pytorch: add squeeze and tensor creation examples

Author: danbev

fundamentals/pytorch/src/squeeze.py

@@ -0,0 +1,25 @@
+import torch
+
+def main():
+    print("Squeeze Tensor Example")
+    t = torch.tensor(
+        [ # first dimension just has one element
+            [1, 2, 3] # second dimension has three elements
+        ] # this would be shape [3, 1] in ggml.
+    )
+    # So what sequeeze does it it removes the empty dimensions.
+    print(t.shape)
+    t = t.squeeze()
+    print(t.shape)
+
+    # And we can also add a dimension back in.
+    t = t.unsqueeze(0)  # Add a dimension at index 0
+    print(t.shape)
+
+    t = t.unsqueeze(0)  # Add a dimension at index 0
+    print(t.shape)
+
+# When a script is run directly the special variable __main__ is set to "__main__"
+print(f"__name__ = {__name__}")
+if __name__ == "__main__":
+    main()

fundamentals/pytorch/src/tensor.py

@@ -0,0 +1,34 @@
+import torch
+
+def print_tensor(t: torch.Tensor):
+    print(t, type(t), t.dtype, t.device, t.requires_grad)
+
+def main():
+    print("Tensor creation examples")
+    # We can create a tensor from a Python list:
+    t = torch.tensor([1, 2])
+    print_tensor(t)
+
+    # From a list of lists:
+    t = torch.tensor([[1, 2], [3, 4]])
+    print_tensor(t)
+
+    t = torch.tensor(5.0)
+    print_tensor(t)
+
+    t = torch.tensor([0, 0, 0],   # Initial data
+             dtype=None,      # Data type (torch.float32, torch.int64, etc.)
+             device=None,     # Device ('cpu', 'cuda', 'cuda:0', etc.)
+             requires_grad=False)  # Track gradients for autograd
+    print_tensor(t)
+
+    t = torch.range(0, 10, 2)  # Start, end, how big each step should be (we know the step size here)
+    print_tensor(t)
+
+    t = torch.linspace(0, 10, 5)  # Start, end, number of steps (we know the number of points needed)
+    print_tensor(t)
+
+
+if __name__ == "__main__":
+    main()
+