finished

Author: TabibitoQZP

README.md

@@ -0,0 +1,77 @@
+# zk-IVF-PQ
+
+## Build
+
+Build and install the Python extension:
+
+```bash
+maturin develop --release
+```
+
+## Experiment 1: Retrieval Utility Evaluation
+
+### Classic ANN (SIFT1M / GIST1M)
+
+```bash
+bash scripts/acc_bench.sh
+```
+
+Results are cached under `data/acc_bench/`.
+
+### IR (MS MARCO passage retrieval, dev)
+
+Run the two evaluation jobs (high-acc and zk-opt) and then aggregate metrics:
+
+```bash
+python -m bench.ms_macro_eval \
+  --num-runs 1 \
+  --n-list 8192 \
+  --n-probe 64 \
+  --M 8 \
+  --K 256 \
+  --cluster-bound 2048 \
+  --top-k 1000 \
+  --out-dir data/exp1/ms_macro_high_acc
+
+python -m bench.ms_macro_eval \
+  --num-runs 1 \
+  --n-list 2048 \
+  --n-probe 16 \
+  --M 8 \
+  --K 256 \
+  --cluster-bound 8192 \
+  --top-k 1000 \
+  --out-dir data/exp1/ms_macro_zk_fast
+
+python -m bench.ms_macro_result data/exp1/ms_macro_high_acc data/exp1/ms_macro_zk_fast
+```
+
+Outputs are written under `data/exp1/`.
+
+## Experiment 2: Proof Cost Evaluation
+
+Script: `scripts/bench_suite.sh` (runs `python -m bench.bench_suite` and caches results under `data/bench_result/`).
+
+```bash
+bash scripts/bench_suite.sh
+```
+
+Optional: if `matplotlib` is available, a summary plot is generated at `data/bench_result/bench_summary.pdf`.
+
+## Experiment 3: Configuration Trade-offs
+
+### Fixed scan budget: sweep `n_list`
+
+```bash
+bash scripts/optimal_config.sh
+```
+
+Outputs are written to `data/optimal_config/`.
+
+### Fixed code budget: gate-only sweep over `K` (and zk-opt selection)
+
+```bash
+bash scripts/optimal_mem_gate_only_ann_ir.sh
+```
+
+Outputs are written to `data/optimal_mem_gate_only/`.

bench/bench_suite.py

@@ -81,17 +81,17 @@ def d(self) -> int:
 
 
 DEFAULT_CONFIGS: List[BenchConfig] = [
-    BenchConfig(
-        name="basic",  # 基础测试, 后面都是以这个为基础
-        N=8192,
-        D=128,
-        M=8,
-        K=16,
-        n_list=256,
-        n_probe=16,
-        top_k=64,
-        merkled=False,
-    ),
+    # BenchConfig(
+    #     name="basic",  # 基础测试, 后面都是以这个为基础
+    #     N=8192,
+    #     D=128,
+    #     M=8,
+    #     K=16,
+    #     n_list=256,
+    #     n_probe=16,
+    #     top_k=64,
+    #     merkled=False,
+    # ),
     BenchConfig(
         name="basic-merkle",  # 基础测试, 后面都是以这个为基础
         N=8192,
@@ -103,17 +103,17 @@ def d(self) -> int:
         top_k=64,
         merkled=True,
     ),
-    BenchConfig(
-        name="low-acc",  # 超低精度测试, 主要是测试circuit是否有机会
-        N=8192,
-        D=128,
-        M=8,
-        K=1,
-        n_list=16,
-        n_probe=1,
-        top_k=1,
-        merkled=False,
-    ),
+    # BenchConfig(
+    #     name="low-acc",  # 超低精度测试, 主要是测试circuit是否有机会
+    #     N=8192,
+    #     D=128,
+    #     M=8,
+    #     K=1,
+    #     n_list=16,
+    #     n_probe=1,
+    #     top_k=1,
+    #     merkled=False,
+    # ),
     BenchConfig(
         name="low-acc-merkle",  # 超低精度测试, 主要是测试circuit是否有机会
         N=8192,
@@ -125,17 +125,17 @@ def d(self) -> int:
         top_k=1,
         merkled=True,
     ),
-    BenchConfig(
-        name="large",  # 大规模, 高精度测试
-        N=65536,
-        D=256,
-        M=16,
-        K=256,
-        n_list=512,
-        n_probe=64,
-        top_k=128,
-        merkled=False,
-    ),
+    # BenchConfig(
+    #     name="large",  # 大规模, 高精度测试
+    #     N=65536,
+    #     D=256,
+    #     M=16,
+    #     K=256,
+    #     n_list=512,
+    #     n_probe=64,
+    #     top_k=128,
+    #     merkled=False,
+    # ),
     BenchConfig(
         name="large-merkle",  # 大规模, 高精度测试
         N=65536,