Compiler and runtime

Why it matters

Connect model conversion, quantization, scheduling, and runtime control to accelerate deployment.

• Start from familiar workflows

  Common model formats and development flows are the starting point, lowering the barrier for your team.

• Hardware-aware optimization

  The compiler partitions, schedules, and places data according to SRAM-CIM and memory hierarchy characteristics.

• An observable runtime

  Task scheduling, resource management, and performance observability help teams locate issues and tune quickly.

How it works

1. Model import

   Import trained models and parse the compute graph and data types.

2. Quantization and adaptation

   Quantize and adapt operators to precision targets, then verify output quality.

3. Compile and schedule

   Map operations onto compute arrays and the memory hierarchy to produce an optimized execution plan.

4. Deploy and observe

   Deploy to the target platform through the runtime and keep monitoring and tuning performance.

What we align on during evaluation

Bring this information to your evaluation request so the VXI team can propose a concrete platform and adoption plan faster.

• Model frameworks, formats, and versions in use
• Accuracy tolerance and quantization preferences
• Operating system, host environment, and integration interfaces
• Team development workflow and update cadence

Frequently asked questions

Do I need to rewrite my models to adopt VXI?

Most models can be converted and adapted directly through the toolchain. For uncommon operators, VXI engineers propose an adaptation plan during evaluation.

Does quantization affect accuracy?

Quantization balances efficiency and precision. The VXI toolchain includes a quantization verification flow so you can confirm output quality meets product requirements before deployment.

What does the runtime provide?

The runtime handles task scheduling, resource management, multi-model execution, and performance observability, giving applications a consistent interface to VXI platforms.

Related technology modules

• SRAM-CIM compute path — Weight residency and near-memory compute that cut data movement
• System co-engineering — An integration path connecting model, software, host, and memory
• Memory hierarchy — Co-placed weights, activations, and cache