Correct the Criteo paragraph against the benchmark logs

The largest streaming vs in-RAM posterior gap was 0.12 on a feature
coefficient, not ~0.1 on the intercept (the intercept gap was 0.01),
and the comparison ran on a 1M-row slice; the OOM extrapolation from
the fitted slope is ~240M rows, not 250M.

Author: YichengYang-Ethan

examples/variational_inference/streaming_dataset.ipynb

@@ -405,9 +405,10 @@
     "standard, publicly available out-of-core learning benchmark. Peak memory for the\n",
     "streaming `DataLoader` stayed flat at about 0.7 GB across a sweep from 1M to 150M\n",
     "rows, while the in-RAM `pm.Minibatch` baseline rose linearly to 15.7 GB at 150M\n",
-    "rows, which extrapolates to out-of-memory around 250M rows on the same 26 GB\n",
-    "machine. The streaming and in-RAM posteriors agreed coefficient for coefficient;\n",
-    "the largest gap was about 0.1, on the intercept.\n",
+    "rows, which extrapolates to out-of-memory around 240M rows on the same 26 GB\n",
+    "machine. On a 1M-row slice, where the in-RAM fit is cheap, the streaming and\n",
+    "in-RAM posterior means agreed coefficient for coefficient (correlation 0.998\n",
+    "across the 14 coefficients); the largest gap on any single coefficient was 0.12.\n",
     "\n",
     "## When to use it\n",
     "\n",

examples/variational_inference/streaming_dataset.myst.md

@@ -222,9 +222,10 @@ features plus the click label) on the
 standard, publicly available out-of-core learning benchmark. Peak memory for the
 streaming `DataLoader` stayed flat at about 0.7 GB across a sweep from 1M to 150M
 rows, while the in-RAM `pm.Minibatch` baseline rose linearly to 15.7 GB at 150M
-rows, which extrapolates to out-of-memory around 250M rows on the same 26 GB
-machine. The streaming and in-RAM posteriors agreed coefficient for coefficient;
-the largest gap was about 0.1, on the intercept.
+rows, which extrapolates to out-of-memory around 240M rows on the same 26 GB
+machine. On a 1M-row slice, where the in-RAM fit is cheap, the streaming and
+in-RAM posterior means agreed coefficient for coefficient (correlation 0.998
+across the 14 coefficients); the largest gap on any single coefficient was 0.12.
 
 ## When to use it