Backports for v2.0.1 and version bump (#369)

*Issue #, if available:*

*Description of changes:*


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

pyproject.toml

@@ -15,7 +15,7 @@ license = { file = "LICENSE" }
 requires-python = ">=3.10"
 dependencies = [
   "torch>=2.0,<3",
-  "transformers>=4.49,<5",
+  "transformers>=4.41,<5",
   "accelerate>=0.34,<2",
   "numpy>=1.21,<3",
   "einops>=0.7.0,<1",

src/chronos/__about__.py

@@ -1 +1 @@
-__version__ = "2.0.0"
+__version__ = "2.0.1"

src/chronos/chronos_bolt.py

@@ -407,6 +407,8 @@ class ChronosBoltPipeline(BaseChronosPipeline):
     def __init__(self, model: ChronosBoltModelForForecasting):
         super().__init__(inner_model=model)  # type: ignore
         self.model = model
+        self.model_context_length: int = self.model.config.chronos_config["context_length"]
+        self.model_prediction_length: int = self.model.config.chronos_config["prediction_length"]
 
     @property
     def quantiles(self) -> List[float]:
@@ -487,14 +489,12 @@ class ChronosBoltPipeline(BaseChronosPipeline):
         """
         context_tensor = self._prepare_and_validate_context(context=inputs)
 
-        model_context_length: int = self.model.config.chronos_config["context_length"]
-        model_prediction_length: int = self.model.config.chronos_config["prediction_length"]
         if prediction_length is None:
-            prediction_length = model_prediction_length
+            prediction_length = self.model_prediction_length
 
-        if prediction_length > model_prediction_length:
+        if prediction_length > self.model_prediction_length:
             msg = (
-                f"We recommend keeping prediction length <= {model_prediction_length}. "
+                f"We recommend keeping prediction length <= {self.model_prediction_length}. "
                 "The quality of longer predictions may degrade since the model is not optimized for it. "
             )
             if limit_prediction_length:
@@ -507,32 +507,46 @@ class ChronosBoltPipeline(BaseChronosPipeline):
 
         # We truncate the context here because otherwise batches with very long
         # context could take up large amounts of GPU memory unnecessarily.
-        if context_tensor.shape[-1] > model_context_length:
-            context_tensor = context_tensor[..., -model_context_length:]
+        if context_tensor.shape[-1] > self.model_context_length:
+            context_tensor = context_tensor[..., -self.model_context_length :]
 
-        # TODO: We unroll the forecast of Chronos Bolt greedily with the full forecast
-        # horizon that the model was trained with (i.e., 64). This results in variance collapsing
-        # every 64 steps.
-        context_tensor = context_tensor.to(
-            device=self.model.device,
-            dtype=torch.float32,
-        )
-        while remaining > 0:
-            with torch.no_grad():
-                prediction = self.model(
-                    context=context_tensor,
-                ).quantile_preds.to(context_tensor)
+        context_tensor = context_tensor.to(device=self.model.device, dtype=torch.float32)
+        # First block prediction
+        with torch.no_grad():
+            prediction: torch.Tensor = self.model(context=context_tensor).quantile_preds.to(context_tensor)
 
             predictions.append(prediction)
             remaining -= prediction.shape[-1]
 
-            if remaining <= 0:
-                break
+        # NOTE: The following heuristic for better prediction intervals with long-horizon forecasts
+        # uses all quantiles generated by the model for the first `model_prediction_length` steps,
+        # concatenating each quantile with the context and generating the next `model_prediction_length` steps.
+        # The `num_quantiles * num_quantiles` "samples" thus generated are then reduced to `num_quantiles`
+        # by computing empirical quantiles. Note that this option scales the batch size by `num_quantiles`
+        # when the `prediction_length` is greater than `model_prediction_length`.
 
-            central_idx = torch.abs(torch.tensor(self.quantiles) - 0.5).argmin()
-            central_prediction = prediction[:, central_idx]
+        if remaining > 0:
+            # Expand the context along quantile axis
+            context_tensor = context_tensor.unsqueeze(1).repeat(1, len(self.quantiles), 1)
 
-            context_tensor = torch.cat([context_tensor, central_prediction], dim=-1)
+        quantile_tensor = torch.tensor(self.quantiles, device=context_tensor.device)
+        while remaining > 0:
+            # Append the prediction to context
+            context_tensor = torch.cat([context_tensor, prediction], dim=-1)[..., -self.model_context_length :]
+            (batch_size, n_quantiles, context_length) = context_tensor.shape
+
+            with torch.no_grad():
+                # Reshape (batch, n_quantiles, context_length) -> (batch * n_quantiles, context_length)
+                prediction = self.model(
+                    context=context_tensor.reshape(batch_size * n_quantiles, context_length)
+                ).quantile_preds.to(context_tensor)
+            # Reshape predictions from (batch * n_quantiles, n_quantiles, model_prediction_length) to (batch, n_quantiles * n_quantiles, model_prediction_length)
+            prediction = prediction.reshape(batch_size, n_quantiles * n_quantiles, -1)
+            # Reduce `n_quantiles * n_quantiles` to n_quantiles and transpose back to (batch_size, n_quantiles, model_prediction_length)
+            prediction = torch.quantile(prediction, q=quantile_tensor, dim=1).transpose(0, 1)
+
+            predictions.append(prediction)
+            remaining -= prediction.shape[-1]
 
         return torch.cat(predictions, dim=-1)[..., :prediction_length].to(dtype=torch.float32, device="cpu")