feat: implement GPU-backed FAISS support and dynamic tokenization scaling#209
Open
wgergely wants to merge 21 commits intoyichuan-w:mainfrom
Open
feat: implement GPU-backed FAISS support and dynamic tokenization scaling#209wgergely wants to merge 21 commits intoyichuan-w:mainfrom
wgergely wants to merge 21 commits intoyichuan-w:mainfrom
Conversation
…ding computation - Force use_server=False to prevent ZMQ connection issues - Add explicit logger for better debugging - Improve code structure and comments
Implements a standalone embedding server for the FAISS backend to prevent ZMQ deadlocks that occur when mixing direct embedding computation (build) and server-based computation (search). - Adds faiss_embedding_server.py: Specialized server reusing leann-core logic. - Updates __init__.py: Exports and registers the new server module.
Adds: - gitignore-parser: For robust .gitignore handling in the CLI. - einops: Required for nomic-embed-text-v1.5 custom implementation.
- api.py: Explicitly separate server-mode (search) vs direct-mode (build) to ensure stability. - embedding_compute.py: Add parallel tokenization, adaptive batch sizing, and support for nomic-embed-text-v1.5. - tests: Add token truncation tests.
- Add gitignore-parser integration for correct file exclusion. - Add suppress_cpp_output context manager to silence noisy FAISS/HNSW backend logs. - Add code-optimized SentenceSplitter configuration.
- metadata_filter.py: Implements comprehensive filtering (comparison, membership, string, boolean) for search results. - tests: Add test suite for metadata filtering logic.
Author
|
Successfully updated the PR with the latest stabilization fixes, metadata enrichment, and MCP protocol v2025 updates. |
…ontext - CodeAnalyzer: Added robust import resolution for JS/TS and Python relative paths - CodeAnalyzer: Improved AST parsing resilience with tree-sitter bindings - Chunking: Integrated context headers for better semantic search retrieval
- API: Standardization of search interfaces and error handling - Chat: Improved RAG context injection flow - Embedding Server: Robust startup/shutdown and process management - CLI: Consistency updates for downstream consumers
WHAT: - Add compute_embeddings_voyage() for Voyage AI API with 32K context - Add resolve_voyage_api_key() to settings.py for API key resolution - Update EMBEDDING_MODEL_LIMITS with voyage-code-3 (32000 tokens) - Add 'voyage' and 'gemini' to CLI --embedding-mode choices - Fix AST chunking import: chunking_utils is in parent package WHY: - Voyage Code 3 provides 77% CoIR score for code retrieval - 32K context window enables Late Chunking strategy - AST chunking was failing due to wrong relative import path (from .chunking_utils should be from ..chunking_utils) IMPACT: - Users can now use: --embedding-mode voyage --embedding-model voyage-code-3 - AST-aware chunking now works correctly in analysis module - Fallback chunking is no longer needed when AST is available
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
PR Update Draft - Leann GPU Backend & Metadata Enrichment
This update builds upon the initial FAISS GPU support by adding robust metadata extraction and stabilizing the environment for production use.
What's New?
1. Metadata-Rich Indexing (Context Headers)
We’ve added a
CodeAnalyzerthat usestree-sitterto extract global context from files. Every code chunk now includes a "Context Header" prepended to its text:2. FAISS Stability (ZMQ Fixes)
To prevent ZMQ deadlocks observed in high-concurrency scenarios, we've implemented an in-process embedding strategy for the FAISS backend. Search operations now compute query embeddings within the same process by default.
3. MCP Protocol v2025 Upgrade
Standardized the codebase to support the latest MCP protocol version (
2025-11-25).4. Better Environment Control
Standardized
LEANN_HOMEandLEANN_DOCShandling across CLI and Server modules. The system now strictly respects these environment variables if provided.trust_remote_code=Trueto supportnomic-embed-text-v1.5out of the box.tree-sitter(0.23+) andgitignore-parserto core requirements.ProcessPoolExecutorfor true CPU parallelism.Verification
Full test suite passed, including new integration tests for the FAISS ZMQ server and metadata analyzer.