- Forums
- Product Forums
- General Purpose MicrocontrollersGeneral Purpose Microcontrollers
- i.MX Forumsi.MX Forums
- QorIQ Processing PlatformsQorIQ Processing Platforms
- Identification and SecurityIdentification and Security
- Power ManagementPower Management
- Wireless ConnectivityWireless Connectivity
- RFID / NFCRFID / NFC
- Advanced AnalogAdvanced Analog
- Neural Processing UnitsNeural Processing Units
- MCX Microcontrollers
- S32G
- S32K
- S32V
- MPC5xxx
- Other NXP Products
- S12 / MagniV Microcontrollers
- Powertrain and Electrification Analog Drivers
- Sensors
- Vybrid Processors
- Digital Signal Controllers
- 8-bit Microcontrollers
- ColdFire/68K Microcontrollers and Processors
- PowerQUICC Processors
- OSBDM and TBDML
- S32M
- S32Z/E
-
- Solution Forums
- Software Forums
- MCUXpresso Software and ToolsMCUXpresso Software and Tools
- CodeWarriorCodeWarrior
- MQX Software SolutionsMQX Software Solutions
- Model-Based Design Toolbox (MBDT)Model-Based Design Toolbox (MBDT)
- FreeMASTER
- eIQ Machine Learning Software
- Embedded Software and Tools Clinic
- S32 SDK
- S32 Design Studio
- GUI Guider
- Zephyr Project
- Voice Technology
- Application Software Packs
- Secure Provisioning SDK (SPSDK)
- Processor Expert Software
- Generative AI & LLMs
-
- Topics
- Mobile Robotics - Drones and RoversMobile Robotics - Drones and Rovers
- NXP Training ContentNXP Training Content
- University ProgramsUniversity Programs
- Rapid IoT
- NXP Designs
- SafeAssure-Community
- OSS Security & Maintenance
- Using Our Community
-
- Cloud Lab Forums
-
- Knowledge Bases
- ARM Microcontrollers
- i.MX Processors
- Identification and Security
- Model-Based Design Toolbox (MBDT)
- QorIQ Processing Platforms
- S32 Automotive Processing Platform
- Wireless Connectivity
- CodeWarrior
- MCUXpresso Suite of Software and Tools
- MQX Software Solutions
- RFID / NFC
- Advanced Analog
- Neural Processing Units
-
- NXP Tech Blogs
- Home
- :
- Translated Content
- :
- Multi Source Translation Content
- :
- 2122841_en-US
2122841_en-US
- Subscribe to RSS Feed
- Mark as New
- Mark as Read
- Bookmark
- Subscribe
- Printer Friendly Page
- Report Inappropriate Content
2122841_en-US
2122841_en-US
Hi Team,
I’m exploring the dm-eiq-genai-flow-demonstrator (GenAI Flow) on the i.MX95 platform working within a Yocto environment. Could you please clarify:
1. Which BSP version (e.g., L6.12.20_2.0.0) is minimum required to run the demo end-to-end?
2. Does this BSP include all necessary elements—such as updated Neutron delegate, TFLite runtime, and tooling—to support LLM inference (RAG)?
3. Are there any additional Yocto meta-layers or image recipes (for example meta-eiq- genai) recommended for this demo?
Thanks in advance for your insights!
Best regards,
Varshil Gandhi
Hi Varshil,
Thanks for your questions.
Let me try to answer them:
1. The package was initially made for Q1 BSP L6.12.3-1.0.0 for i.MX95 rev Ax.
2. This Q1 BSP does not directly include what's necessary to leverage Neutron on LLM inferences. That's why the package provides a meta-layer.
3. This meta-eiq-genai-flow meta-layer provides the possibility to add a Onnxruntime Neutron Execution Provider to accelerate the LLM operations on Neutron and free up some CPU resources, with associated patches/files for Kernel and Neutron Fw. Please note that we are using onnxruntime and not TF lite for this project.
That being said, Q2 BSP (e.g., L6.12.20_2.0.0) is being released this days. It supports the B0 revision of i.MX95 but without LLM acceleration on Neutron, only CPU mode is functional.
Thanks and best regards,
Pierre