facebookresearch/fairscale

Health v2 · maintenance only

Flags: archived no_license

Adoption not part of the score

3407 stars · 293 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

FairScale is a PyTorch extension library providing composable modules and APIs for high-performance, large-scale distributed training, including techniques like Fully Sharded Data Parallel (FSDP). Its FSDP implementation has been upstreamed to PyTorch, so this library now serves mainly as a historical reference and research playground for new scaling ideas.

Use cases

• train large neural network models that don't fit on a single GPU
• shard model parameters and optimizer states across GPUs
• experiment with distributed training scaling techniques
• scale model training with limited GPU memory
• use model parallelism and pipeline parallelism in PyTorch

When to choose

• you need to experiment with research-stage scaling techniques beyond what PyTorch ships
• you want composable distributed training modules with simple APIs
• you're studying how FSDP works or prototyping new sharding ideas

When to avoid

• you just need FSDP in production - use the version built into modern PyTorch
• you need a actively developed library with new features
• you're not using PyTorch

Facets

library · maturity maintenance

machine-learning llm-training gpu-computing deep-learning machine-learning large-language-models gpu-computing python cross-platform pytorch distributed-training fsdp model-parallelism data-parallelism facebook-research gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="facebookresearch/fairscale")