feat(rerank): add TEI provider

[evaldass]

Summary

• add a dedicated Hugging Face Text Embeddings Inference (TEI) rerank provider
• support TEI base URLs and full /rerank endpoint URLs, optional Bearer auth, extra headers, and optional model names
• document TEI rerank configuration and refresh stale rerank config tests for current auto-detect behavior

Tests

• PYTHONPATH=. .venv/bin/python -m pytest -o addopts='' --confcutdir=tests/unit tests/unit/test_tei_rerank.py tests/unit/test_cohere_rerank.py tests/unit/models/rerank/test_openai_rerank_extra_headers.py tests/unit/config/test_rerank_extra_headers_config.py -q
• PYTHONPATH=. .venv/bin/python -m pytest -o addopts='' --confcutdir=tests/misc tests/misc/test_rerank_openai.py -q
• .venv/bin/python -m ruff check openviking/models/rerank/tei_rerank.py openviking/models/rerank/init.py openviking/models/rerank/volcengine_rerank.py openviking_cli/utils/config/rerank_config.py tests/unit/test_tei_rerank.py tests/misc/test_rerank_openai.py
• .venv/bin/python -m ruff format --check openviking/models/rerank/tei_rerank.py openviking/models/rerank/init.py openviking/models/rerank/volcengine_rerank.py openviking_cli/utils/config/rerank_config.py tests/unit/test_tei_rerank.py tests/misc/test_rerank_openai.py
• python3 -m compileall openviking/models/rerank openviking_cli/utils/config/rerank_config.py

Smoke test

• Verified against a live TEI reranker using BAAI/bge-reranker-v2-m3: POST /rerank returns ordered scores through TEIRerankClient.

[github-actions]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 3 🔵🔵🔵⚪⚪

🏅 Score: 75

🧪 PR contains tests

🔒 No security concerns identified

✅ No TODO sections

🔀 Multiple PR themes Sub-PR theme: Add TEI rerank provider Relevant files: openviking/models/rerank/tei_rerank.py openviking/models/rerank/init.py openviking/models/rerank/volcengine_rerank.py openviking_cli/utils/config/rerank_config.py tests/unit/test_tei_rerank.py Sub-PR theme: Refresh stale rerank config tests Relevant files: tests/misc/test_rerank_openai.py Sub-PR theme: Update rerank configuration docs Relevant files: docs/en/guides/01-configuration.md

⚡ Recommended focus areas for review Blocking I/O in Rerank Method TEIRerankClient.rerank_batch uses blocking requests.post which may starve the event loop if called from an async context. Use httpx.AsyncClient instead for async HTTP calls. response = requests.post( url=self.rerank_url, headers=headers, json=req_body, timeout=30, ) TEI Auto-Detection Missing RerankConfig._effective_provider does not auto-detect TEI provider. TEI only requires api_base, but current logic skips it, requiring explicit provider="tei" to work. if self.api_key and self.api_base: return "openai" if self.api_key: return "cohere" if self.ak and self.sk: return "vikingdb" return None Overly Broad Exception Catch Bare except Exception catches all errors including unexpected ones. Narrow to specific exceptions like requests.exceptions.RequestException and json.JSONDecodeError. except Exception as e: logger.error(f"[TEIRerankClient] Rerank failed: {e}") return None

[github-actions]

PR Code Suggestions ✨

No code suggestions found for the PR.

[evaldass]

Follow-up on the automated review notes:

• TEI auto-detection is addressed in the latest push: api_base without api_key now resolves to provider=tei, with a unit test and docs note. api_key + api_base still resolves to openai for backwards compatibility, so TEI with auth should set provider=tei explicitly.
• The synchronous requests.post call is consistent with the existing rerank client interface and the OpenAI/Cohere rerank implementations in this package.
• The broad exception fallback is also consistent with the existing rerank clients, which return None on provider failure so retrieval can fall back gracefully.