{"data":{"slug":"deepspeedai-deepspeed","name":"DeepSpeed","tagline":"Deep learning optimization library for efficient distributed training and inference","github_url":"https://github.com/deepspeedai/DeepSpeed","owner":"deepspeedai","repo":"DeepSpeed","owner_avatar_url":"https://avatars.githubusercontent.com/u/74068820?v=4","primary_language":"Python","stars":42685,"forks":4883,"topics":["billion-parameters","compression","data-parallelism","deep-learning","gpu","inference","machine-learning","mixture-of-experts","model-parallelism","pipeline-parallelism","pytorch","trillion-parameters","zero"],"archived":false,"github_pushed_at":"2026-07-11T10:17:46+00:00","maintenance_label":"Very active","url":"https://www.graphcanon.com/tools/deepspeedai-deepspeed","markdown_url":"https://www.graphcanon.com/tools/deepspeedai-deepspeed.md","api_url":"https://www.graphcanon.com/api/graphcanon/tools/deepspeedai-deepspeed","graph_url":"https://www.graphcanon.com/api/graphcanon/graph?tool=deepspeedai-deepspeed","description":"DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.","homepage_url":"https://www.deepspeed.ai/","license":"Apache-2.0","open_issues":1302,"watchers":359,"ai_summary":"DeepSpeed is a Python-based deep learning library aimed at facilitating efficient distributed training and inference, supporting PyTorch with optimizations like compression, data parallelism, model parallelism, and pipeline parallelism.","readme_excerpt":"# Installation\n\nThe quickest way to get started with DeepSpeed is via pip, this will install\nthe latest release of DeepSpeed which is not tied to specific PyTorch or CUDA\nversions. DeepSpeed includes several C++/CUDA extensions that we commonly refer\nto as our 'ops'.  By default, all of these extensions/ops will be built\njust-in-time (JIT) using [torch's JIT C++ extension loader that relies on\nninja](https://pytorch.org/docs/stable/cpp_extension.html) to build and\ndynamically link them at runtime.\n\n---\n\n## Requirements\n* [PyTorch](https://pytorch.org/) must be installed _before_ installing DeepSpeed.\n* For full feature support we recommend a version of PyTorch that is >= 2.0 and ideally the latest PyTorch stable release.\n* A CUDA or ROCm compiler such as [nvcc](https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/#introduction) or [hipcc](https://github.com/ROCm-Developer-Tools/HIPCC) used to compile C++/CUDA/HIP extensions.\n* Specific GPUs we develop and test against are listed below, this doesn't mean your GPU will not work if it doesn't fall into this category it's just DeepSpeed is most well tested on the following:\n  * NVIDIA: Pascal, Volta, Ampere, and Hopper architectures\n  * AMD: MI100 and MI200","github_created_at":"2020-01-23T18:35:18+00:00","created_at":"2026-07-11T10:35:35.701092+00:00","updated_at":"2026-07-11T12:35:00.88786+00:00","categories":[{"slug":"inference-serving","name":"Inference & Serving","url":"https://www.graphcanon.com/categories/inference-serving","markdown_url":"https://www.graphcanon.com/categories/inference-serving.md","api_url":"https://www.graphcanon.com/api/graphcanon/categories/inference-serving"},{"slug":"model-training","name":"Model Training","url":"https://www.graphcanon.com/categories/model-training","markdown_url":"https://www.graphcanon.com/categories/model-training.md","api_url":"https://www.graphcanon.com/api/graphcanon/categories/model-training"}],"tags":[{"slug":"billion-parameters","name":"billion-parameters"},{"slug":"compression","name":"compression"},{"slug":"data-parallelism","name":"data-parallelism"},{"slug":"deep-learning","name":"deep-learning"},{"slug":"gpu","name":"gpu"},{"slug":"inference","name":"inference"},{"slug":"machine-learning","name":"machine-learning"},{"slug":"mixture-of-experts","name":"mixture-of-experts"}],"trust":{"provenance":{"is_fork":false,"github_id":235860204,"owner_type":"Organization","methodology":"github_public_v1","parent_repo":null,"near_duplicate_slugs":[]},"computed_at":"2026-07-11T10:35:36.379Z","maintenance":{"label":"Very active","score":96,"methodology":"github_public_v1","releases_90d":3,"days_since_push":0,"last_release_at":"2026-06-16T20:52:17Z"},"security_summary":{"status":"no_lockfile","scanner":null,"low_count":0,"high_count":0,"last_scan_at":"2026-07-11T10:35:37.128Z","medium_count":0,"scan_profile":"none","critical_count":0}},"capability_facts":{"scan":{"source":"repo_scan","observed_at":"2026-07-11T12:34:33.645Z"},"languages":{"value":["python"],"source":"github.language","observed_at":"2026-07-11T12:34:33.645Z"},"license_spdx":{"value":"Apache-2.0","source":"github.license","observed_at":"2026-07-11T12:34:33.645Z"}},"decision_facts":{"hosting":null,"pricing":null,"requirements":null,"constraints":null,"when_to_use":["- When training or inferring with PyTorch on large datasets or complex deep learning models (up to trillion parameters)","- For efficient distributed training requiring specific optimizations like ZeRO, which significantly reduces memory usage for massive models","- If your infrastructure includes NVIDIA GPUs from the Pascal, Volta, Ampere, and Hopper architectures or AMD GPUs such as MI100 and MI200"],"when_not_to_use":["- When you are working in an environment that only supports CPU-based training without access to CUDA or ROCm compatible GPUs","- If your project's PyTorch version is less than 2.0, DeepSpeed may not support all of its features and optimizations effectively"],"source":"enrich:decision_facts","observed_at":"2026-07-11T12:35:00.523Z"},"constraint_facets":null,"decision_summary":[{"label":"Adopt for","value":"Decisions for DeepSpeed use are driven by its capacity to handle large models efficiently using techniques such as data parallelism, model parallelism, pipeline parallelism, and compression."}]}}