<-Reference Software Go to ToC
This chapter is informative.
1 Implementations
1.1 A resource-constrained Implementation
- The Controller is a single process, or a library function on a computing board, that runs the Module and operates on interrupt call-backs.
- The AIMs are in the Controller’s process; the Module is built from AIM Instance Metadata held locally.
- The Controller is reached through a secure communication interface.
- The User Agent acts on the Module at the request of its user.
1.2 An Implementation that is not resource-constrained
- The Controller obtains the AIM Instance Metadata of the Module, and the Implementations it names, from the MPAI Store, after the two have authenticated each other, and verifies them.
- The Controller instantiates the AIMs: in its own process, in other processes, or on AIM hosts on other machines, which it admits by the Trust Protocol.
- The Controller establishes the Channels, runs the Module by exchange or continuously, and keeps its Trace.
- User Agents drive the Module through the Controller API, locally or remotely over MPAI-MAS.
2 Examples of types
Table 1 – Examples of types
| Type | Meaning |
| byte[] bitstream_t | An array of bytes of variable length. |
| {int32 frameNumber; int16 x; int16 y; byte[] frame} frame_t | A struct_type with four members, frameNumber, x, y and frame: an int32, an int16, an int16 and an array of bytes of variable length. |
| {int32 i32 | int64 i64} variant_t | A variant_type that is either an int32 or an int64. |
3 Example of AIM Metadata
The AIM Metadata of Automatic Speech Recognition (MMC-ASR-V2.5) of MPAI-MMC: a Basic AIM whose Ports are identified by Data Type and, where a Data Type recurs on one side, Port Number. The fields that an AIM Instance Metadata fills in are shown as comments.
{ "Identifier": {"ImplementerID": "IIDRA-assigned", "ImplementationID": "/* String from Implementer */", "AIMName": "MMC-ASR-V2.5"},
"Header": "MMC-ASR-V2.5",
"APIProfile": "Basic",
"Description": "Automatic Speech Recognition: receives an Audio Object and provides a Text Object and Speech Descriptors.",
"InternalTypes": [ ],
"ExternalPorts": [{"Name": "LanguageSelector", "Direction": "Input", "DataType": "OSD-SEL-V1.5", "IsOptional": true, "Technology": "", "Protocol": "", "IsRemote": ""},
{"Name": "AuxiliaryText", "Direction": "Input", "DataType": "OSD-TXO-V1.5", "IsOptional": true, "Technology": "", "Protocol": "", "IsRemote": ""},
{"Name": "SpeakerID", "Direction": "Input", "DataType": "OSD-IID-V1.5", "IsOptional": true, "Technology": "", "Protocol": "", "IsRemote": ""},
{"Name": "SpeechOverlap", "Direction": "Input", "DataType": "MMC-SOL-V2.5", "IsOptional": true, "Technology": "", "Protocol": "", "IsRemote": ""},
{"Name": "RecognisedText", "Direction": "Output", "DataType": ["OSD-BTO-V1.5", "OSD-TXO-V1.5"], "PortNumber": 1, "Technology": "", "Protocol": "", "IsRemote": ""},
{"Name": "SpeechDescriptors", "Direction": "Output", "DataType": "MMC-SDO-V2.5", "IsOptional": true, "Technology": "", "Protocol": "", "IsRemote": ""},
{"Name": "ImageSpeech", "Direction": "Input", "DataType": "OSD-BSO-V1.5", "PortNumber": 1, "IsOptional": true, "Technology": "", "Protocol": "", "IsRemote": ""},
{"Name": "ResponseSpeech", "Direction": "Input", "DataType": "OSD-BSO-V1.5", "PortNumber": 2, "IsOptional": true, "Technology": "", "Protocol": "", "IsRemote": ""},
{"Name": "TextQuestion", "Direction": "Output", "DataType": "OSD-BTO-V1.5", "PortNumber": 2, "Technology": "", "Protocol": "", "IsRemote": ""}],
"SubAIMs": [ ],
"Topology": [ ],
"Implementations": [{ "ImplementationURI":"/* absolute URI locating the Implementation */",
"BinaryName": "/* provided by implementer */",
"Architecture": "/* x86-64 | ARM64 */",
"MASPort": 443,
"Firewall": [{ "Protocol": "TCP | UDP", "Version": "IPV4 | IPV6", "Direction": "Ingress | Egress | Both", "Range": "9990[-9999]", "Source": "0.0.0.0/0" }],
"OperatingSystem": "/* Windows | Linux | MacOS */",
"Version": "/* implementer version */",
"Source": "/* AIMStorage | MPAIStore */",
"Destination": "/* provided by implementer */" }],
"ResourcePolicies":[{ "Name": "Memory", "Minimum": "/* GB */", "Maximum": "/* GB */", "Request": "/* GB */" },
{ "Name": "CPU:Number", "Minimum": "/* integer, 0=all */", "Maximum": "/* integer, 0=all */", "Request": "/* integer */" },
{ "Name": "CPU:Class", "Minimum": "/* string */", "Maximum": "/* string */", "Request": "/* string */" },
{ "Name": "GPU:Number", "Minimum": "/* integer, 0=all */", "Maximum": "/* integer, 0=all */", "Request": "/* integer */" },
{ "Name": "GPU:CUDA:Class", "Minimum": "/* string */", "Maximum": "/* string */", "Request": "/* string */" },
{ "Name": "GPU:CUDA:FrameBuffer", "Minimum": "/* GB */", "Maximum": "/* GB */", "Request": "/* GB */" },
{ "Name": "GPU:CUDA:MemorySpeed", "Minimum": "/* GB/s */", "Maximum": "/* GB/s */", "Request": "/* GB/s */" } ],
"DataXMData": { },
"Documentation": [{"Type": "Tutorial", "URI": "https://mpai.community/standards/mpai-mmc/v2-5/"}]}
4 Standards using MPAI-AIF V3.0
The following MPAI Standards specify AIMs and Composite AIMs whose Metadata is AIM Metadata of MPAI-AIF V3.0:
Table 2 – Standards using MPAI-AIF V3.0
| Standard | Title |
| MPAI-MMC | Multimodal Conversation |
| MPAI-PAF | Portable Avatar Format |
| MPAI-OSD | Object and Scene Description |
| MPAI-CAV | Connected Autonomous Vehicle – Technologies |
| MPAI-AIH | AI for Health – Health Secure Platform |
An example of a Workflow is given in Workflow Description.
