From aade91ce8e424eee2570ea1fb2ce21f9759516ad Mon Sep 17 00:00:00 2001
From: "Bintang Alam Semesta W.A.M"
 <23573683+bintang-aswam@users.noreply.github.com>
Date: Thu, 10 Jul 2025 08:35:46 +0700
Subject: [PATCH] zero the gradients after updating weights

Manually zero the gradients after updating weights by using machine epsilon for standard float (64-bit)
---
 .../examples_autograd/polynomial_custom_function.py  | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/beginner_source/examples_autograd/polynomial_custom_function.py b/beginner_source/examples_autograd/polynomial_custom_function.py
index 39057c8fd7..72df7c3f1f 100755
--- a/beginner_source/examples_autograd/polynomial_custom_function.py
+++ b/beginner_source/examples_autograd/polynomial_custom_function.py
@@ -74,7 +74,7 @@ def backward(ctx, grad_output):
 d = torch.full((), 0.3, device=device, dtype=dtype, requires_grad=True)
 
 learning_rate = 5e-6
-for t in range(2000):
+for t in range(int(1/(learning_rate))):
     # To apply our Function, we use Function.apply method. We alias this as 'P3'.
     P3 = LegendrePolynomial3.apply
 
@@ -98,9 +98,11 @@ def backward(ctx, grad_output):
         d -= learning_rate * d.grad
 
         # Manually zero the gradients after updating weights
-        a.grad = None
-        b.grad = None
-        c.grad = None
-        d.grad = None
+        # by using machine epsilon for standard float (64-bit)
+        import sys        
+        a.grad = loss*sys.float_info.epsilon
+        b.grad = loss*sys.float_info.epsilon
+        c.grad = loss*sys.float_info.epsilon
+        d.grad = loss*sys.float_info.epsilon
 
 print(f'Result: y = {a.item()} + {b.item()} * P3({c.item()} + {d.item()} x)')