|
Download experiments/postbenchmark_optimizations/README.md from Angshul/SparseGeometricRAG: direct link, hf CLI and curl.
- Browser
- Download file 1.12 kB
-
https://huggingface.co/datasets/Angshul/SparseGeometricRAG/resolve/main/experiments/postbenchmark_optimizations/README.md
- Command line
-
hf download hf://datasets/Angshul/SparseGeometricRAG/experiments/postbenchmark_optimizations/README.md
-
curl -L -o README.md https://huggingface.co/datasets/Angshul/SparseGeometricRAG/resolve/main/experiments/postbenchmark_optimizations/README.md
1.12 kB
| # Post-benchmark optimization branch | |
| `final_two_optimizations_ARCHIVE.py` is the exact recovered source supplied after the frozen six-dataset `P=100 -> top-10` benchmark was produced. | |
| It is preserved for provenance, not used to generate `RESULTS_FINAL_6DATASETS.md`. Its original one-off scale harness imported `eval_scale_rag` and `opt_two_variants` from an experiment working directory; those two transient modules did not survive as standalone files. The reusable/frozen benchmark implementation is therefore the package code in `geomretrieval/`, especially `geomretrieval/rag_top10.py`, plus the runners in `experiments/beir/` and the full-scale scripts in `experiments/msmarco_scale/`. | |
| The recovered optimization branch introduced: | |
| 1. O(K) generation-stamp aggregation instead of `np.unique` sorting. | |
| 2. A pre-lexical tail gate `min(route_docs, max(10000, 40*P))`. | |
| 3. Query-conditioned 16-coordinate residual-code sign matching with default weight `wcode=0.25`. | |
| Recorded post-benchmark outputs are stored in `results/final_100_to_10/nq_two_optimizations_merged.json` and `hotpotqa_two_optimizations_merged.json`. | |