diff --git a/neurips21.html b/neurips21.html
index d2c6b4d9..e7bc65f3 100644
--- a/neurips21.html
+++ b/neurips21.html
@@ -277,7 +277,7 @@
Benchmark Datasets
1B points |
100M base points |
10K queries |
- link |
+ link |
CC0 |
@@ -298,10 +298,10 @@ Benchmark Datasets
100 |
L2 |
k-NN |
- 1B points |
+ 1B points |
N/A |
- 100K queries |
- link |
+ 100K queries |
+ link |
link to terms |
@@ -310,10 +310,10 @@ Benchmark Datasets
100 |
L2 |
k-NN |
- 1B points |
- 100M base points |
- 29.3K queries |
- link |
+ 1B points |
+ 100M base points |
+ 29.3K queries |
+ link |
O-UDA |
@@ -378,7 +378,7 @@ Metrics
- Baseline DiskANN indices for T2 can be downloaded using "azcopy copy 'https://comp21storage.blob.core.windows.net/publiccontainer/comp21/diskann-T2-baseline-indices' 'local_folder' --recursive".
+ Baseline DiskANN indices for T2 can be downloaded using "azcopy copy 'https://comp21storage.z5.web.core.windows.net/comp21/diskann-T2-baseline-indices' 'local_folder' --recursive".
Note that this would take some time as the indices are large. All indices were built using R and L parameters set to 100.
Search for T2 used 16 threads and beamwidth 4. The Ls parameter was varied to tune recall vs QPS.
Update: T2 baseline results have been modified after measuring via pybind11 interface on docker. There was a 30-40% QPS loss using this interface