Spaces:
Running
Running
| # Migration: v1.3 -> v1.4 | |
| v1.4 is a source/UI/evaluation upgrade. No corpus or database migration is required. | |
| ## Main changes | |
| - Ask immediately shows `Processing...` and disables duplicate submissions. | |
| - Evaluation immediately shows a running state and disables duplicate benchmark runs. | |
| - Source-level AP/nDCG metrics deduplicate repeated chunks from one file. | |
| - Evaluation bypasses response caching and does not mutate chat history. | |
| - Deep judge scores are citation-calibrated. | |
| - Letter grades use quality gates. | |
| - Structured table lookup routing is clarified. | |
| - The demo benchmark expands to v1.4. | |
| - Architecture + API becomes an interactive runtime/API reference tab. | |
| - New session-status and benchmark-metadata API endpoints are available. | |
| ## Upgrade | |
| Extract the v1.4 patch over a clean v1.3 repository, then: | |
| ```bash | |
| git add . | |
| git commit -m "Upgrade RAGForge to v1.4 evaluation and API observability" | |
| git push origin main | |
| ``` | |
| The patch intentionally excludes `demo_documents/NIST_AI_RMF_1.0.pdf`, so an existing Hugging Face Xet/LFS setup is not disturbed. | |
| ## Post-deploy checks | |
| 1. Click Ask and confirm the button changes to `Processing...` immediately. | |
| 2. Start Quick evaluation and confirm `Run evaluation` becomes disabled with a visible running message. | |
| 3. Run Standard and verify source AP@5 never exceeds 1.0. | |
| 4. Run Standard followed by Deep and verify Deep pipeline latency is not near zero due to cache hits. | |
| 5. Inspect diagnostics for Text2SQL/citation/reranker findings. | |
| 6. Open Architecture + API, click `Refresh runtime view`, and verify the current workspace and curl examples appear. | |
| 7. Open `/docs` to confirm the FastAPI v1.4 endpoints. | |