feat: LLM - Improved the handling of temperature and top_p in streaming

The `temperature` and `top_p` parameters have `float` type. However some people might pass integer values to these parameters. The streaming API is sensitive to the `int` vs. `float` parameter type and will throw an error. This CL mitigates that by converting integer values to floats.

PiperOrigin-RevId: 562738676

vertexai/language_models/_language_models.py

@@ -759,9 +759,13 @@ def predict_streaming(
             prediction_parameters["maxDecodeSteps"] = max_output_tokens
 
         if temperature is not None:
+            if isinstance(temperature, int):
+                temperature = float(temperature)
             prediction_parameters["temperature"] = temperature
 
         if top_p:
+            if isinstance(top_p, int):
+                top_p = float(top_p)
             prediction_parameters["topP"] = top_p
 
         if top_k:
@@ -1389,10 +1393,14 @@ def _prepare_request(
         if temperature is None:
             temperature = self._temperature
         if temperature is not None:
+            if isinstance(temperature, int):
+                temperature = float(temperature)
             prediction_parameters["temperature"] = temperature
 
         top_p = top_p or self._top_p
         if top_p:
+            if isinstance(top_p, int):
+                top_p = float(top_p)
             prediction_parameters["topP"] = top_p
 
         top_k = top_k or self._top_k
@@ -1749,6 +1757,8 @@ def _create_prediction_request(
         prediction_parameters = {}
 
         if temperature is not None:
+            if isinstance(temperature, int):
+                temperature = float(temperature)
             prediction_parameters["temperature"] = temperature
 
         if max_output_tokens: