Relevant Experience & Education Highlights

Brings deep hands-on expertise in MLIR dialect design, PyTorch compiler integration via Torch-MLIR, and building end-to-end compilation stacks for heterogeneous ML accelerators, directly aligning with extending Mythic's MLIR ecosystem and PyTorch infrastructure.

1. Direct, hands-on experience with MLIR, including dialect design, compiler passes, or lowering pipelines

Led MLIR-based compiler development to bridge high-level frameworks to hardware targets.

• Co-created and maintained Tensor Compute Primitives (mlir-tcp), an open-source mid-level MLIR dialect for ML programs, at a leading autonomous vehicle technology company.
• Co-maintained Torch-MLIR, providing first-class compiler support from PyTorch to MLIR ecosystems.
• Tech-led ML compiler frontend connecting state-of-the-art PyTorch models to MLIR for optimization toward heterogeneous compute platforms.
• Built MLIR-enabled compilation stack simplifying ML model deployments onto commodity and in-house ML accelerators.

2. Experience architecting complete MLIR flows: from frontend dialects down to hardware-aware dialects

Architected full-stack compiler infrastructures optimizing PyTorch workloads for production inference.

• Created GraFX, a source-to-source compiler framework accelerating PyTorch model development, optimization, and deployment lifecycles.
• Developed lowering pipelines from high-level PyTorch operations through MLIR middle-end to latency-critical accelerator targets.
• Worked at intersection of compilers and deep learning algorithms for accelerated ML deployments on autonomous systems.

3. Familiarity with PyTorch compiler technologies, especially Torch-MLIR and integration paths with PyTorch 2.0

Delivered seamless PyTorch-to-MLIR connectivity enabling efficient hardware targeting.

• Integrated PyTorch ecosystem with MLIR compiler passes for optimized inference on diverse accelerators.
• Led frontend development ensuring compatibility with advanced PyTorch models and compiler technologies.
• Developed advanced quantization algorithms like TQT for hardware-friendly fixed-point inference at a programmable logic company specializing in FPGAs and AI Engines.

4. Background in heterogeneous or specialized accelerators

Optimized ML workloads for varied compute platforms including in-house and commodity accelerators.

• Targeted compilation stacks for autonomous vehicle inference on heterogeneous platforms.
• Explored novel architectures and algorithms for efficient ML acceleration on FPGAs and AI Engines in cloud and edge use cases.
• Advanced algorithms for ML acceleration achieving high-accuracy quantized representations for constrained hardware.

Requirements & Candidate Alignment