will be used on next execution of `Context`, either drawn from default shapes in the `.mlmodelc`
function or supplied by a preceding call to
`setDynamicShapes(_:forFunction:)` or `setBatchSize(_:forFunction:)`.
Otherwise the function represents dynamically-sized dimensions with a value of `-1` in either shape or stride as appropriate.
- Note: This function supports `Float`, `Float16`, `Bool`, and `Int32` datatypes only and returns `nil` for other types.
- SeeAlso: `BNNSGraphContextGetTensor`
Synchronously returns a new context that wraps a graph object which represents the compiled `.mlmodelc` file.
- Parameter path: The path to the `.mlmodelc` program file.
- Parameter functionName: The name of a specific function to be compiled. Pass `nil` to specify that the function compiles all functions.
- Parameter options: The compilation options.
- SeeAlso: `BNNSGraphCompileFromFile`
- SeeAlso: `BNNSGraphContextMake`
Returns a new context that wraps a graph object which represents the compiled `.mlmodelc` file.
- Parameter path: The path to the `.mlmodelc` program file.
- Parameter functionName: The name of a specific function to be compiled. Pass `nil` to specify that the function compiles all functions.
- Parameter options: The compilation options.
- SeeAlso: `BNNSGraphCompileFromFile`
- SeeAlso: `BNNSGraphContextMake`
Synchronously executes the specified function with the provided context.
If the underlying model contains dynamic shaped inputs or outputs, these must be set prior to calling this routine through a call
to either `setDynamicShapes(_:forFunction:)` or `setBatchSize(_:forFunction:)`.
Don't modify the shapes again until this routine has returned.
- Parameter function: The specific function to execute. You may set this to `nil` if there is only one function.
- Parameter arguments: The output and input arguments.
The same context must only be used by a single thread at a time.
- SeeAlso: `BNNSGraphContextExecute`
Omit the output parameters from the call. The result contains $ret and $out, with arguments inside $out.
Executes the specified function with the provided context.
If the underlying model contains dynamic shaped inputs or outputs, these must be set prior to calling this routine through a call
to either `setDynamicShapes(_:forFunction:)` or `setBatchSize(_:forFunction:)`.
Don't modify the shapes again until this routine has returned.
- Parameter function: The specific function to execute. You may set this to `nil` if there is only one function.
- Parameter arguments: The output and input arguments. Note that the arguments may not be in the same order
as the original `mlpackage` or `mlmodelc` file. Use the `argumentPosition(forFunction:argument:)` to
get the correct position in the `arguments` array for a given argument.
The same context must only be executed by a single thread at a time.
- SeeAlso: `BNNSGraphContextExecute`
Omit the output parameters from the call. The result contains $ret and $out, with arguments inside $out.
method · iOS 18.0+
argumentPosition
TypeScript
argumentPosition(function_: (string) | null, argument: string): number
Swift declaration
func argumentPosition(forFunction function: String? = nil, argument: String) -> Int
Returns the position in the `arguments` array for a call to `executeFunction(_:arguments:)` for the given argument name.
- Parameter function: The specific function to query. You may set this to `nil` if there is only one function.
- Parameter argument: The input or output argument name that the function queries.
- SeeAlso: `BNNSGraphContextGetArgumentPosition`
Specifies the dynamic shapes for a graph and infers (if possible) output shapes.
- Parameter shapes: The array of shapes for outputs and inputs, in the same order you pass to `executeFunction(_:arguments:)`.
- Parameter function: The specific function to set shapes for. You may set this to `nil` if there is only one function.
On entry, the function reads the input shapes from this array, unless indicated otherwise by setting `shapes[idx].rank = 0`, in which
case the constant or default value from the `.mlmodelc` source is used. If you specify a non-zero value that does not match a constant
size in the `.mlmodelc` source, the function generates an error and the output shapes are not read.
On exit, the function returns output shapes that have `shapes[idx].rank != 0` to an upper bound on the expected output shape.
If the function can't deduce the output shape as it depends on input data values, the value of `shapes[idx].size[d]` is set
to zero for that dimension.
get · iOS 26.0+
streamingAdvanceCount
TypeScript
get streamingAdvanceCount(): number
Swift declaration
var streamingAdvanceCount: Int { get set }
Sets streaming advancement amount for cases with dynamically shaped inputs.
- Parameters:
- context: The graph context.
- advance_count: An integer value that specifies the number of elements that the function advances the internal state pointer.
For models compiled with the `BNNSOption` attribute `StateMode=Streaming` enabled, where `slice_update` operations use an update parameter of dynamic shape, BNNS can't unambigiously determine the streaming advancement size. In this case, call this function before calling ``BNNSGraphContextExecute`` to set the advancement size for each frame.
This function advances the internal state pointer by `advance_count` elements in the streaming dimension before returning from ``BNNSGraphContextExecute``.
> Note: The BNNS streaming APIs do not support models that require different advancement amounts for different states.
set · iOS 26.0+
streamingAdvanceCount
TypeScript
set streamingAdvanceCount(value: number)
Swift declaration
var streamingAdvanceCount: Int { get set }
Sets streaming advancement amount for cases with dynamically shaped inputs.
- Parameters:
- context: The graph context.
- advance_count: An integer value that specifies the number of elements that the function advances the internal state pointer.
For models compiled with the `BNNSOption` attribute `StateMode=Streaming` enabled, where `slice_update` operations use an update parameter of dynamic shape, BNNS can't unambigiously determine the streaming advancement size. In this case, call this function before calling ``BNNSGraphContextExecute`` to set the advancement size for each frame.
This function advances the internal state pointer by `advance_count` elements in the streaming dimension before returning from ``BNNSGraphContextExecute``.
> Note: The BNNS streaming APIs do not support models that require different advancement amounts for different states.
get · iOS 18.0+
checkForNaNsAndInfinities
TypeScript
get checkForNaNsAndInfinities(): boolean
Swift declaration
var checkForNaNsAndInfinities: Bool { get set }
Enables debug mode, checks intermediate tensors for nans and infinities.
> - Not for use in production code.
- SeeAlso: `BNNSGraphContextEnableNanAndInfChecks`
set · iOS 18.0+
checkForNaNsAndInfinities
TypeScript
set checkForNaNsAndInfinities(value: boolean)
Swift declaration
var checkForNaNsAndInfinities: Bool { get set }
Enables debug mode, checks intermediate tensors for nans and infinities.
> - Not for use in production code.
- SeeAlso: `BNNSGraphContextEnableNanAndInfChecks`
Returns an unallocated tensor for a given function argument.
- Parameter function: The specific function to create the tensor for. You may set this to `nil` if there is only one function.
- Parameter argument: The argument to create the tensor for.
- Parameter fillKnownDynamicShapes: if `true`, the function replaces any dynamic shapes
with shapes that will be used on next execution of `Context`, either drawn from default shapes in the `.mlmodelc`
function or supplied by a preceding call to
`setDynamicShapes(_:forFunction:)` or `setBatchSize(_:forFunction:)`.
Otherwise the function represents dynamically-sized dimensions with a value of `-1` in either shape or stride as appropriate.
- SeeAlso: `BNNSGraphContextGetTensor`
Skipped declarations
1 declarations were excluded from these bindings. Each reason comes from the generator or the iOS SDK compiler.
BNNSGraph.Context.executeFunction(_:arguments:)
This declaration needs Swift generic type arguments. (no canonical witness for T)
