fix all test warnings (#7024)

* fix pytorch warning in nn.conv2d for same padding

* fix future warning in torch load

* fix overflow warning in tensor list test: https://github.com/numpy/numpy/issues/23606#issuecomment-1512752172

* fix floating point warnings in dtype tests using docs https://numpy.org/doc/stable/reference/generated/numpy.errstate.html and a neat solution https://stackoverflow.com/questions/53634965/change-np-seterr-behavior-inside-a-function-only

* put err state in one place; comment taken care of by function hover

* enter np errstate context manager on test setup

* put decorator on class

test/test_dtype.py

@@ -71,6 +71,7 @@ class TestDType(unittest.TestCase):
   def test_to_np(self):
     _test_to_np(Tensor(self.DATA, dtype=self.DTYPE), _to_np_dtype(self.DTYPE), np.array(self.DATA, dtype=_to_np_dtype(self.DTYPE)))
 
+  @np.errstate(all='ignore')
   def test_casts_to(self): list(map(
     lambda dtype: _test_cast(Tensor(self.DATA, dtype=dtype), self.DTYPE),
     get_available_cast_dtypes(self.DTYPE)
@@ -420,6 +421,7 @@ class TestTypeSpec(unittest.TestCase):
       subprocess.run(['DEFAULT_FLOAT=TYPO python3 -c "from tinygrad import dtypes"'],
                       shell=True, check=True)
 
+  @np.errstate(all='ignore')
   def test_dtype_str_arg(self):
     n = np.random.normal(0, 1, (10, 10)).astype(np.float32)
     tested = 0

test/test_dtype_alu.py

@@ -92,6 +92,7 @@ def universal_test_midcast(a, b, c, op1, op2, d1:DType, d2:DType):
   numpy_value = op2[1](op1[1](an, bn).astype(_to_np_dtype(d2)), cn)
   np.testing.assert_allclose(tensor_value, numpy_value, rtol=1e-6 if getenv("PTX") else 1e-7)
 
+@np.errstate(all='ignore')
 class TestDTypeALU(unittest.TestCase):
   @unittest.skipUnless(is_dtype_supported(dtypes.float64, Device.DEFAULT), f"no float64 on {Device.DEFAULT}")
   @given(ht.float64, ht.float64, strat.sampled_from(binary_operations))
@@ -181,6 +182,7 @@ class TestFromFuzzer(unittest.TestCase):
     _test_value(np.pi / 2)
      # worst case of ulp 1.5
     _test_value(np.pi * 2, unit=1.5)
+  @np.errstate(all='ignore')
   @given(strat.sampled_from(dtypes_float))
   def test_log2(self, dtype):
     if not is_dtype_supported(dtype): return

test/test_nn.py

@@ -211,7 +211,7 @@ class TestNN(unittest.TestCase):
 
   def test_conv2d_same_padding_odd_input(self):
     BS, C1, H, W = 16, 16, 29, 31
-    C2, K, S, P = 32, 4, 1, 'same'
+    C2, K, S, P = 32, 5, 1, 'same'
     self._run_conv2d_same_padding_test(BS, C1, C2, H, W, K, S, P)
 
   def test_conv2d_same_padding_large_kernel(self):

test/test_tensor.py

@@ -399,7 +399,7 @@ class TestTinygrad(unittest.TestCase):
     if is_dtype_supported(dtypes.float16):
       data = [math.nan, -math.inf, 65504, 65519, 65519.999, 65520, 65520.1]
       data = data + [-x for x in data]
-      np.testing.assert_allclose(Tensor(data, dtype=dtypes.float16).numpy(), np.array(data).astype(np.float16))
+      with np.errstate(over='ignore'): np.testing.assert_allclose(Tensor(data, dtype=dtypes.float16).numpy(), np.array(data).astype(np.float16))
 
     # uint32
     data = [1 << 33, 1 << 32, 1 << 32 - 1, 1]

test/unit/test_disk_tensor.py

@@ -13,7 +13,7 @@ def compare_weights_both(url):
   import torch
   fn = fetch(url)
   tg_weights = get_state_dict(torch_load(fn))
-  torch_weights = get_state_dict(torch.load(fn, map_location=torch.device('cpu')), tensor_type=torch.Tensor)
+  torch_weights = get_state_dict(torch.load(fn, map_location=torch.device('cpu'), weights_only=True), tensor_type=torch.Tensor)
   assert list(tg_weights.keys()) == list(torch_weights.keys())
   for k in tg_weights:
     if tg_weights[k].dtype == dtypes.bfloat16: tg_weights[k] = torch_weights[k].float() # numpy doesn't support bfloat16