From ac5948cb489abb2c32d949a37cb25f2ab0179b98 Mon Sep 17 00:00:00 2001
From: Jian Chen <cjian@microsoft.com>
Date: Thu, 20 Oct 2022 12:33:07 -0400
Subject: [PATCH] Fix bug for percentile calibration module. (#13376)

### Description
Fix bug for percentile calibration module.


### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
---
 .../python/tools/quantization/calibrate.py    | 21 ++++++++++++++++---
 1 file changed, 18 insertions(+), 3 deletions(-)

diff --git a/onnxruntime/python/tools/quantization/calibrate.py b/onnxruntime/python/tools/quantization/calibrate.py
index 51497e9058..b431647313 100644
--- a/onnxruntime/python/tools/quantization/calibrate.py
+++ b/onnxruntime/python/tools/quantization/calibrate.py
@@ -570,14 +570,23 @@ class HistogramCollector(CalibrationDataCollector):
         for tensor, data_arr in name_to_arr.items():
             data_arr = np.asarray(data_arr)
             data_arr = data_arr.flatten()
+            if data_arr.size > 0:
+                min_value = np.min(data_arr)
+                max_value = np.max(data_arr)
+            else:
+                min_value = 0
+                max_value = 0
+
             data_arr = np.absolute(data_arr)  # only consider absolute value
 
             if tensor not in self.histogram_dict:
                 # first time it uses num_bins to compute histogram.
                 hist, hist_edges = np.histogram(data_arr, bins=self.num_bins)
-                self.histogram_dict[tensor] = (hist, hist_edges)
+                self.histogram_dict[tensor] = (hist, hist_edges, min_value, max_value)
             else:
                 old_histogram = self.histogram_dict[tensor]
+                old_min = old_histogram[2]
+                old_max = old_histogram[3]
                 old_hist = old_histogram[0]
                 old_hist_edges = old_histogram[1]
                 temp_amax = np.max(data_arr)
@@ -589,7 +598,7 @@ class HistogramCollector(CalibrationDataCollector):
                     old_hist_edges = np.hstack((old_hist_edges, new_bin_edges))
                 hist, hist_edges = np.histogram(data_arr, bins=old_hist_edges)
                 hist[: len(old_hist)] += old_hist
-                self.histogram_dict[tensor] = (hist, hist_edges)
+                self.histogram_dict[tensor] = (hist, hist_edges, min(old_min, min_value), max(old_max, max_value))
 
     def collect_value(self, name_to_arr):
         """
@@ -688,6 +697,7 @@ class HistogramCollector(CalibrationDataCollector):
             cdf = np.cumsum(hist / total)
             if self.symmetric:
                 idx_right = np.searchsorted(cdf, percentile / 100.0)
+
                 thresholds_dict[tensor] = (
                     -float(hist_edges[idx_right]),
                     float(hist_edges[idx_right]),
@@ -700,7 +710,12 @@ class HistogramCollector(CalibrationDataCollector):
                     float(hist_edges[idx_left]),
                     float(hist_edges[idx_right]),
                 )
-
+            min_value = histogram[2]
+            max_value = histogram[3]
+            if thresholds_dict[tensor][0] < min_value:
+                thresholds_dict[tensor] = (min_value, thresholds_dict[tensor][1])
+            if thresholds_dict[tensor][1] > max_value:
+                thresholds_dict[tensor] = (thresholds_dict[tensor][0], max_value)
             # Plot histogram for debug only
             if False:
                 apply_plot(hist, hist_edges)