Fixed range validation for text generation penalty parameters (#3623)

ludwig/schema/llms/generation.py

@@ -165,39 +165,31 @@ class LLMGenerationConfig(schema_utils.BaseMarshmallowConfig):
         "depending on the size of the model.",
     )
 
-    diversity_penalty: Optional[float] = schema_utils.FloatRange(
+    diversity_penalty: Optional[float] = schema_utils.NonNegativeFloat(
         default=0.0,
-        min=0.0,
-        max=1.0,
         allow_none=True,
         description="The value used to control the diversity of the generated text. The higher the value, the more "
-        "diverse the text will be. The value should be between 0 and 1.0. If set to 0, no diversity is enforced."
+        "diverse the text will be. If set to 0, no diversity is enforced."
         "This value is subtracted from a beam(s) score if it generates a token same as any beam from other group at a"
         "particular time. Note that diversity_penalty is only effective if group beam search is enabled.",
     )
 
-    repetition_penalty: Optional[float] = schema_utils.FloatRange(
+    repetition_penalty: Optional[float] = schema_utils.NonNegativeFloat(
         default=1.0,
-        min=0.0,
-        max=1.0,
         allow_none=True,
         description="The parameter for repetition penalty. 1.0 means no penalty. "
         "See [this paper](https://arxiv.org/pdf/1909.05858.pdf) for more details.",
     )
 
-    encoder_repetition_penalty: Optional[float] = schema_utils.FloatRange(
+    encoder_repetition_penalty: Optional[float] = schema_utils.NonNegativeFloat(
         default=1.0,
-        min=0.0,
-        max=1.0,
         allow_none=True,
         description="The paramater for encoder_repetition_penalty. An exponential penalty on sequences that are not"
         " in the original input. 1.0 means no penalty.",
     )
 
-    length_penalty: Optional[float] = schema_utils.FloatRange(
+    length_penalty: Optional[float] = schema_utils.Float(
         default=1.0,
-        min=0.0,
-        max=1.0,
         allow_none=True,
         description="Exponential penalty to the length that is used with beam-based generation. It is applied as an "
         "exponent to the sequence length, which in turn is used to divide the score of the sequence. Since the score is"

ludwig/schema/utils.py

@@ -602,6 +602,35 @@ def IntegerRange(
     )
 
 
+@DeveloperAPI
+def Float(
+    default: Union[None, int],
+    allow_none=False,
+    description="",
+    parameter_metadata: ParameterMetadata = None,
+):
+    """Returns a dataclass field with marshmallow metadata strictly enforcing float inputs."""
+    if default is not None:
+        try:
+            assert isinstance(default, float) or isinstance(default, int)
+        except Exception:
+            raise ValidationError(f"Invalid default: `{default}`")
+    return field(
+        metadata={
+            "marshmallow_field": fields.Float(
+                allow_none=allow_none,
+                load_default=default,
+                dump_default=default,
+                metadata={
+                    "description": description,
+                    "parameter_metadata": convert_metadata_to_json(parameter_metadata),
+                },
+            )
+        },
+        default=default,
+    )
+
+
 @DeveloperAPI
 def NonNegativeFloat(
     default: Union[None, float],