[claudespice]

Problem

ONNX inference was removed from the runtime in vNext (spiceai#11684) — there is no ONNX runtime left in the build at all: Cargo.lock on trunk contains zero ort / tract-onnx / tract-core entries, and the only text-embeddings backend resolved is text-embeddings-backend-candle (no ORT backend). The embeddings docs still advertise ONNX:

• The embedding-provider table has an "ML Format(s)" column listing ONNX for the file and huggingface sources. Every other row is -, so the whole column now describes a capability that no longer exists.
• Both local-embedding deployment guides list .onnx as an accepted model-weight format, so a reader would ship ONNX weights that cannot load.

Separately, the data-ingestion page's edge example still describes "the loaded machine learning model" performing inference on OTel observations — the ONNX inference path that removal deleted.

Changes (vNext only)

• components/embeddings/index.md: dropped the now-empty "ML Format(s)" column and renamed the remaining "LLM Format(s)" column to Format(s). Its cell values are carried over verbatim — the GGUF, GGML, SafeTensor entries were not re-verified here, so flag them if that list is also stale. (Also dropped the \* footnote marker, which had no footnote on the page.)
• components/embeddings/local/deployment.md, components/embeddings/huggingface/deployment.md: removed .onnx from the accepted weight formats.
• features/data-ingestion/index.md: reframed the anomaly-detection example around querying fresh ingested data instead of ONNX inference.

Not propagated to versioned_docs/: ort/tract are present in Cargo.lock at v2.0.1 and v2.1.1, so the ONNX claims are correct in the versioned snapshots.

Verified against

spiceai/spiceai @ origin/trunk (7bc4175f59):

• Cargo.lock — grep -cE '^name = "(ort|tract-onnx|tract-core)"' → 0 on trunk, 2 at both v2.0.1 and v2.1.1.
• Cargo.lock — the only TEI backend crates present are text-embeddings-backend, -backend-candle, -backend-core, -core.
• crates/llms/src/embeddings/candle/ — the local/HF embedding loaders go through TeiEmbed (candle + safetensors); no ONNX path.
• features/machine-learning-models already records the removal; these pages were missed by the earlier sweeps (docs docs(models): remove traditional ML (ONNX) inference (spiceai/spiceai#11684) #1914, docs(api): remove ML prediction endpoints from the API reference (vNext) #1981).

npm run build passes locally.

[github-actions]

✅ Pull with Spice Passed

Passing checks:

• ✅ Title meets minimum length requirement (10 characters)
• ✅ Has at least one of the required labels: area/blog, area/docs, area/cookbook, dependencies
• ✅ No banned labels detected
• ✅ Has at least one assignee: claudespice

[claudespice]

@copilot review

[github-actions]

🚀 deployed to https://3d657576.spiceai-org-website.pages.dev