dmlc/nnvm

Health v2 · maintenance only

Flags: archived

Adoption not part of the score

1649 stars · 274 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

NNVM was a computation-graph intermediate representation and compiler front-end for deep learning frameworks, serving as the graph-level optimization and deployment layer of the TVM stack. Development has ended and all source code now lives in the TVM repository (apache/tvm).

Use cases

• compile deep learning models to GPU backends like CUDA, OpenCL, ROCm, and Metal
• optimize computation graphs before hardware-specific code generation
• deploy trained neural network models to diverse hardware targets
• convert models from frameworks like MXNet into a portable graph IR

When to choose

• you are maintaining legacy code that already depends on the NNVM graph IR
• you need to study the historical design of TVM's graph compiler

When to avoid

• starting any new project - use TVM (Relay/Relax) instead
• you need current model support, maintenance, or bug fixes
• you want an actively developed deep learning compiler

Facets

library · maturity abandoned

compiler machine-learning deep-learning llm-inference deep-learning compilers machine-learning windows cpp python computation-graph tvm compiler-stack model-deployment gpu opencl rocm metal merged-into-tvm linux macos cuda

Member repositories

For agents

markdown · JSON · MCP: product_card(name="dmlc/nnvm")