Update README.md

Changing floating point data types to match that of the model: The input types are cast to float32, but the weights themselves are float16.

README.md

@@ -38,12 +38,12 @@ traced_model = torch.jit.trace(wrapped_model, (dummy_input['input_ids'], dummy_i
 model_from_torch = ct.convert(
     traced_model,
     inputs=[
-        ct.TensorType(name="input_ids", shape=(1, ct.RangeDim(1, 512))),
-        ct.TensorType(name="attention_mask", shape=(1, ct.RangeDim(1, 512)))
+        ct.TensorType(name="input_ids", shape=(1, ct.RangeDim(1, 512)), dtype=np.float32),
+        ct.TensorType(name="attention_mask", shape=(1, ct.RangeDim(1, 512)), dtype=np.float32)
     ],
     minimum_deployment_target=ct.target.iOS17,
     convert_to="mlprogram",
-    compute_precision=ct.precision.FLOAT32
+    compute_precision=ct.precision.FLOAT16
 )
 
 # Save the CoreML model as an mlpackage