Chronos-2: Change default fine-tuning learning rate and remove experimental label (#381)

*Issue #, if available:*

*Description of changes:* Lower learning rates generally appear to be
working better. This is probably because we are doing full fine-tuning
of a model with 120M params.


By submitting this pull request, I confirm that you can use, modify,
copy, and redistribute this contribution, under the terms of your
choice.

src/chronos/chronos2/pipeline.py

@@ -100,7 +100,7 @@ class Chronos2Pipeline(BaseChronosPipeline):
         | Sequence[Mapping[str, TensorOrArray | Mapping[str, TensorOrArray | None]]]
         | None = None,
         context_length: int | None = None,
-        learning_rate: float = 1e-5,
+        learning_rate: float = 1e-6,
         num_steps: int = 1000,
         batch_size: int = 256,
         output_dir: Path | str | None = None,
@@ -126,7 +126,7 @@ class Chronos2Pipeline(BaseChronosPipeline):
         context_length
             The maximum context length used during fine-tuning, by default set to the model's default context length
         learning_rate
-            The learning rate for the optimizer, by default 1e-5
+            The learning rate for the optimizer, by default 1e-6
         num_steps
             The number of steps to fine-tune for, by default 1000
         batch_size
@@ -153,12 +153,6 @@ class Chronos2Pipeline(BaseChronosPipeline):
 
         from chronos.chronos2.trainer import Chronos2Trainer, EvaluateAndSaveFinalStepCallback
 
-        warnings.warn(
-            "Fine-tuning support is experimental and may be changed in future versions.",
-            category=FutureWarning,
-            stacklevel=2,
-        )
-
         # Create a copy of the model to avoid modifying the original
         config = deepcopy(self.model.config)
         model = Chronos2Model(config).to(self.model.device)  # type: ignore