TinySearch

v0.1.1 Breaking

This release includes breaking changes for platform teams planning a safe upgrade.

Published 20d Search Engines

View tool

✓ No known CVEs patched

Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

ReleasePort's take

Light signal

editorial:auto 13d

TinySearch v0.1.1 moves ONNX and tokenizer blobs from repository to first-start download. Embedding runtime adds token_type_ids support for graph operations.

Why it matters: Reduces repository footprint for faster clones and checkouts. First deployment may experience download latency; test in dev. Routine maintenance; no breaking changes or migration deadline.

Summary

AI summary

Updates Highlights, Notes for upgraders, and TinySearch v0.1.1 across a mixed release.

Changes in this release

Type	Severity	Summary	CVE
Feature
Feature	Medium	ONNX bundle downloads on first start when embedding_backend is default. ONNX bundle downloads on first start when embedding_backend is default. Source: llm_adapter@2026-05-21 Confidence: high	—
Feature	Medium	Embedding runtime supports graphs expecting token_type_ids (zeros when required). Embedding runtime supports graphs expecting token_type_ids (zeros when required). Source: llm_adapter@2026-05-21 Confidence: high	—
Feature	Medium	Export script shares HF model ID via services/onnx_bundle_constants.py. Export script shares HF model ID via services/onnx_bundle_constants.py. Source: llm_adapter@2026-05-21 Confidence: low	—
Performance	Medium	Benchmark prefetches ONNX bundle using ensure logic when required. Benchmark prefetches ONNX bundle using ensure logic when required. Source: llm_adapter@2026-05-21 Confidence: low	—
Deprecation	Medium	ONNX and tokenizer blobs removed from git repository. ONNX and tokenizer blobs removed from git repository. Source: llm_adapter@2026-05-21 Confidence: low	—

Full changelog

TinySearch v0.1.1

Highlights

ONNX bundle on first start: When embedding_backend is default, MCP and FastAPI call ensure_onnx_bundle_sync() so the MiniLM ONNX bundle is downloaded once into models/all-minilm-l6-v2-onnx/ (or TINYSEARCH_ONNX_MODEL_DIR) instead of shipping large weights in the repo.
Smaller repo / clearer licensing: ONNX and tokenizer blobs are removed from git (.gitkeep keeps the folder). README adds a License section: MIT for project code; downloaded / optional ONNX weights remain Apache-2.0 (Hugging Face model cards apply).
Embedding runtime: ONNX inference supports graphs that expect token_type_ids (zeros when required), matching the prebuilt bundle from onnx-models/all-MiniLM-L6-v2-onnx.
Export script: Shares HF model id via services/onnx_bundle_constants.py with the export path.
Benchmark: scripts/benchmark_mcp_research.py prefetches the bundle with the same ensure logic when _REQUIRE_ONNX_BUNDLE is true and the resolved backend is default; errors if ONNX is required but the config uses a non-default embedding backend.

Notes for upgraders

After pull, run the app or benchmark once (with network) so the bundle downloads, or run scripts/export_embedding_onnx.py to generate files locally.
.gitignore now ignores the bundle files under models/all-minilm-l6-v2-onnx/; do not expect model.onnx in the clone anymore.

View diff on GitHub

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Share on X Share on Bluesky

Track TinySearch

Get notified when new releases ship.

About TinySearch

All releases →