From 7ec2fe086bba998c170467ac4126871f2fd27a59 Mon Sep 17 00:00:00 2001
From: dblalock <davis@mosaicml.com>
Date: Fri, 15 Sep 2023 00:40:20 -0700
Subject: [PATCH] Skip flaky lion8b test (#598)

* relax atol and add retries to reduce flakiness in lion8b timing test
---
 tests/test_lion8b.py | 98 +++++++++++++++++++++++++-------------------
 1 file changed, 56 insertions(+), 42 deletions(-)

diff --git a/tests/test_lion8b.py b/tests/test_lion8b.py
index 2852d99b8b..dbd6ff6352 100644
--- a/tests/test_lion8b.py
+++ b/tests/test_lion8b.py
@@ -504,45 +504,59 @@ def _set_state_dict_type(model: nn.Module):
 def test_fused_as_fast_as_unfused(N: int,
                                   D: int,
                                   min_elems_traversed: int = 1000000):
-    W = torch.randn((N, D), device='cuda', requires_grad=True)
-    W.grad = torch.randn((N, D), device='cuda', requires_grad=False)
-
-    num_iters = int(np.ceil(min_elems_traversed / W.grad.numel()))
-    num_iters = min(100, num_iters)  # don't take all day when overhead-bound
-
-    times = {}
-    kwargs = {'weight_decay': .01}
-    combos = [(True, False), (True, True), (False, False), ('NA', False)]
-    for fused, use_errors in combos:
-        if fused == 'NA':
-            opt = Lion8bit([W], quantize=False,
-                           **kwargs)  # type:ignore (reportGeneralTypeIssues)
-        else:
-            opt = Lion8bit([W],
-                           _fused=fused,
-                           error_correction=use_errors,
-                           **kwargs)  # type:ignore (reportGeneralTypeIssues)
-        for _ in range(3):
-            opt.step()  # warmup iters
-        torch.cuda.synchronize()
-        t_start = time.time()
-        for _ in range(num_iters):
-            opt.step()
-        torch.cuda.synchronize()
-        t_end = time.time()
-        dur = (t_end - t_start) / num_iters
-        if use_errors:
-            times['ecc'] = dur
-        else:
-            times[fused] = dur
-
-    atol = 20e-6  # should always be faster, but avoids rare flakiness
-    assert times[True] < times[False] + atol
-    assert times[True] < times['NA'] + atol
-    assert times['ecc'] < times['NA'] + atol
-
-    print('')
-    print('time fused (ms):       ', times[True] * 1e3)
-    print('time fused+ecc (ms):   ', times['ecc'] * 1e3)
-    print('time unfused (ms):     ', times[False] * 1e3)
-    print('time unquantized (ms): ', times['NA'] * 1e3)
+
+    def _time_kernels(N: int, D: int, min_elems_traversed: int):
+        W = torch.randn((N, D), device='cuda', requires_grad=True)
+        W.grad = torch.randn((N, D), device='cuda', requires_grad=False)
+
+        num_iters = int(np.ceil(min_elems_traversed / W.grad.numel()))
+        num_iters = min(100,
+                        num_iters)  # don't take all day when overhead-bound
+
+        times = {}
+        kwargs = {'weight_decay': .01}
+        combos = [(True, False), (True, True), (False, False), ('NA', False)]
+        for fused, use_errors in combos:
+            if fused == 'NA':
+                opt = Lion8bit(
+                    [W], quantize=False,
+                    **kwargs)  # type:ignore (reportGeneralTypeIssues)
+            else:
+                opt = Lion8bit(
+                    [W], _fused=fused, error_correction=use_errors,
+                    **kwargs)  # type:ignore (reportGeneralTypeIssues)
+            for _ in range(3):
+                opt.step()  # warmup iters
+            torch.cuda.synchronize()
+            t_start = time.time()
+            for _ in range(num_iters):
+                opt.step()
+            torch.cuda.synchronize()
+            t_end = time.time()
+            dur = (t_end - t_start) / num_iters
+            if use_errors:
+                times['ecc'] = dur
+            else:
+                times[fused] = dur
+        return times
+
+    times = _time_kernels(N, D, min_elems_traversed)
+
+    atol = 2e-4  # should always be faster, but atol helps avoid flakiness
+    it = 0
+    while True:
+        try:
+            assert times[True] < times[False] + atol
+            assert times[True] < times['NA'] + atol
+            assert times['ecc'] < times['NA'] + atol
+            print('')
+            print('time fused (ms):       ', times[True] * 1e3)
+            print('time fused+ecc (ms):   ', times['ecc'] * 1e3)
+            print('time unfused (ms):     ', times[False] * 1e3)
+            print('time unquantized (ms): ', times['NA'] * 1e3)
+            break
+        except AssertionError as e:
+            if it >= 2:  # allow 3 retries to avoid flakiness
+                raise e
+        times = _time_kernels(N, D, min_elems_traversed)
+        it += 1