Weights & Biases alternatives

Experiment-tracking and ML-ops platform (metrics, artifacts, sweeps, Weave tracing/evals). Acquired by CoreWeave (completed 2025-05-05, ~$1.7B); now part of CoreWeave Forge. Full Weights & Biases record →

  1. 1 FSDP (PyTorch) Training infrastructure

    also in Training infrastructure

    OSS ✓ Self-host ✓ Managed ✓ MCP ? ★ 105k released 2026-09-30 BSD-3-Clause (PyTorch) Compare with Weights & Biases →

  2. 2 Unsloth Training infrastructure

    also in Training infrastructure

    OSS ✓ Self-host ✓ Managed ✓ MCP ? ★ 79k released 2026-10-26 Apache-2.0 (free tier) Compare with Weights & Biases →

  3. 3 DeepSpeed Training infrastructure

    also in Training infrastructure

    OSS ✓ Self-host ✓ Managed ✓ MCP ? ★ 43k released 2026-10-20 Apache-2.0 Compare with Weights & Biases →

  4. 4 Label Studio (HumanSignal) Training infrastructure

    also in Training infrastructure

    OSS ✓ Self-host ✓ Managed ✓ MCP ? ★ 29k released 2026-09-29 Apache-2.0 (Community); proprietary Enterprise Compare with Weights & Biases →

  5. 5 MLflow Training infrastructure

    also in Training infrastructure

    OSS ✓ Self-host ✓ Managed ? MCP ✓ ★ 29k released 2026-10-21 Apache-2.0 Compare with Weights & Biases →

  6. 6 Prefect Training infrastructure

    also in Training infrastructure

    OSS ✓ Self-host ✓ Managed ✓ MCP ? ★ 24k released 2026-10-21 Apache-2.0 Compare with Weights & Biases →

  7. 7 Hugging Face Datasets Training infrastructure

    also in Training infrastructure

    OSS ✓ Self-host ✓ Managed ✓ MCP ? ★ 22k released 2026-10-05 Apache-2.0 Compare with Weights & Biases →

  8. 8 PEFT (Hugging Face) Training infrastructure

    also in Training infrastructure

    OSS ✓ Self-host ✓ Managed ✓ MCP ? ★ 22k released 2026-10-01 Apache-2.0 Compare with Weights & Biases →

Ranked by shared memory design (storage, retrieval, conflict handling, memory unit), same section and shared must-haves (open source, self-hosting, MCP), with adoption as a tie-breaker. ✓ confirmed · — no · ? not in the catalog yet. Want picks for your own requirements? Use the chooser. Checking a whole project? Check your stack.