Spaces:
Running on Zero
Running on Zero
Commit History
-
feat: Enhance answer generation and synthesis fragment processing with improved filtering and snippet extraction Running
Bellok0ee3d53 -
feat: Implement session-local synthesis artifact search functionality
Bellok215653b -
feat: Add synthesis functionality with session summary and artifact ingestion
Bellok9fcde48 -
feat: Implement document synthesis endpoint with hierarchical summarization
Belloke7d33ec -
Update README.md
Verified signature9cb8f84 -
Update README.md
Verified signature39f8350 -
feat: Add synthesis endpoint and request/response models for document synthesis
Belloka1474c6 -
feat: Implement grounded answer generation and integrate with query endpoint
Bellok79b957e -
feat: Add developmental artifacts to .dockerignore and .gitignore
Bellok02cbf32 -
feat: Add support for remote pack loading with environment configuration and enhanced metadata handling
Bellok1c68bde -
fix(ui): update system stats display logic in Gradio interface
Bellok87dbd82 -
feat: Implement RemotePackLoader for downloading Warbler packs from Hugging Face dataset repo
Bellok9823f9d
-
feat: Add HFDatasetRepoManager for Hugging Face dataset repo management
Bellok3523a4a -
feat: Add neurodivergent-memory workflow instructions and prompts
Bellok- Introduced `nd-memory-workflow.instructions.md` for memory-driven development guidelines. - Added prompts for exploring memory city and executing GitHub issues with memory logging. - Implemented setup prompt for configuring memory-driven project customizations. - Enhanced `app.py` to support hosted environments and improved embedding provider handling. - Updated `requirements.txt` and `pyproject.toml` to include necessary dependencies. - Enhanced `PackLoader` to respect include/exclude patterns and added environment-based configuration. - Implemented lexical indexing in `RetrievalAPI` for improved search capabilities. - Added tests for new functionalities in `PackLoader` and `RetrievalAPI`.93b8585
-
refactor(stats): replace separate successful retrievals and error rate with combined retrieval success rate
Bellok- Simplify system stats display by using a single "Retrieval Success Rate" metric instead of two separate fields for successful retrievals and error rate - Improves readability and consolidates related performance data into one percentage-based indicatord5e328f -
fix(ui): remove mismatched labels from radio query options
Bellok- Removed custom labels from gr.Radio component in "Query" tab - Labels outnumbered inputs, causing potential UI inconsistencies - Now relies on default input-based labeling for cleaner interface3919620
-
feat: Integrate ConflictDetector into RetrievalAPI
Bellok- Add import and initialization of ConflictDetector ('Bob the Skeptic') for conflict detection - Incorporate conflict_detector into RetrievalAPI constructor - Update retrieval examples with adjusted use_hybrid flags and added labels for better demo clarity2d1bbbf -
feat: enhance system statistics with comprehensive real-time reporting
BellokExpand get_system_stats() function to provide detailed, formatted markdown output including sections for document store, query performance, conflict detection (Bob the Skeptic), FractalStat intelligence, system health, and recent activity. This improves monitoring and debugging by offering real-time metrics and component statuses via the Gradio interface.25bfc9e -
docs: enhance README with search mode guides and app info updates, add entanglement resonance feature
Bellok- Add detailed explanations for semantic and hybrid search modes in README - Update query examples with semantic and FractalStat hybrid searches - Enhance app description with document count and fallback features - Implement entanglement_resonance function for cross-coordinate conceptual connections in fractalstat_rag_bridge.pyf22e6ff -
feat(rag): enable additional datasets and add hybrid search fallback
Bellok- Uncommented and enabled 'arxiv', 'edustories', 'manuals', 'enterprise', and 'prompt-report' dataset packs for broader knowledge coverage - Added hybrid fallback mechanism in query_warbler() to perform pure semantic search if fractalstat hybrid yields no results, improving retrieval reliability - Updated document count in Gradio UI from '2.6M+' to '100k+' for accuracy - Removed auto-sharing from demo launch for controlled access This expands the RAG system's data sources, enhances query robustness with fallback logic, corrects misinformation, and secures the demo deployment.1635a41 -
Merge branch 'main' of https://huggingface.co/spaces/Bellok/warbler-cda
Bellok81c2b00 -
revert(app): limit HF datasets to 50k arxiv and core packs for space constraints
BellokReduced dataset ingestion in README.md and app.py to exclude most HF datasets except for a balanced 50k arxiv papers, novels, and npc-dialogue, lowering from ~2.6M to ~100k documents. Updated Gradio SDK to 6.0.2 and added thumbnail for deployment efficiency on HuggingFace Spaces. Addresses resource limits while maintaining core functionality.2c1a23f -
Update README_HF.md
Verified signature0210d4e -
refactor: reduce arxiv dataset limit to 50k for deployment efficiency
Bellok- Lowered arxiv_limit from 100,000 to 50,000 to balance paper coverage with faster deployment times - Updated comments to reflect the new balanced capacity, ensuring sufficient knowledge base without excessive ingestion duration58c8726 -
feat: expand HF dataset downloads for diverse knowledge and clean debug code
Bellok- Updated comment and expanded `datasets_to_download` list to include additional packs like novels, manuals, enterprise content, and more for broader knowledge diversity. - Removed DEBUG logging block that printed embedding summaries, as it's no longer needed after testing.6abf740 -
Update app.py
Verified signature4c8cef9 -
Update README.md
Verified signature7319a05 -
feat(app): increase ArXiv dataset ingestion limit to 100k papers
BellokScale up from 10k to support a massive knowledge base, now that CPU embeddings are solid and reliable.9692d79 -
refactor: increase arxiv download limit to 10k for full capacity
BellokChanged arxiv_limit from 500 back to 10000 papers, as 10k works with CPU embeddings, restoring full download capacity for52b882f -
again
Bellok6976944 -
arxiv papers
Bellokde3ab38 -
feat(query): update confidence threshold and enhance debug logging
BellokAdjust confidence threshold from 0.1 to 0.3 in query_warbler function to restore normal threshold now that real HF data might be available. Enhance debug output to include top 3 relevance scores, threshold values, and no-results info for better troubleshooting.424ee90 -
docs: correct grammar in import comment
BellokFixed grammatical error by adding "the" to the comment on the HuggingFace Spaces GPU import for clarity and correctness.914f7c2 -
here
Bellok2ca23fd -
and again
Belloka8a243e
-
again
Bellok7ad1b0d -
again
Bellok1384c05 -
chore-debug: add debug prints to inspect query and document embeddings
BellokDuring debugging, added stderr prints to output first 5 values of query embeddings and compare the first 10 values of the first two document embeddings to check for duplicates or issues in the retrieval process.e26c40f -
feat(ingestion): limit arxiv dataset downloads to prevent HF timeout
BellokRestrict arxiv ingestion to 10k physics-focused papers to avoid Hugging Face timeouts. Comment out non-essential datasets (novels, manuals, etc.) to prioritize astronomy queries and enable faster demo loading. Maintains local packs unchanged for reliable access.5e6fb91 -
again...
Belloke57e902 -
again
Belloka10a16f -
again
Bellokac56231 -
refactor: disable hybrid search and lower confidence threshold for semantic search debugging
BellokTemporarily set fractalstat_hybrid to False and reduced confidence_threshold to 0.1 to isolate issues in semantic search functionality. This allows focused testing without hybrid mode interference.e3d3e05 -
feat: enhance debug output for embedding and scoring diagnostics
Bellok- Add debug logging in app.py to print document ID and embedding details (zero or dimensions) during document addition. - Update debug output in retrieval_api.py with comprehensive diagnostics, including semantic, fractalstat, hybrid scores with higher precision, and content length. This improves visibility for troubleshooting retrieval scoring.66bbb24 -
refactor(retrieval_api): apply confidence threshold to hybrid scores and add debug logging
Bellok- Moved fractalstat resonance calculation and hybrid scoring before threshold check, ensuring all documents are processed for hybrid filtering instead of semantic-only. - Renamed variable for brevity (fractstat_resonance). - Added debug prints to log query parameters and per-document scores for troubleshooting retrieval accuracy.b98a532 -
feat(query_warbler): add confidence_threshold parameter
BellokAdded a confidence_threshold of 0.3 to the query_warbler function to improve testing by better distinguishing academic content from casual queries. This parameter helps filter results based on query semantic confidence.4de684c -
feat: add hybrid semantic FractalStat retrieval mode
BellokAdd new HYBRID_SEMANTIC_FRACTALSTAT retrieval mode to the RetrievalMode enum and corresponding handling in RetrievalAPI. This mode enables hybrid semantic search enhanced with FractalStat scoring, initially implemented by calling the semantic similarity retrieval method. Closes enhancement request for advanced hybrid retrieval strategies.3f55674 -
feat: implement batch embedding for faster document ingestion
BellokReplace individual document ingestion with parallel batch embedding using multiple workers to significantly improve ingestion speed. Also, integrate pre-computed FractalStat coordinates for each document during ingestion.f5f6029 -
feat(app/docker): integrate committed packs and optimize embedding provider
BellokConfigure embedding provider with 4 workers, batch size 64, and caching for better throughout. Prioritize loading local committed packs (faction-politics, wisdom-scrolls) over HuggingFace downloads, suppress progress bars, and copy packs into Dockerfile. Update READMEs to fix dataset name references. Reduces startup time and ensures essential packs are always available locally.82f450e -
refactor: simplify system initialization by removing function wrapper
BellokRemove the `initialize_system()` function and `@GPU` decorator, moving initialization steps directly into the script body with added progress prints for better readability and streamlined execution.382376e