From 7dbee4edb08448858b10f529d5a91df685431a88 Mon Sep 17 00:00:00 2001
From: Metrax Authors <no-reply@google.com>
Date: Wed, 17 Jun 2026 12:28:35 -0700
Subject: [PATCH] The cl/926112932 introduced a loop vectorization bug
 (b/520028111). This cl disables the llvm loop vectorization for successful
 submission of metrax `nlp_metrics_test`.

PiperOrigin-RevId: 933863640
---
 src/metrax/nlp_metrics_test.py | 66 ++++++++++++++++++++++------------
 1 file changed, 43 insertions(+), 23 deletions(-)

diff --git a/src/metrax/nlp_metrics_test.py b/src/metrax/nlp_metrics_test.py
index 8152553..a7db800 100644
--- a/src/metrax/nlp_metrics_test.py
+++ b/src/metrax/nlp_metrics_test.py
@@ -15,8 +15,20 @@
 """Tests for metrax nlp metrics."""
 
 import os
-os.environ['KERAS_BACKEND'] = 'jax'
 
+os.environ["KERAS_BACKEND"] = "jax"
+
+# Disable platform dependent math to work around loop vectorizer assertions in
+# LLVM (e.g., LoopVectorizationCostModel::getVectorCallCost) under compiler
+# integrations (such as cl/926112932).
+# See go/llvm-bad-integrate for details.
+# TODO(ggaonkar): Remove this once the bug is fixed.
+xla_flags = os.environ.get("XLA_FLAGS", "")
+if "--xla_backend_extra_options=-vectorize-loops=false" not in xla_flags:
+  os.environ["XLA_FLAGS"] = (
+      xla_flags + " --xla_backend_extra_options=-vectorize-loops=false"
+  ).strip()
+# pylint: disable=g-import-not-at-top
 from absl.testing import absltest
 from absl.testing import parameterized
 import jax.numpy as jnp
@@ -217,28 +229,28 @@ def test_rougen_merge_fails_on_different_max_order(self):
 
   @parameterized.named_parameters(
       (
-          'basic',
+          "basic",
           np.random.randint(10, size=[2, 5, 10]),
           np.random.uniform(size=(2, 5, 10, 20)),
           None,
           False,
       ),
       (
-          'weighted',
+          "weighted",
           np.random.randint(10, size=[2, 5, 10]),
           np.random.uniform(size=(2, 5, 10, 20)),
           np.random.randint(2, size=(2, 5, 10)).astype(np.float32),
           False,
       ),
       (
-          'negative_values',
+          "negative_values",
           np.random.randint(10, size=[2, 5, 10]),
           np.random.uniform(size=(2, 5, 10, 20), low=-2, high=2),
           None,
           False,
       ),
       (
-          'from_logits',
+          "from_logits",
           np.random.randint(10, size=[2, 5, 10]),
           np.random.uniform(size=(2, 5, 10, 20), low=-2, high=2),
           None,
@@ -258,8 +270,8 @@ def test_perplexity(self, y_true, y_pred, sample_weights, from_logits):
           sample_weights=weights,
           from_logits=from_logits,
       )
-      metrax_metric = update if metrax_metric is None else metrax_metric.merge(
-          update
+      metrax_metric = (
+          update if metrax_metric is None else metrax_metric.merge(update)
       )
 
     expected = keras_metric.result()
@@ -273,14 +285,14 @@ def test_perplexity(self, y_true, y_pred, sample_weights, from_logits):
   def test_wer(self):
     """Tests that WER metric computes correct values with tokenized and untokenized inputs."""
     string_preds = [
-      "the cat sat on the mat",
-      "a quick brown fox jumps over the lazy dog",
-      "hello world"
+        "the cat sat on the mat",
+        "a quick brown fox jumps over the lazy dog",
+        "hello world",
     ]
     string_refs = [
-      "the cat sat on the hat",
-      "the quick brown fox jumps over the lazy dog",
-      "hello beautiful world"
+        "the cat sat on the hat",
+        "the quick brown fox jumps over the lazy dog",
+        "hello beautiful world",
     ]
     tokenized_preds = [sentence.split() for sentence in string_preds]
     tokenized_refs = [sentence.split() for sentence in string_refs]
@@ -288,31 +300,39 @@ def test_wer(self):
     metrax_token_metric = None
     keras_metric = keras_hub.metrics.EditDistance(normalize=True)
     for pred, ref in zip(tokenized_preds, tokenized_refs):
-      metrax_update = metrax.WER.from_model_output(pred,ref)
+      metrax_update = metrax.WER.from_model_output(pred, ref)
       keras_metric.update_state(ref, pred)
-      metrax_token_metric = metrax_update if metrax_token_metric is None else metrax_token_metric.merge(metrax_update)
+      metrax_token_metric = (
+          metrax_update
+          if metrax_token_metric is None
+          else metrax_token_metric.merge(metrax_update)
+      )
 
     np.testing.assert_allclose(
         metrax_token_metric.compute(),
         keras_metric.result(),
         rtol=1e-05,
         atol=1e-05,
-        err_msg="String-based WER should match keras_hub EditDistance"
+        err_msg="String-based WER should match keras_hub EditDistance",
     )
 
     metrax_string_metric = None
     for pred, ref in zip(string_preds, string_refs):
       update = metrax.WER.from_model_output(predictions=pred, references=ref)
-      metrax_string_metric = update if metrax_string_metric is None else metrax_string_metric.merge(update)
+      metrax_string_metric = (
+          update
+          if metrax_string_metric is None
+          else metrax_string_metric.merge(update)
+      )
 
     np.testing.assert_allclose(
-    metrax_string_metric.compute(),
-    metrax_token_metric.compute(),
-    rtol=1e-05,
-    atol=1e-05,
-    err_msg="String input and tokenized input should produce the same WER"
+        metrax_string_metric.compute(),
+        metrax_token_metric.compute(),
+        rtol=1e-05,
+        atol=1e-05,
+        err_msg="String input and tokenized input should produce the same WER",
     )
 
 
-if __name__ == '__main__':
+if __name__ == "__main__":
   absltest.main()