tensorflow/mesh

Health v2 · maintenance only

Flags: archived

Adoption not part of the score

1630 stars · 255 forks observed · 2026-08-28

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

Mesh TensorFlow is a Python library and embedded language for specifying distributed tensor computations, letting users define how model dimensions are split across a mesh of processors. It is implemented as a layer over TensorFlow and supports model parallelism, data parallelism, and mixed strategies for training and inference.

Use cases

• train a model whose parameters don't fit on a single device
• split a large language model across multiple GPUs or TPUs
• run model parallelism with data parallelism combined
• perform low-latency parallel inference at batch size 1
• train models with activations too large for one device
• specify custom tensor sharding layouts across a processor mesh

When to choose

• you need model parallelism for very large models in TensorFlow
• you want declarative control over how tensors are split across devices
• you are training huge language models on TPU/GPU meshes

When to avoid

• you only need simple data-parallel training
• you use PyTorch or JAX instead of TensorFlow
• you want actively developed tooling - the project is in maintenance mode

Facets

library · maturity maintenance

machine-learning deep-learning llm-training gpu-computing machine-learning deep-learning large-language-models microservices python model-parallelism distributed-training tensorflow tensor-sharding data-parallelism gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="tensorflow/mesh")