Expand description
Benchmark and evaluation harness for FibQuant.
Provides a typed benchmark receipt (FibBenchmarkReceiptV1) capturing
recall@K, nDCG@K, compression ratio, cosine similarity, MSE, and timing
for a corpus of vectors queried against a FibQuant-quantized database.
Structs§
- FibBenchmark
Corpus - A corpus of database vectors, queries, and ground-truth top-K labels.
- FibBenchmark
Receipt V1 - Typed benchmark receipt produced by
run_benchmark.
Constants§
- BENCHMARK_
SCHEMA - Schema version stamp for benchmark receipts.
Functions§
- ndcg_
at_ k - nDCG@K: normalized Discounted Cumulative Gain at K.
- recall_
at_ k - Recall@K: fraction of true top-K items found in the approximate top-K.
- run_
benchmark - Run a full benchmark on a corpus with a FibQuantizer.