Jittor/jittor

Health v2 · maintenance only

Adoption not part of the score

3229 stars · 324 forks observed · 2026-08-28

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

Jittor is a high-performance deep learning framework from Tsinghua University based on just-in-time (JIT) compilation and meta-operators, with a Python front-end and CUDA/C++ back-end. It includes an integrated op compiler and tuner plus model libraries for vision, detection, segmentation, GANs, point clouds, and LLM inference.

Use cases

• train neural networks with a pytorch-like api
• run deep learning models with jit-compiled optimized ops
• run large language model inference on limited hardware
• train image segmentation and detection models
• do differentiable rendering research
• run deep learning on gpu with automatic op tuning

When to choose

• you want a pytorch-style dynamic graph framework with jit-compiled performance
• you need to run LLMs on modest hardware via JittorLLMs
• you want to customize operators without writing low-level code
• you work in computer vision research areas covered by its model libraries

When to avoid

• your team and ecosystem depend on pytorch or tensorflow
• you need broad multi-vendor hardware support beyond cuda/cpu
• you require a large community and third-party library ecosystem
• you need windows-first or mobile deployment support

Facets

framework · maturity active

deep-learning machine-learning llm-inference gpu-computing deep-learning machine-learning computer-vision gpu-computing python cpp jit-compilation meta-operators pytorch-alternative tsinghua model-zoo differentiable-rendering linux gpu cuda

Member repositories

For agents

markdown · JSON · MCP: product_card(name="Jittor/jittor")