LlamaMBS.MaxDevices as UInt64

Type	Topic	Plugin	Version	macOS	Windows	Linux	iOS	Targets
property	Llama	MBS Tools Plugin	25.5	✅ Yes	✅ Yes	✅ Yes	✅ Yes	All

Returns the maximum number of hardware devices (e.g. GPUs or accelerators) that the runtime can utilize.

May return 1 for CPU-only configurations or higher if multi-GPU support is available.
(Read only property)

LlamaMBS.MaxParallelSequences as UInt64

Type	Topic	Plugin	Version	macOS	Windows	Linux	iOS	Targets
property	Llama	MBS Tools Plugin	25.5	✅ Yes	✅ Yes	✅ Yes	✅ Yes	All

Returns the maximum number of sequences that can be processed in parallel during inference.

Useful for determining batching or parallel decoding strategies in server or multi-user environments.
(Read only property)

LlamaMBS.MaxTensorBufTOverrides as UInt64

Type	Topic	Plugin	Version	macOS	Windows	Linux	iOS	Targets
property	Llama	MBS Tools Plugin	26.1	✅ Yes	✅ Yes	✅ Yes	✅ Yes	All

Maximum tensor buft overrides.

Should be 4096.
(Read only property)

LlamaMBS.SupportsGPUOffload as Boolean

Type	Topic	Plugin	Version	macOS	Windows	Linux	iOS	Targets
property	Llama	MBS Tools Plugin	25.5	✅ Yes	✅ Yes	✅ Yes	✅ Yes	All

Returns whether the current build supports GPU acceleration/offloading.

Returns true if GPU offloading is available, false otherwise.

Enabled if compiled with CUDA, Metal, or similar GPU backends.
(Read only property)

LlamaMBS.SupportsMLock as Boolean

Type	Topic	Plugin	Version	macOS	Windows	Linux	iOS	Targets
property	Llama	MBS Tools Plugin	25.5	✅ Yes	✅ Yes	✅ Yes	✅ Yes	All

Checks if the runtime supports mlock, which allows locking memory pages to avoid swapping them to disk.

Returns true if mlock is supported and enabled, false otherwise.

Locking model weights in RAM can improve performance consistency and prevent paging.
(Read only property)

LlamaMBS.SupportsMMap as Boolean

Type	Topic	Plugin	Version	macOS	Windows	Linux	iOS	Targets
property	Llama	MBS Tools Plugin	25.5	✅ Yes	✅ Yes	✅ Yes	✅ Yes	All

Checks if the current build/runtime supports memory-mapped model loading.

Returns: true if mmap is supported, false otherwise.

mmap allows efficient loading of large models without reading them entirely into memory.
(Read only property)

LlamaMBS.SupportsRPC as Boolean

Type	Topic	Plugin	Version	macOS	Windows	Linux	iOS	Targets
property	Llama	MBS Tools Plugin	25.5	✅ Yes	✅ Yes	✅ Yes	✅ Yes	All

Indicates whether the runtime supports RPC (Remote Procedure Call) functionality.

Returns true if RPC support is compiled in, false otherwise.

Useful for distributed inference setups where parts of the model or computation may be hosted remotely.
(Read only property)

LlamaMBS.TimeUS as Int64

Type	Topic	Plugin	Version	macOS	Windows	Linux	iOS	Targets
property	Llama	MBS Tools Plugin	25.5	✅ Yes	✅ Yes	✅ Yes	✅ Yes	All

Returns the current time in microseconds since an unspecified starting point (usually process start or epoch, depending on implementation).

Useful for benchmarking or profiling model performance.
(Read only property)

Some examples using this property:

• /Tools/Llama/LLama Simple
• /Tools/Llama/LLama Simple console