From 3001faee42e395c7e5cacffc6f8ecd9de895d6dc Mon Sep 17 00:00:00 2001
From: Echo Xiao <echoxiao666@gmail.com>
Date: Sat, 1 Aug 2026 13:55:38 -0700
Subject: [PATCH 1/4] [DOC] Update tslib v2 tutorial to use the package classes

Rewrite the tslib v2 example to use the high-level pkg API
(TimeXer_pkg_v2 / DLinear_pkg_v2) instead of manually wiring up the
datamodule, model and Trainer.

- Configure via model_cfg / datamodule_cfg / trainer_cfg + .fit() / .predict()
- Add a DLinear example to show the same pipeline with another tslib model
- Fix future_known_feature typo in the TimeSeries definition
- Fix nhead -> n_heads (the old name was silently swallowed by **kwargs,
  so the model was running with the default number of heads)
- Make context_length divisible by patch_length to drop the
  'some time steps will not be used' warning
- Remove categorical_encoders, which TslibDataModule does not accept
- Note that TimeXer v2 does not yet consume categorical features and that
  the datamodule does not encode them
- Leave logging_metrics off the QuantileLoss variant: point metrics cannot
  be computed on quantile output and raise an assertion during fit
- Use accelerator='cpu' so the small demo is reproducible on any machine
---
 docs/source/tutorials/tslib_v2_example.ipynb | 1395 ++++++++----------
 1 file changed, 611 insertions(+), 784 deletions(-)

diff --git a/docs/source/tutorials/tslib_v2_example.ipynb b/docs/source/tutorials/tslib_v2_example.ipynb
index 1a4d8ad94..a87c9ef53 100644
--- a/docs/source/tutorials/tslib_v2_example.ipynb
+++ b/docs/source/tutorials/tslib_v2_example.ipynb
@@ -2,7 +2,7 @@
  "cells": [
   {
    "cell_type": "markdown",
-   "id": "b5d44943",
+   "id": "662cccd2",
    "metadata": {},
    "source": [
     "# TSLib for v2 - Example notebook for full pipeline"
@@ -10,40 +10,48 @@
   },
   {
    "cell_type": "markdown",
-   "id": "b7d27b55",
+   "id": "5b43e1d5",
    "metadata": {},
    "source": [
     "## Basic imports for getting started\n",
     "\n",
-    "This notebook is a basic vignette for the usage of the `tslib` data module on the `TimeXer` model for the v2 of PyTorch Forecasting. This is an experimental version and is an unstable version of the API.\n",
+    "This notebook is an end-to-end vignette for the `tslib` models in v2 of PyTorch Forecasting, using the high-level **package (`pkg`) classes**. You give the package class three config dicts — `model_cfg`, `datamodule_cfg`, `trainer_cfg` — and call `.fit()` / `.predict()`; it wires up the datamodule, model and `Trainer` for you.\n",
     "\n",
-    "Feedback and suggestions on this pipeline - PR [#1836](https://github.com/sktime/pytorch-forecasting/pull/1836)"
+    "This is an experimental, unstable API. Feedback welcome on issue [#1836](https://github.com/sktime/pytorch-forecasting/pull/1836)."
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": null,
-   "id": "550a3fbf",
-   "metadata": {},
+   "execution_count": 1,
+   "id": "18c542de",
+   "metadata": {
+    "execution": {
+     "iopub.execute_input": "2026-08-01T23:05:31.044514Z",
+     "iopub.status.busy": "2026-08-01T23:05:31.044420Z",
+     "iopub.status.idle": "2026-08-01T23:05:34.235414Z",
+     "shell.execute_reply": "2026-08-01T23:05:34.234976Z"
+    },
+    "jupyter": {
+     "is_executing": true
+    }
+   },
    "outputs": [],
    "source": [
     "import numpy as np\n",
     "import pandas as pd\n",
     "from sklearn.preprocessing import StandardScaler\n",
-    "import torch\n",
     "\n",
-    "from pytorch_forecasting.data.data_module import TslibDataModule\n",
-    "from pytorch_forecasting.data.encoders import (\n",
-    "    NaNLabelEncoder,\n",
-    "    TorchNormalizer,\n",
-    ")\n",
+    "from pytorch_forecasting.data.data_module import TslibDataModule  # only for the metadata peek below\n",
+    "from pytorch_forecasting.data.encoders import TorchNormalizer\n",
     "from pytorch_forecasting.data.timeseries import TimeSeries\n",
-    "from pytorch_forecasting.models.timexer._timexer_v2 import TimeXer"
+    "from pytorch_forecasting.metrics import MAE, SMAPE, QuantileLoss\n",
+    "from pytorch_forecasting.models.timexer import TimeXer_pkg_v2\n",
+    "from pytorch_forecasting.models.dlinear import DLinear_pkg_v2"
    ]
   },
   {
    "cell_type": "markdown",
-   "id": "2625ed3d",
+   "id": "07365bf1",
    "metadata": {},
    "source": [
     "## Construct a time series dataset\n",
@@ -54,8 +62,15 @@
   {
    "cell_type": "code",
    "execution_count": 2,
-   "id": "a0058487",
-   "metadata": {},
+   "id": "1c932107",
+   "metadata": {
+    "execution": {
+     "iopub.execute_input": "2026-08-01T23:05:34.237862Z",
+     "iopub.status.busy": "2026-08-01T23:05:34.237703Z",
+     "iopub.status.idle": "2026-08-01T23:05:34.264007Z",
+     "shell.execute_reply": "2026-08-01T23:05:34.263659Z"
+    }
+   },
    "outputs": [
     {
      "data": {
@@ -93,55 +108,55 @@
        "      <th>0</th>\n",
        "      <td>0</td>\n",
        "      <td>0</td>\n",
-       "      <td>0.177658</td>\n",
-       "      <td>0.181124</td>\n",
+       "      <td>-0.010135</td>\n",
+       "      <td>0.211194</td>\n",
        "      <td>0</td>\n",
        "      <td>1.000000</td>\n",
-       "      <td>0.409581</td>\n",
+       "      <td>0.065485</td>\n",
        "      <td>0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>1</th>\n",
        "      <td>0</td>\n",
        "      <td>1</td>\n",
-       "      <td>0.181124</td>\n",
-       "      <td>0.314081</td>\n",
+       "      <td>0.211194</td>\n",
+       "      <td>0.570354</td>\n",
        "      <td>0</td>\n",
        "      <td>0.995004</td>\n",
-       "      <td>0.409581</td>\n",
+       "      <td>0.065485</td>\n",
        "      <td>0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>2</th>\n",
        "      <td>0</td>\n",
        "      <td>2</td>\n",
-       "      <td>0.314081</td>\n",
-       "      <td>0.601934</td>\n",
+       "      <td>0.570354</td>\n",
+       "      <td>0.433613</td>\n",
        "      <td>0</td>\n",
        "      <td>0.980067</td>\n",
-       "      <td>0.409581</td>\n",
+       "      <td>0.065485</td>\n",
        "      <td>0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>3</th>\n",
        "      <td>0</td>\n",
        "      <td>3</td>\n",
-       "      <td>0.601934</td>\n",
-       "      <td>0.733805</td>\n",
+       "      <td>0.433613</td>\n",
+       "      <td>0.615071</td>\n",
        "      <td>0</td>\n",
        "      <td>0.955336</td>\n",
-       "      <td>0.409581</td>\n",
+       "      <td>0.065485</td>\n",
        "      <td>0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>4</th>\n",
        "      <td>0</td>\n",
        "      <td>4</td>\n",
-       "      <td>0.733805</td>\n",
-       "      <td>0.768843</td>\n",
+       "      <td>0.615071</td>\n",
+       "      <td>0.858715</td>\n",
        "      <td>0</td>\n",
        "      <td>0.921061</td>\n",
-       "      <td>0.409581</td>\n",
+       "      <td>0.065485</td>\n",
        "      <td>0</td>\n",
        "    </tr>\n",
        "  </tbody>\n",
@@ -150,18 +165,18 @@
       ],
       "text/plain": [
        "   series_id  time_idx         x         y  category  future_known_feature  \\\n",
-       "0          0         0  0.177658  0.181124         0              1.000000   \n",
-       "1          0         1  0.181124  0.314081         0              0.995004   \n",
-       "2          0         2  0.314081  0.601934         0              0.980067   \n",
-       "3          0         3  0.601934  0.733805         0              0.955336   \n",
-       "4          0         4  0.733805  0.768843         0              0.921061   \n",
+       "0          0         0 -0.010135  0.211194         0              1.000000   \n",
+       "1          0         1  0.211194  0.570354         0              0.995004   \n",
+       "2          0         2  0.570354  0.433613         0              0.980067   \n",
+       "3          0         3  0.433613  0.615071         0              0.955336   \n",
+       "4          0         4  0.615071  0.858715         0              0.921061   \n",
        "\n",
        "   static_feature  static_feature_cat  \n",
-       "0        0.409581                   0  \n",
-       "1        0.409581                   0  \n",
-       "2        0.409581                   0  \n",
-       "3        0.409581                   0  \n",
-       "4        0.409581                   0  "
+       "0        0.065485                   0  \n",
+       "1        0.065485                   0  \n",
+       "2        0.065485                   0  \n",
+       "3        0.065485                   0  \n",
+       "4        0.065485                   0  "
       ]
      },
      "execution_count": 2,
@@ -197,7 +212,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "c7c04ff5",
+   "id": "db37eeee",
    "metadata": {},
    "source": [
     "## Feature Categories and Definitions\n",
@@ -246,14 +261,21 @@
   {
    "cell_type": "code",
    "execution_count": 3,
-   "id": "89a5adbe",
-   "metadata": {},
+   "id": "7b9016cb",
+   "metadata": {
+    "execution": {
+     "iopub.execute_input": "2026-08-01T23:05:34.265375Z",
+     "iopub.status.busy": "2026-08-01T23:05:34.265299Z",
+     "iopub.status.idle": "2026-08-01T23:05:34.271403Z",
+     "shell.execute_reply": "2026-08-01T23:05:34.271057Z"
+    }
+   },
    "outputs": [
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "/home/aryan/pytorch-forecasting/pytorch_forecasting/data/timeseries/_timeseries_v2.py:105: UserWarning: TimeSeries is part of an experimental rework of the pytorch-forecasting data layer, scheduled for release with v2.0.0. The API is not stable and may change without prior warning. For beta testing, but not for stable production use. Feedback and suggestions are very welcome in pytorch-forecasting issue 1736, https://github.com/sktime/pytorch-forecasting/issues/1736\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/data/timeseries/_timeseries_v2.py:104: UserWarning: TimeSeries is part of an experimental rework of the pytorch-forecasting data layer, scheduled for release with v2.0.0. The API is not stable and may change without prior warning. For beta testing, but not for stable production use. Feedback and suggestions are very welcome in pytorch-forecasting issue 1736, https://github.com/sktime/pytorch-forecasting/issues/1736\n",
       "  warn(\n"
      ]
     }
@@ -264,7 +286,7 @@
     "    time=\"time_idx\",\n",
     "    target=\"y\",\n",
     "    group=[\"series_id\"],\n",
-    "    num=[\"x\", \"future_know_feature\", \"static_feature\"],\n",
+    "    num=[\"x\", \"future_known_feature\", \"static_feature\"],\n",
     "    cat=[\"category\", \"static_feature_cat\"],\n",
     "    known=[\"future_known_feature\"],\n",
     "    unknown=[\"x\", \"category\"],\n",
@@ -274,56 +296,91 @@
   },
   {
    "cell_type": "markdown",
-   "id": "f8753a6a",
+   "id": "99d80bd2",
+   "metadata": {},
+   "source": [
+    "> **Note on categorical features.** The `cat` columns above are declared and carried through the `TslibDataModule` (as `history_cat` / `future_cat` / `static_categorical_features`), but **TimeXer v2 currently uses only the continuous features in the context window** — categorical variables are not yet consumed by the model (`enc_in` is set to the continuous dimension only).\n",
+    "\n",
+    "Also note the datamodule does **not** encode categoricals: every feature is cast to `float32` and sliced by index, so categorical columns must already be numeric (string categories would fail). There is no categorical-encoder step in this pipeline yet. Support for proper categorical handling is planned for a future release. We keep the `cat` declaration here to show the full data definition, not because it affects the forecast today."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "5ab5e1e9",
    "metadata": {},
    "source": [
-    "## Initialise the `TslibDataModule` using the dataset\n",
+    "## The high-level package (`pkg`) API\n",
     "\n",
-    "This steps initialises a basic data module built specially for `tslib` modules and provides all the metadata required to train and implement the `tslib` of your choice!\n",
-    "You can refer the implementation for `TslibDataModule` for more information."
+    "Every v2 model has a package class (e.g. `TimeXer_pkg_v2`, `DLinear_pkg_v2`). It is a thin wrapper that manages the model, its `TslibDataModule` and the Lightning `Trainer`. You configure it with three dicts and drive it with `.fit(data)` and `.predict(data)` — no manual object wiring, and checkpoints are saved automatically under `checkpoints/`."
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 4,
-   "id": "5eae9035",
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/home/aryan/pytorch-forecasting/pytorch_forecasting/data/_tslib_data_module.py:275: UserWarning: TslibDataModule is experimental and subject to change. The API is not stable and may change without prior warning.\n",
-      "  warnings.warn(\n"
-     ]
+   "id": "36fe8aba",
+   "metadata": {
+    "execution": {
+     "iopub.execute_input": "2026-08-01T23:05:34.273984Z",
+     "iopub.status.busy": "2026-08-01T23:05:34.273895Z",
+     "iopub.status.idle": "2026-08-01T23:05:34.275739Z",
+     "shell.execute_reply": "2026-08-01T23:05:34.275436Z"
     }
-   ],
+   },
+   "outputs": [],
    "source": [
-    "data_module = TslibDataModule(\n",
-    "    time_series_dataset=dataset,\n",
-    "    context_length=30,\n",
+    "datamodule_cfg = dict(\n",
+    "    context_length=32,  # divisible by patch_length(4) -> no dropped-timestep warning\n",
     "    prediction_length=1,\n",
     "    add_relative_time_idx=True,\n",
     "    target_normalizer=TorchNormalizer(),\n",
-    "    categorical_encoders={\n",
-    "        \"category\": NaNLabelEncoder(add_nan=True),\n",
-    "        \"static_feature_cat\": NaNLabelEncoder(add_nan=True),\n",
-    "    },\n",
     "    scalers={\n",
     "        \"x\": StandardScaler(),\n",
     "        \"future_known_feature\": StandardScaler(),\n",
     "        \"static_feature\": StandardScaler(),\n",
     "    },\n",
     "    batch_size=32,\n",
+    ")\n",
+    "\n",
+    "trainer_cfg = dict(\n",
+    "    max_epochs=5,\n",
+    "    accelerator=\"cpu\",\n",
+    "    devices=1,\n",
+    "    enable_progress_bar=False,\n",
+    "    enable_model_summary=True,\n",
     ")"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "id": "a6081661",
+   "metadata": {},
+   "source": [
+    "## Peek at the metadata\n",
+    "\n",
+    "The `pkg` classes build the `TslibDataModule` for you internally. If you want to inspect the metadata that gets passed to the model, you can build one explicitly:"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 5,
-   "id": "b1843233",
-   "metadata": {},
+   "id": "8da49230",
+   "metadata": {
+    "execution": {
+     "iopub.execute_input": "2026-08-01T23:05:34.276848Z",
+     "iopub.status.busy": "2026-08-01T23:05:34.276784Z",
+     "iopub.status.idle": "2026-08-01T23:05:34.345367Z",
+     "shell.execute_reply": "2026-08-01T23:05:34.344958Z"
+    }
+   },
    "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/data/data_module/_tslib_data_module.py:331: UserWarning: TslibDataModule is experimental and subject to change. The API is not stable and may change without prior warning.\n",
+      "  warnings.warn(\n"
+     ]
+    },
     {
      "data": {
       "text/plain": [
@@ -355,7 +412,7 @@
        "  'all': 5,\n",
        "  'static_categorical': 1,\n",
        "  'static_continuous': 1},\n",
-       " 'context_length': 30,\n",
+       " 'context_length': 32,\n",
        " 'prediction_length': 1,\n",
        " 'freq': 'h',\n",
        " 'features': 'MS'}"
@@ -367,283 +424,152 @@
     }
    ],
    "source": [
-    "data_module.metadata"
+    "_dm_preview = TslibDataModule(time_series_dataset=dataset, **datamodule_cfg)\n",
+    "_dm_preview.setup(stage=\"fit\")\n",
+    "_dm_preview.metadata"
    ]
   },
   {
    "cell_type": "markdown",
-   "id": "dd9451ee",
+   "id": "b19ce3e3",
    "metadata": {},
    "source": [
-    "## Initialise the model\n",
-    "\n",
-    "We shall try out two versions of this model, one using `MAE()` and one with `QuantileLoss()`.\n",
+    "## TimeXer with MAE (point forecast) via the package class\n",
     "\n",
-    "Let us quickly import the required packages for the next steps."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "f6b568a5",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from pytorch_forecasting.metrics import MAE, SMAPE, QuantileLoss"
+    "Instead of wiring up the model, datamodule and `Trainer` by hand, we hand three config dicts to `TimeXer_pkg_v2` and call `.fit()` / `.predict()`."
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 7,
-   "id": "429b5f15",
-   "metadata": {},
+   "execution_count": 6,
+   "id": "92a9d2a6",
+   "metadata": {
+    "execution": {
+     "iopub.execute_input": "2026-08-01T23:05:34.346619Z",
+     "iopub.status.busy": "2026-08-01T23:05:34.346541Z",
+     "iopub.status.idle": "2026-08-01T23:05:34.349410Z",
+     "shell.execute_reply": "2026-08-01T23:05:34.348947Z"
+    }
+   },
    "outputs": [
     {
-     "name": "stderr",
+     "name": "stdout",
      "output_type": "stream",
      "text": [
-      "/home/aryan/pytorch-forecasting/pytorch_forecasting/models/base/_base_model_v2.py:61: UserWarning: The Model 'TimeXer' is part of an experimental reworkof the pytorch-forecasting model layer, scheduled for release with v2.0.0. The API is not stable and may change without prior warning. This class is intended for beta testing and as a basic skeleton, but not for stable production use. Feedback and suggestions are very welcome in pytorch-forecasting issue 1736, https://github.com/sktime/pytorch-forecasting/issues/1736\n",
-      "  warn(\n",
-      "/home/aryan/pytorch-forecasting/pytorch_forecasting/models/base/_tslib_base_model_v2.py:60: UserWarning: The Model 'TimeXer' is part of an experimental implementationof the pytorch-forecasting model layer for Time Series Library, scheduledfor release with v2.0.0. The API is not stableand may change without prior warning. This class is intended for betatesting, not for stable production use.\n",
-      "  warn(\n",
-      "/home/aryan/pytorch-forecasting/pytorch_forecasting/models/timexer/_timexer_v2.py:133: UserWarning: TimeXer is an experimental model implemented on TslibBaseModelV2. It is an unstable version and maybe subject to unannouced changes.Please use with caution. Feedback on the design and implementation iswelcome. On the issue #1833 - https://github.com/sktime/pytorch-forecasting/issues/1833\n",
-      "  warn.warn(\n",
-      "/home/aryan/pytorch-forecasting/pytorch_forecasting/models/timexer/_timexer_v2.py:179: UserWarning: Context length (30) is not divisible by patch length. This may lead to unexpected behavior, as sometime steps will not be used in the model.\n",
-      "  warn.warn(\n"
+      "{'loss': MAE(), 'hidden_size': 64, 'n_heads': 4, 'e_layers': 2, 'd_ff': 256, 'dropout': 0.1, 'patch_length': 4, 'logging_metrics': [MAE(), SMAPE()], 'optimizer': 'adam', 'optimizer_params': {'lr': 0.001}, 'lr_scheduler': 'reduce_lr_on_plateau', 'lr_scheduler_params': {'mode': 'min', 'factor': 0.5, 'patience': 5}}\n"
      ]
     }
    ],
    "source": [
-    "model1 = TimeXer(\n",
-    "    loss=MAE(),\n",
-    "    hidden_size=64,\n",
-    "    nhead=4,\n",
-    "    e_layers=2,\n",
-    "    d_ff=256,\n",
-    "    dropout=0.1,\n",
-    "    patch_length=4,\n",
-    "    logging_metrics=[MAE(), SMAPE()],\n",
-    "    optimizer=\"adam\",\n",
-    "    optimizer_params={\"lr\": 1e-3},\n",
-    "    lr_scheduler=\"reduce_lr_on_plateau\",\n",
-    "    lr_scheduler_params={\n",
-    "        \"mode\": \"min\",\n",
-    "        \"factor\": 0.5,\n",
-    "        \"patience\": 5,\n",
-    "    },\n",
-    "    metadata=data_module.metadata,\n",
-    ")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 8,
-   "id": "0aa21f48",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "model2 = TimeXer(\n",
-    "    loss=QuantileLoss(quantiles=[0.1, 0.5, 0.9]),  # quantiles of 0.1, 0.5 and 0.9 used.\n",
-    "    hidden_size=64,\n",
-    "    nhead=4,\n",
-    "    e_layers=2,\n",
-    "    d_ff=256,\n",
-    "    dropout=0.1,\n",
-    "    patch_length=4,\n",
-    "    optimizer=\"adam\",\n",
-    "    optimizer_params={\"lr\": 1e-3},\n",
-    "    lr_scheduler=\"reduce_lr_on_plateau\",\n",
-    "    lr_scheduler_params={\n",
-    "        \"mode\": \"min\",\n",
-    "        \"factor\": 0.5,\n",
-    "        \"patience\": 5,\n",
-    "    },\n",
-    "    metadata=data_module.metadata,\n",
+    "pkg_mae = TimeXer_pkg_v2(\n",
+    "    model_cfg=dict(\n",
+    "        loss=MAE(),\n",
+    "        hidden_size=64,\n",
+    "        n_heads=4,\n",
+    "        e_layers=2,\n",
+    "        d_ff=256,\n",
+    "        dropout=0.1,\n",
+    "        patch_length=4,\n",
+    "        logging_metrics=[MAE(), SMAPE()],\n",
+    "        optimizer=\"adam\",\n",
+    "        optimizer_params={\"lr\": 1e-3},\n",
+    "        lr_scheduler=\"reduce_lr_on_plateau\",\n",
+    "        lr_scheduler_params={\"mode\": \"min\", \"factor\": 0.5, \"patience\": 5},\n",
+    "    ),\n",
+    "    datamodule_cfg=datamodule_cfg,\n",
+    "    trainer_cfg=trainer_cfg,\n",
     ")"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 9,
-   "id": "02605f9b",
-   "metadata": {},
+   "execution_count": 7,
+   "id": "a051e0bf",
+   "metadata": {
+    "execution": {
+     "iopub.execute_input": "2026-08-01T23:05:34.350621Z",
+     "iopub.status.busy": "2026-08-01T23:05:34.350544Z",
+     "iopub.status.idle": "2026-08-01T23:05:44.127364Z",
+     "shell.execute_reply": "2026-08-01T23:05:44.126961Z"
+    }
+   },
    "outputs": [
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "GPU available: True (cuda), used: True\n",
-      "TPU available: False, using: 0 TPU cores\n",
-      "HPU available: False, using: 0 HPUs\n",
-      "GPU available: True (cuda), used: True\n",
-      "TPU available: False, using: 0 TPU cores\n",
-      "HPU available: False, using: 0 HPUs\n"
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/data/data_module/_tslib_data_module.py:331: UserWarning: TslibDataModule is experimental and subject to change. The API is not stable and may change without prior warning.\n",
+      "  warnings.warn(\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/models/base/_base_model_v2.py:85: UserWarning: The Model 'TimeXer' is part of an experimental reworkof the pytorch-forecasting model layer, scheduled for release with v2.0.0. The API is not stable and may change without prior warning. This class is intended for beta testing and as a basic skeleton, but not for stable production use. Feedback and suggestions are very welcome in pytorch-forecasting issue 1736, https://github.com/sktime/pytorch-forecasting/issues/1736\n",
+      "  warn(\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/models/base/_tslib_base_model_v2.py:63: UserWarning: The Model 'TimeXer' is part of an experimental implementationof the pytorch-forecasting model layer for Time Series Library, scheduledfor release with v2.0.0. The API is not stableand may change without prior warning. This class is intended for betatesting, not for stable production use.\n",
+      "  warn(\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/models/timexer/_timexer_v2.py:134: UserWarning: TimeXer is an experimental model implemented on TslibBaseModelV2. It is an unstable version and maybe subject to unannouced changes.Please use with caution. Feedback on the design and implementation iswelcome. On the issue #1833 - https://github.com/sktime/pytorch-forecasting/issues/1833\n",
+      "  warn.warn(\n",
+      "GPU available: True (mps), used: False\n"
      ]
-    }
-   ],
-   "source": [
-    "from lightning.pytorch import Trainer\n",
-    "\n",
-    "trainer1 = Trainer(\n",
-    "    max_epochs=5,\n",
-    "    accelerator=\"auto\",\n",
-    "    devices=1,\n",
-    "    enable_progress_bar=True,\n",
-    "    enable_model_summary=True,\n",
-    ")\n",
-    "\n",
-    "trainer2 = Trainer(\n",
-    "    max_epochs=4,\n",
-    "    accelerator=\"auto\",\n",
-    "    devices=1,\n",
-    "    enable_progress_bar=True,\n",
-    "    enable_model_summary=True,\n",
-    ")"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "e22756b2",
-   "metadata": {},
-   "source": [
-    "## Fit the trainer on the model and feed data using the data module"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 10,
-   "id": "6e9117d2",
-   "metadata": {},
-   "outputs": [
+    },
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "You are using a CUDA device ('NVIDIA GeForce RTX 4050 Laptop GPU') that has Tensor Cores. To properly utilize them, you should set `torch.set_float32_matmul_precision('medium' | 'high')` which will trade-off precision for performance. For more details, read https://pytorch.org/docs/stable/generated/torch.set_float32_matmul_precision.html#torch.set_float32_matmul_precision\n",
-      "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n",
-      "\n",
-      "  | Name         | Type                   | Params | Mode \n",
-      "----------------------------------------------------------------\n",
-      "0 | loss         | MAE                    | 0      | train\n",
-      "1 | en_embedding | EnEmbedding            | 320    | train\n",
-      "2 | ex_embedding | DataEmbedding_inverted | 2.0 K  | train\n",
-      "3 | encoder      | Encoder                | 133 K  | train\n",
-      "4 | head         | FlattenHead            | 513    | train\n",
-      "----------------------------------------------------------------\n",
-      "136 K     Trainable params\n",
-      "0         Non-trainable params\n",
-      "136 K     Total params\n",
-      "0.546     Total estimated model params size (MB)\n",
-      "57        Modules in train mode\n",
-      "0         Modules in eval mode\n"
+      "TPU available: False, using: 0 TPU cores\n"
      ]
     },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "f26d868819404cb0a48cc030aefef48c",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Sanity Checking: |          | 0/? [00:00<?, ?it/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "/home/aryan/pytorch-forecasting/.venv/lib/python3.12/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:425: The 'val_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=15` in the `DataLoader` to improve performance.\n",
-      "/home/aryan/pytorch-forecasting/.venv/lib/python3.12/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:425: The 'train_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=15` in the `DataLoader` to improve performance.\n",
-      "/home/aryan/pytorch-forecasting/.venv/lib/python3.12/site-packages/lightning/pytorch/loops/fit_loop.py:310: The number of training batches (42) is smaller than the logging interval Trainer(log_every_n_steps=50). Set a lower value for log_every_n_steps if you want to see logs for the training epoch.\n"
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/setup.py:175: GPU available but not used. You can set it by doing `Trainer(accelerator='gpu')`.\n",
+      "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n"
      ]
     },
     {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "db0fdcf01b784ddf8bf0d93fe1cc23e4",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Training: |          | 0/? [00:00<?, ?it/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "26073d89e3014850ba0c4d7f01104047",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Validation: |          | 0/? [00:00<?, ?it/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "bf42294210dc4d53a066967bc54b1282",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Validation: |          | 0/? [00:00<?, ?it/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n"
+     ]
     },
     {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "27ed3f4556284bf0a918d22554e8e9c2",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Validation: |          | 0/? [00:00<?, ?it/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "\n",
+      "  | Name            | Type                   | Params | Mode  | FLOPs\n",
+      "---------------------------------------------------------------------------\n",
+      "0 | loss            | MAE                    | 0      | train | 0    \n",
+      "1 | logging_metrics | ModuleList             | 0      | train | 0    \n",
+      "2 | en_embedding    | EnEmbedding            | 320    | train | 0    \n",
+      "3 | ex_embedding    | DataEmbedding_inverted | 2.1 K  | train | 0    \n",
+      "4 | encoder         | Encoder                | 133 K  | train | 0    \n",
+      "5 | head            | FlattenHead            | 577    | train | 0    \n",
+      "---------------------------------------------------------------------------\n",
+      "136 K     Trainable params\n",
+      "0         Non-trainable params\n",
+      "136 K     Total params\n",
+      "0.547     Total estimated model params size (MB)\n",
+      "60        Modules in train mode\n",
+      "0         Modules in eval mode\n",
+      "0         Total Flops\n"
+     ]
     },
     {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "0e338f265f554e6d96020135a3f64f64",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Validation: |          | 0/? [00:00<?, ?it/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'val_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n"
+     ]
     },
     {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "46bbd28e869740449aad2c7c1aaf5da7",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Validation: |          | 0/? [00:00<?, ?it/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'train_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/loops/fit_loop.py:321: The number of training batches (38) is smaller than the logging interval Trainer(log_every_n_steps=50). Set a lower value for log_every_n_steps if you want to see logs for the training epoch.\n"
+     ]
     },
     {
      "name": "stderr",
@@ -654,603 +580,504 @@
     }
    ],
    "source": [
-    "trainer1.fit(model1, data_module)"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "e8fb4f31",
-   "metadata": {},
-   "source": [
-    "Now let us train the model using `QuantileLoss`."
+    "pkg_mae.fit(dataset, save_ckpt=False)"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 11,
-   "id": "3c67d86f",
-   "metadata": {},
+   "execution_count": 8,
+   "id": "cb5aa00f",
+   "metadata": {
+    "execution": {
+     "iopub.execute_input": "2026-08-01T23:05:44.128653Z",
+     "iopub.status.busy": "2026-08-01T23:05:44.128567Z",
+     "iopub.status.idle": "2026-08-01T23:05:45.858557Z",
+     "shell.execute_reply": "2026-08-01T23:05:45.858157Z"
+    }
+   },
    "outputs": [
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n",
-      "\n",
-      "  | Name         | Type                   | Params | Mode \n",
-      "----------------------------------------------------------------\n",
-      "0 | loss         | QuantileLoss           | 0      | train\n",
-      "1 | en_embedding | EnEmbedding            | 320    | train\n",
-      "2 | ex_embedding | DataEmbedding_inverted | 2.0 K  | train\n",
-      "3 | encoder      | Encoder                | 133 K  | train\n",
-      "4 | head         | FlattenHead            | 1.5 K  | train\n",
-      "----------------------------------------------------------------\n",
-      "137 K     Trainable params\n",
-      "0         Non-trainable params\n",
-      "137 K     Total params\n",
-      "0.550     Total estimated model params size (MB)\n",
-      "57        Modules in train mode\n",
-      "0         Modules in eval mode\n"
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/data/data_module/_tslib_data_module.py:331: UserWarning: TslibDataModule is experimental and subject to change. The API is not stable and may change without prior warning.\n",
+      "  warnings.warn(\n",
+      "GPU available: True (mps), used: False\n"
      ]
     },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "acba85c2e99b4dc391315e10471e1e8c",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Sanity Checking: |          | 0/? [00:00<?, ?it/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "/home/aryan/pytorch-forecasting/.venv/lib/python3.12/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:425: The 'val_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=15` in the `DataLoader` to improve performance.\n",
-      "/home/aryan/pytorch-forecasting/.venv/lib/python3.12/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:425: The 'train_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=15` in the `DataLoader` to improve performance.\n",
-      "/home/aryan/pytorch-forecasting/.venv/lib/python3.12/site-packages/lightning/pytorch/loops/fit_loop.py:310: The number of training batches (42) is smaller than the logging interval Trainer(log_every_n_steps=50). Set a lower value for log_every_n_steps if you want to see logs for the training epoch.\n"
+      "TPU available: False, using: 0 TPU cores\n"
      ]
     },
     {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "77d8d477ad3040aea894b76ca2df8130",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Training: |          | 0/? [00:00<?, ?it/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "f1bbc1f476b046e788d827e5dc02d888",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Validation: |          | 0/? [00:00<?, ?it/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "1233ff4c4fbe4bde9e7717c54e28567d",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Validation: |          | 0/? [00:00<?, ?it/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/setup.py:175: GPU available but not used. You can set it by doing `Trainer(accelerator='gpu')`.\n",
+      "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n"
+     ]
     },
     {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "334cb342267b47b1b4e1e2ca1038569f",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Validation: |          | 0/? [00:00<?, ?it/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n"
+     ]
     },
     {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "26260a641c604444811ec1bf00c80edb",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Validation: |          | 0/? [00:00<?, ?it/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'predict_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n"
+     ]
     },
     {
-     "name": "stderr",
+     "name": "stdout",
      "output_type": "stream",
      "text": [
-      "`Trainer.fit` stopped: `max_epochs=4` reached.\n"
+      "Point prediction shape: torch.Size([1700, 1])\n",
+      "Point prediction: tensor([[ 0.2108],\n",
+      "        [ 0.3451],\n",
+      "        [ 0.4759],\n",
+      "        ...,\n",
+      "        [-0.0950],\n",
+      "        [-0.3101],\n",
+      "        [-0.4429]])\n"
      ]
     }
    ],
    "source": [
-    "trainer2.fit(model2, data_module)"
+    "preds_mae = pkg_mae.predict(dataset, mode=\"prediction\", trainer_kwargs={\"accelerator\": \"cpu\", \"enable_progress_bar\": False})\n",
+    "print(\"Point prediction shape:\", preds_mae[\"prediction\"].shape)\n",
+    "print(\"Point prediction:\", preds_mae[\"prediction\"])"
    ]
   },
   {
    "cell_type": "markdown",
-   "id": "16e2d445",
+   "id": "4bc8ddd8",
    "metadata": {},
    "source": [
-    "## Test the model"
+    "## TimeXer with QuantileLoss (interval forecast)\n",
+    "\n",
+    "Swap the loss to `QuantileLoss` to get quantile / prediction-interval outputs."
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 12,
-   "id": "dbf1ace6",
-   "metadata": {},
+   "execution_count": 9,
+   "id": "27a5689f",
+   "metadata": {
+    "execution": {
+     "iopub.execute_input": "2026-08-01T23:05:45.859695Z",
+     "iopub.status.busy": "2026-08-01T23:05:45.859624Z",
+     "iopub.status.idle": "2026-08-01T23:05:45.862143Z",
+     "shell.execute_reply": "2026-08-01T23:05:45.861730Z"
+    }
+   },
    "outputs": [
     {
-     "name": "stderr",
+     "name": "stdout",
      "output_type": "stream",
      "text": [
-      "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n",
-      "/home/aryan/pytorch-forecasting/.venv/lib/python3.12/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:425: The 'test_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=15` in the `DataLoader` to improve performance.\n"
-     ]
-    },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "d8e51a5bb1564edab4e0d32743a16299",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Testing: |          | 0/? [00:00<?, ?it/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────\n",
-      "       Test metric             DataLoader 0\n",
-      "────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────\n",
-      "        test_MAE            0.13894660770893097\n",
-      "       test_SMAPE           0.40041154623031616\n",
-      "        test_loss           0.13894660770893097\n",
-      "────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────\n"
+      "{'loss': QuantileLoss(quantiles=[0.1, 0.5, 0.9]), 'hidden_size': 64, 'n_heads': 4, 'e_layers': 2, 'd_ff': 256, 'dropout': 0.1, 'patch_length': 4, 'optimizer': 'adam', 'optimizer_params': {'lr': 0.001}, 'lr_scheduler': 'reduce_lr_on_plateau', 'lr_scheduler_params': {'mode': 'min', 'factor': 0.5, 'patience': 5}}\n"
      ]
     }
    ],
    "source": [
-    "test_metrics = trainer1.test(model1, data_module)"
+    "pkg_q = TimeXer_pkg_v2(\n",
+    "    model_cfg=dict(\n",
+    "        loss=QuantileLoss(quantiles=[0.1, 0.5, 0.9]),\n",
+    "        hidden_size=64,\n",
+    "        n_heads=4,\n",
+    "        e_layers=2,\n",
+    "        d_ff=256,\n",
+    "        dropout=0.1,\n",
+    "        patch_length=4,\n",
+    "        # NOTE: no logging_metrics here — point metrics (MAE/SMAPE) cannot be\n",
+    "        # computed on QuantileLoss output (extra quantile dim) and would raise\n",
+    "        # \"Prediction should only have one extra dimension\" during fit.\n",
+    "        optimizer=\"adam\",\n",
+    "        optimizer_params={\"lr\": 1e-3},\n",
+    "        lr_scheduler=\"reduce_lr_on_plateau\",\n",
+    "        lr_scheduler_params={\"mode\": \"min\", \"factor\": 0.5, \"patience\": 5},\n",
+    "    ),\n",
+    "    datamodule_cfg=datamodule_cfg,\n",
+    "    trainer_cfg=trainer_cfg,\n",
+    ")"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 13,
-   "id": "250b128a",
-   "metadata": {},
+   "execution_count": 10,
+   "id": "c89820fc",
+   "metadata": {
+    "execution": {
+     "iopub.execute_input": "2026-08-01T23:05:45.863145Z",
+     "iopub.status.busy": "2026-08-01T23:05:45.863067Z",
+     "iopub.status.idle": "2026-08-01T23:05:55.640465Z",
+     "shell.execute_reply": "2026-08-01T23:05:55.639906Z"
+    }
+   },
    "outputs": [
     {
-     "data": {
-      "text/plain": [
-       "TimeXer(\n",
-       "  (loss): MAE()\n",
-       "  (en_embedding): EnEmbedding(\n",
-       "    (value_embedding): Linear(in_features=4, out_features=64, bias=False)\n",
-       "    (position_embedding): PositionalEmbedding()\n",
-       "    (dropout): Dropout(p=0.1, inplace=False)\n",
-       "  )\n",
-       "  (ex_embedding): DataEmbedding_inverted(\n",
-       "    (value_embedding): Linear(in_features=30, out_features=64, bias=True)\n",
-       "    (dropout): Dropout(p=0.1, inplace=False)\n",
-       "  )\n",
-       "  (encoder): Encoder(\n",
-       "    (layers): ModuleList(\n",
-       "      (0-1): 2 x EncoderLayer(\n",
-       "        (self_attention): AttentionLayer(\n",
-       "          (inner_attention): FullAttention(\n",
-       "            (dropout): Dropout(p=0.1, inplace=False)\n",
-       "          )\n",
-       "          (query_projection): Linear(in_features=64, out_features=64, bias=True)\n",
-       "          (key_projection): Linear(in_features=64, out_features=64, bias=True)\n",
-       "          (value_projection): Linear(in_features=64, out_features=64, bias=True)\n",
-       "          (out_projection): Linear(in_features=64, out_features=64, bias=True)\n",
-       "        )\n",
-       "        (cross_attention): AttentionLayer(\n",
-       "          (inner_attention): FullAttention(\n",
-       "            (dropout): Dropout(p=0.1, inplace=False)\n",
-       "          )\n",
-       "          (query_projection): Linear(in_features=64, out_features=64, bias=True)\n",
-       "          (key_projection): Linear(in_features=64, out_features=64, bias=True)\n",
-       "          (value_projection): Linear(in_features=64, out_features=64, bias=True)\n",
-       "          (out_projection): Linear(in_features=64, out_features=64, bias=True)\n",
-       "        )\n",
-       "        (conv1): Conv1d(64, 256, kernel_size=(1,), stride=(1,))\n",
-       "        (conv2): Conv1d(256, 64, kernel_size=(1,), stride=(1,))\n",
-       "        (norm1): LayerNorm((64,), eps=1e-05, elementwise_affine=True)\n",
-       "        (norm2): LayerNorm((64,), eps=1e-05, elementwise_affine=True)\n",
-       "        (norm3): LayerNorm((64,), eps=1e-05, elementwise_affine=True)\n",
-       "        (dropout): Dropout(p=0.1, inplace=False)\n",
-       "      )\n",
-       "    )\n",
-       "    (norm): LayerNorm((64,), eps=1e-05, elementwise_affine=True)\n",
-       "  )\n",
-       "  (head): FlattenHead(\n",
-       "    (flatten): Flatten(start_dim=-2, end_dim=-1)\n",
-       "    (linear): Linear(in_features=512, out_features=1, bias=True)\n",
-       "    (dropout): Dropout(p=0.1, inplace=False)\n",
-       "  )\n",
-       ")"
-      ]
-     },
-     "execution_count": 13,
-     "metadata": {},
-     "output_type": "execute_result"
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/data/data_module/_tslib_data_module.py:331: UserWarning: TslibDataModule is experimental and subject to change. The API is not stable and may change without prior warning.\n",
+      "  warnings.warn(\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/models/base/_base_model_v2.py:85: UserWarning: The Model 'TimeXer' is part of an experimental reworkof the pytorch-forecasting model layer, scheduled for release with v2.0.0. The API is not stable and may change without prior warning. This class is intended for beta testing and as a basic skeleton, but not for stable production use. Feedback and suggestions are very welcome in pytorch-forecasting issue 1736, https://github.com/sktime/pytorch-forecasting/issues/1736\n",
+      "  warn(\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/models/base/_tslib_base_model_v2.py:63: UserWarning: The Model 'TimeXer' is part of an experimental implementationof the pytorch-forecasting model layer for Time Series Library, scheduledfor release with v2.0.0. The API is not stableand may change without prior warning. This class is intended for betatesting, not for stable production use.\n",
+      "  warn(\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/models/timexer/_timexer_v2.py:134: UserWarning: TimeXer is an experimental model implemented on TslibBaseModelV2. It is an unstable version and maybe subject to unannouced changes.Please use with caution. Feedback on the design and implementation iswelcome. On the issue #1833 - https://github.com/sktime/pytorch-forecasting/issues/1833\n",
+      "  warn.warn(\n",
+      "GPU available: True (mps), used: False\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "TPU available: False, using: 0 TPU cores\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/setup.py:175: GPU available but not used. You can set it by doing `Trainer(accelerator='gpu')`.\n",
+      "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "\n",
+      "  | Name            | Type                   | Params | Mode  | FLOPs\n",
+      "---------------------------------------------------------------------------\n",
+      "0 | loss            | QuantileLoss           | 0      | train | 0    \n",
+      "1 | logging_metrics | ModuleList             | 0      | train | 0    \n",
+      "2 | en_embedding    | EnEmbedding            | 320    | train | 0    \n",
+      "3 | ex_embedding    | DataEmbedding_inverted | 2.1 K  | train | 0    \n",
+      "4 | encoder         | Encoder                | 133 K  | train | 0    \n",
+      "5 | head            | FlattenHead            | 1.7 K  | train | 0    \n",
+      "---------------------------------------------------------------------------\n",
+      "137 K     Trainable params\n",
+      "0         Non-trainable params\n",
+      "137 K     Total params\n",
+      "0.551     Total estimated model params size (MB)\n",
+      "58        Modules in train mode\n",
+      "0         Modules in eval mode\n",
+      "0         Total Flops\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'val_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'train_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/loops/fit_loop.py:321: The number of training batches (38) is smaller than the logging interval Trainer(log_every_n_steps=50). Set a lower value for log_every_n_steps if you want to see logs for the training epoch.\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "`Trainer.fit` stopped: `max_epochs=5` reached.\n"
+     ]
     }
    ],
    "source": [
-    "model1.eval()"
+    "pkg_q.fit(dataset, save_ckpt=False)"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 14,
-   "id": "f730b49a",
-   "metadata": {},
+   "execution_count": 11,
+   "id": "8240e1fb",
+   "metadata": {
+    "execution": {
+     "iopub.execute_input": "2026-08-01T23:05:55.641813Z",
+     "iopub.status.busy": "2026-08-01T23:05:55.641734Z",
+     "iopub.status.idle": "2026-08-01T23:05:57.364930Z",
+     "shell.execute_reply": "2026-08-01T23:05:57.364434Z"
+    }
+   },
    "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/data/data_module/_tslib_data_module.py:331: UserWarning: TslibDataModule is experimental and subject to change. The API is not stable and may change without prior warning.\n",
+      "  warnings.warn(\n",
+      "GPU available: True (mps), used: False\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "TPU available: False, using: 0 TPU cores\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/setup.py:175: GPU available but not used. You can set it by doing `Trainer(accelerator='gpu')`.\n",
+      "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'predict_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n"
+     ]
+    },
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Prediction: tensor([[[ 0.1253]],\n",
-      "\n",
-      "        [[ 0.2623]],\n",
-      "\n",
-      "        [[ 0.4591]],\n",
-      "\n",
-      "        [[ 0.6304]],\n",
-      "\n",
-      "        [[ 0.7916]],\n",
-      "\n",
-      "        [[ 0.9132]],\n",
-      "\n",
-      "        [[ 1.0252]],\n",
-      "\n",
-      "        [[ 1.1069]],\n",
-      "\n",
-      "        [[ 1.1370]],\n",
-      "\n",
-      "        [[ 1.1317]],\n",
+      "Quantile prediction shape (batch, pred_len, n_quantiles): torch.Size([1700, 1, 3])\n",
+      "Quantile prediction: tensor([[[ 0.1863,  0.2792,  0.4827]],\n",
       "\n",
-      "        [[ 1.0659]],\n",
+      "        [[ 0.3067,  0.4198,  0.6217]],\n",
       "\n",
-      "        [[ 0.9617]],\n",
+      "        [[ 0.4261,  0.5496,  0.7150]],\n",
       "\n",
-      "        [[ 0.8297]],\n",
+      "        ...,\n",
       "\n",
-      "        [[ 0.6622]],\n",
+      "        [[-0.1460,  0.1544,  0.1308]],\n",
       "\n",
-      "        [[ 0.5254]],\n",
+      "        [[-0.3319, -0.0281, -0.0082]],\n",
       "\n",
-      "        [[ 0.3310]],\n",
-      "\n",
-      "        [[ 0.1579]],\n",
-      "\n",
-      "        [[-0.0506]],\n",
-      "\n",
-      "        [[-0.1999]],\n",
-      "\n",
-      "        [[ 0.0740]],\n",
-      "\n",
-      "        [[ 0.2787]],\n",
-      "\n",
-      "        [[ 0.4506]],\n",
-      "\n",
-      "        [[ 0.6381]],\n",
-      "\n",
-      "        [[ 0.7867]],\n",
-      "\n",
-      "        [[ 0.9343]],\n",
-      "\n",
-      "        [[ 1.0370]],\n",
-      "\n",
-      "        [[ 1.1286]],\n",
-      "\n",
-      "        [[ 1.1737]],\n",
-      "\n",
-      "        [[ 1.1367]],\n",
-      "\n",
-      "        [[ 1.0765]],\n",
-      "\n",
-      "        [[ 0.9569]],\n",
-      "\n",
-      "        [[ 0.8583]]])\n"
+      "        [[-0.5041, -0.2163, -0.1586]]])\n"
      ]
     }
    ],
    "source": [
-    "with torch.no_grad():\n",
-    "    test_batch = next(iter(data_module.test_dataloader()))\n",
-    "    x_test, y_test = test_batch\n",
-    "    y_pred = model1(x_test)\n",
-    "\n",
-    "    print(\"Prediction:\", y_pred[\"prediction\"])"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 15,
-   "id": "e316c047",
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "torch.Size([32, 1, 1])"
-      ]
-     },
-     "execution_count": 15,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "y_pred[\"prediction\"].shape"
+    "preds_q = pkg_q.predict(dataset, mode=\"quantiles\", trainer_kwargs={\"accelerator\": \"cpu\", \"enable_progress_bar\": False})\n",
+    "print(\"Quantile prediction shape (batch, pred_len, n_quantiles):\", preds_q[\"prediction\"].shape)\n",
+    "print(\"Quantile prediction:\", preds_q[\"prediction\"])"
    ]
   },
   {
    "cell_type": "markdown",
-   "id": "a01927d4",
+   "id": "cb9492cf",
    "metadata": {},
    "source": [
-    "Let us do the same for `QuantileLoss` predictions."
+    "## Same pipeline, different model: DLinear\n",
+    "\n",
+    "`DLinear` is the other model built on the `tslib` base, so it runs through the exact same pipeline — only the package class and its `model_cfg` change."
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 16,
-   "id": "22bd191f",
-   "metadata": {},
+   "execution_count": 12,
+   "id": "98090c16",
+   "metadata": {
+    "execution": {
+     "iopub.execute_input": "2026-08-01T23:05:57.366294Z",
+     "iopub.status.busy": "2026-08-01T23:05:57.366198Z",
+     "iopub.status.idle": "2026-08-01T23:06:03.812697Z",
+     "shell.execute_reply": "2026-08-01T23:06:03.812248Z"
+    }
+   },
    "outputs": [
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n",
-      "/home/aryan/pytorch-forecasting/.venv/lib/python3.12/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:425: The 'test_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=15` in the `DataLoader` to improve performance.\n"
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/data/data_module/_tslib_data_module.py:331: UserWarning: TslibDataModule is experimental and subject to change. The API is not stable and may change without prior warning.\n",
+      "  warnings.warn(\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/models/base/_base_model_v2.py:85: UserWarning: The Model 'DLinear' is part of an experimental reworkof the pytorch-forecasting model layer, scheduled for release with v2.0.0. The API is not stable and may change without prior warning. This class is intended for beta testing and as a basic skeleton, but not for stable production use. Feedback and suggestions are very welcome in pytorch-forecasting issue 1736, https://github.com/sktime/pytorch-forecasting/issues/1736\n",
+      "  warn(\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/models/base/_tslib_base_model_v2.py:63: UserWarning: The Model 'DLinear' is part of an experimental implementationof the pytorch-forecasting model layer for Time Series Library, scheduledfor release with v2.0.0. The API is not stableand may change without prior warning. This class is intended for betatesting, not for stable production use.\n",
+      "  warn(\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/models/dlinear/_dlinear_v2.py:96: UserWarning: DLinear is an experimental model implemented on TslibBaseModelV2. It is an unstable version and may be subject to unannounced changes. Please use with caution.\n",
+      "  warnings.warn(\n",
+      "GPU available: True (mps), used: False\n"
      ]
     },
     {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "f29e92e8bae645c0bc6692d18c8bdd1f",
-       "version_major": 2,
-       "version_minor": 0
-      },
-      "text/plain": [
-       "Testing: |          | 0/? [00:00<?, ?it/s]"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "TPU available: False, using: 0 TPU cores\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/setup.py:175: GPU available but not used. You can set it by doing `Trainer(accelerator='gpu')`.\n",
+      "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "\n",
+      "  | Name            | Type                | Params | Mode  | FLOPs\n",
+      "------------------------------------------------------------------------\n",
+      "0 | loss            | MAE                 | 0      | train | 0    \n",
+      "1 | logging_metrics | ModuleList          | 0      | train | 0    \n",
+      "2 | decomposition   | SeriesDecomposition | 0      | train | 0    \n",
+      "3 | linear_seasonal | Linear              | 33     | train | 0    \n",
+      "4 | linear_trend    | Linear              | 33     | train | 0    \n",
+      "------------------------------------------------------------------------\n",
+      "66        Trainable params\n",
+      "0         Non-trainable params\n",
+      "66        Total params\n",
+      "0.000     Total estimated model params size (MB)\n",
+      "9         Modules in train mode\n",
+      "0         Modules in eval mode\n",
+      "0         Total Flops\n"
+     ]
     },
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────\n",
-      "       Test metric             DataLoader 0\n",
-      "────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────\n",
-      "        test_loss           0.07047828286886215\n",
-      "────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────\n"
+      "{'loss': MAE(), 'moving_avg': 25, 'individual': False, 'logging_metrics': [MAE(), SMAPE()], 'optimizer': 'adam', 'optimizer_params': {'lr': 0.001}}\n"
      ]
-    }
-   ],
-   "source": [
-    "test_metrics = trainer2.test(model2, data_module)"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 17,
-   "id": "a1d857db",
-   "metadata": {},
-   "outputs": [
+    },
     {
-     "data": {
-      "text/plain": [
-       "TimeXer(\n",
-       "  (loss): QuantileLoss(quantiles=[0.1, 0.5, 0.9])\n",
-       "  (en_embedding): EnEmbedding(\n",
-       "    (value_embedding): Linear(in_features=4, out_features=64, bias=False)\n",
-       "    (position_embedding): PositionalEmbedding()\n",
-       "    (dropout): Dropout(p=0.1, inplace=False)\n",
-       "  )\n",
-       "  (ex_embedding): DataEmbedding_inverted(\n",
-       "    (value_embedding): Linear(in_features=30, out_features=64, bias=True)\n",
-       "    (dropout): Dropout(p=0.1, inplace=False)\n",
-       "  )\n",
-       "  (encoder): Encoder(\n",
-       "    (layers): ModuleList(\n",
-       "      (0-1): 2 x EncoderLayer(\n",
-       "        (self_attention): AttentionLayer(\n",
-       "          (inner_attention): FullAttention(\n",
-       "            (dropout): Dropout(p=0.1, inplace=False)\n",
-       "          )\n",
-       "          (query_projection): Linear(in_features=64, out_features=64, bias=True)\n",
-       "          (key_projection): Linear(in_features=64, out_features=64, bias=True)\n",
-       "          (value_projection): Linear(in_features=64, out_features=64, bias=True)\n",
-       "          (out_projection): Linear(in_features=64, out_features=64, bias=True)\n",
-       "        )\n",
-       "        (cross_attention): AttentionLayer(\n",
-       "          (inner_attention): FullAttention(\n",
-       "            (dropout): Dropout(p=0.1, inplace=False)\n",
-       "          )\n",
-       "          (query_projection): Linear(in_features=64, out_features=64, bias=True)\n",
-       "          (key_projection): Linear(in_features=64, out_features=64, bias=True)\n",
-       "          (value_projection): Linear(in_features=64, out_features=64, bias=True)\n",
-       "          (out_projection): Linear(in_features=64, out_features=64, bias=True)\n",
-       "        )\n",
-       "        (conv1): Conv1d(64, 256, kernel_size=(1,), stride=(1,))\n",
-       "        (conv2): Conv1d(256, 64, kernel_size=(1,), stride=(1,))\n",
-       "        (norm1): LayerNorm((64,), eps=1e-05, elementwise_affine=True)\n",
-       "        (norm2): LayerNorm((64,), eps=1e-05, elementwise_affine=True)\n",
-       "        (norm3): LayerNorm((64,), eps=1e-05, elementwise_affine=True)\n",
-       "        (dropout): Dropout(p=0.1, inplace=False)\n",
-       "      )\n",
-       "    )\n",
-       "    (norm): LayerNorm((64,), eps=1e-05, elementwise_affine=True)\n",
-       "  )\n",
-       "  (head): FlattenHead(\n",
-       "    (flatten): Flatten(start_dim=-2, end_dim=-1)\n",
-       "    (linear): Linear(in_features=512, out_features=3, bias=True)\n",
-       "    (dropout): Dropout(p=0.1, inplace=False)\n",
-       "  )\n",
-       ")"
-      ]
-     },
-     "execution_count": 17,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "model2.eval()"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 18,
-   "id": "52e2a36a",
-   "metadata": {},
-   "outputs": [
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'val_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'train_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/loops/fit_loop.py:321: The number of training batches (38) is smaller than the logging interval Trainer(log_every_n_steps=50). Set a lower value for log_every_n_steps if you want to see logs for the training epoch.\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "`Trainer.fit` stopped: `max_epochs=5` reached.\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/data/data_module/_tslib_data_module.py:331: UserWarning: TslibDataModule is experimental and subject to change. The API is not stable and may change without prior warning.\n",
+      "  warnings.warn(\n",
+      "GPU available: True (mps), used: False\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "TPU available: False, using: 0 TPU cores\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/setup.py:175: GPU available but not used. You can set it by doing `Trainer(accelerator='gpu')`.\n",
+      "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'predict_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n"
+     ]
+    },
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Prediction: tensor([[[-0.1025, -0.0489,  0.0900]],\n",
-      "\n",
-      "        [[ 0.0680,  0.0936,  0.2504]],\n",
-      "\n",
-      "        [[ 0.2310,  0.2605,  0.4298]],\n",
-      "\n",
-      "        [[ 0.3604,  0.3968,  0.5679]],\n",
-      "\n",
-      "        [[ 0.4935,  0.5408,  0.7165]],\n",
-      "\n",
-      "        [[ 0.6274,  0.6697,  0.8745]],\n",
-      "\n",
-      "        [[ 0.7192,  0.7940,  0.9812]],\n",
-      "\n",
-      "        [[ 0.7555,  0.8650,  1.0313]],\n",
-      "\n",
-      "        [[ 0.7602,  0.8706,  1.0427]],\n",
-      "\n",
-      "        [[ 0.7532,  0.8524,  1.0308]],\n",
-      "\n",
-      "        [[ 0.7003,  0.7784,  0.9995]],\n",
-      "\n",
-      "        [[ 0.5987,  0.6807,  0.9390]],\n",
-      "\n",
-      "        [[ 0.4757,  0.5814,  0.7966]],\n",
-      "\n",
-      "        [[ 0.3432,  0.4587,  0.6614]],\n",
-      "\n",
-      "        [[ 0.1659,  0.2931,  0.5039]],\n",
-      "\n",
-      "        [[-0.0338,  0.0983,  0.3208]],\n",
-      "\n",
-      "        [[-0.1989, -0.0829,  0.1821]],\n",
-      "\n",
-      "        [[-0.3732, -0.2402,  0.0121]],\n",
-      "\n",
-      "        [[-0.5151, -0.3600, -0.1606]],\n",
-      "\n",
-      "        [[-0.0789, -0.0406,  0.0908]],\n",
-      "\n",
-      "        [[ 0.0495,  0.0830,  0.2585]],\n",
-      "\n",
-      "        [[ 0.2185,  0.2520,  0.4223]],\n",
-      "\n",
-      "        [[ 0.3870,  0.4209,  0.5818]],\n",
-      "\n",
-      "        [[ 0.5243,  0.5766,  0.7636]],\n",
-      "\n",
-      "        [[ 0.6293,  0.6854,  0.8715]],\n",
-      "\n",
-      "        [[ 0.7055,  0.7854,  0.9698]],\n",
-      "\n",
-      "        [[ 0.7722,  0.8390,  1.0474]],\n",
-      "\n",
-      "        [[ 0.8323,  0.9074,  1.0969]],\n",
-      "\n",
-      "        [[ 0.8132,  0.8968,  1.1051]],\n",
-      "\n",
-      "        [[ 0.6892,  0.8067,  1.0172]],\n",
-      "\n",
-      "        [[ 0.5896,  0.7130,  0.9167]],\n",
-      "\n",
-      "        [[ 0.4989,  0.5976,  0.8067]]])\n"
+      "DLinear prediction shape: torch.Size([1700, 1])\n",
+      "DLinear prediction: tensor([[ 0.2997],\n",
+      "        [ 0.4742],\n",
+      "        [ 0.6185],\n",
+      "        ...,\n",
+      "        [ 0.0846],\n",
+      "        [-0.1148],\n",
+      "        [-0.2949]])\n"
      ]
     }
    ],
    "source": [
-    "with torch.no_grad():\n",
-    "    test_batch = next(iter(data_module.test_dataloader()))\n",
-    "    x_test, y_test = test_batch\n",
-    "    y_pred = model2(x_test)\n",
-    "\n",
-    "    print(\"Prediction:\", y_pred[\"prediction\"])"
+    "pkg_dl = DLinear_pkg_v2(\n",
+    "    model_cfg=dict(\n",
+    "        loss=MAE(),\n",
+    "        moving_avg=25,\n",
+    "        individual=False,\n",
+    "        logging_metrics=[MAE(), SMAPE()],\n",
+    "        optimizer=\"adam\",\n",
+    "        optimizer_params={\"lr\": 1e-3},\n",
+    "    ),\n",
+    "    datamodule_cfg=datamodule_cfg,\n",
+    "    trainer_cfg=trainer_cfg,\n",
+    ")\n",
+    "pkg_dl.fit(dataset, save_ckpt=False)\n",
+    "preds_dl = pkg_dl.predict(dataset, mode=\"prediction\", trainer_kwargs={\"accelerator\": \"cpu\", \"enable_progress_bar\": False})\n",
+    "print(\"DLinear prediction shape:\", preds_dl[\"prediction\"].shape)\n",
+    "print(\"DLinear prediction:\", preds_dl[\"prediction\"])"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 19,
-   "id": "a4e6e4b1",
+   "execution_count": null,
+   "id": "9d06d4ce8d057e44",
    "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "torch.Size([32, 1, 3])"
-      ]
-     },
-     "execution_count": 19,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "y_pred[\"prediction\"].shape"
-   ]
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "markdown",
+   "id": "9fe81e3e1d43d1f2",
+   "metadata": {},
+   "source": []
   }
  ],
  "metadata": {
@@ -1269,7 +1096,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.12.3"
+   "version": "3.14.6"
   }
  },
  "nbformat": 4,

From 174da668e373cfcc750cf8f0dc32efbf623906eb Mon Sep 17 00:00:00 2001
From: Echo Xiao <echoxiao666@gmail.com>
Date: Sat, 1 Aug 2026 22:30:46 -0700
Subject: [PATCH 2/4] ...

---
 docs/source/tutorials/tslib_v2_example.ipynb | 27 ++++++++++++++++----
 1 file changed, 22 insertions(+), 5 deletions(-)

diff --git a/docs/source/tutorials/tslib_v2_example.ipynb b/docs/source/tutorials/tslib_v2_example.ipynb
index a87c9ef53..bf6b6fd21 100644
--- a/docs/source/tutorials/tslib_v2_example.ipynb
+++ b/docs/source/tutorials/tslib_v2_example.ipynb
@@ -41,7 +41,9 @@
     "import pandas as pd\n",
     "from sklearn.preprocessing import StandardScaler\n",
     "\n",
-    "from pytorch_forecasting.data.data_module import TslibDataModule  # only for the metadata peek below\n",
+    "from pytorch_forecasting.data.data_module import (\n",
+    "    TslibDataModule,\n",
+    ")  # only for the metadata peek below\n",
     "from pytorch_forecasting.data.encoders import TorchNormalizer\n",
     "from pytorch_forecasting.data.timeseries import TimeSeries\n",
     "from pytorch_forecasting.metrics import MAE, SMAPE, QuantileLoss\n",
@@ -651,7 +653,11 @@
     }
    ],
    "source": [
-    "preds_mae = pkg_mae.predict(dataset, mode=\"prediction\", trainer_kwargs={\"accelerator\": \"cpu\", \"enable_progress_bar\": False})\n",
+    "preds_mae = pkg_mae.predict(\n",
+    "    dataset,\n",
+    "    mode=\"prediction\",\n",
+    "    trainer_kwargs={\"accelerator\": \"cpu\", \"enable_progress_bar\": False},\n",
+    ")\n",
     "print(\"Point prediction shape:\", preds_mae[\"prediction\"].shape)\n",
     "print(\"Point prediction:\", preds_mae[\"prediction\"])"
    ]
@@ -880,8 +886,15 @@
     }
    ],
    "source": [
-    "preds_q = pkg_q.predict(dataset, mode=\"quantiles\", trainer_kwargs={\"accelerator\": \"cpu\", \"enable_progress_bar\": False})\n",
-    "print(\"Quantile prediction shape (batch, pred_len, n_quantiles):\", preds_q[\"prediction\"].shape)\n",
+    "preds_q = pkg_q.predict(\n",
+    "    dataset,\n",
+    "    mode=\"quantiles\",\n",
+    "    trainer_kwargs={\"accelerator\": \"cpu\", \"enable_progress_bar\": False},\n",
+    ")\n",
+    "print(\n",
+    "    \"Quantile prediction shape (batch, pred_len, n_quantiles):\",\n",
+    "    preds_q[\"prediction\"].shape,\n",
+    ")\n",
     "print(\"Quantile prediction:\", preds_q[\"prediction\"])"
    ]
   },
@@ -1060,7 +1073,11 @@
     "    trainer_cfg=trainer_cfg,\n",
     ")\n",
     "pkg_dl.fit(dataset, save_ckpt=False)\n",
-    "preds_dl = pkg_dl.predict(dataset, mode=\"prediction\", trainer_kwargs={\"accelerator\": \"cpu\", \"enable_progress_bar\": False})\n",
+    "preds_dl = pkg_dl.predict(\n",
+    "    dataset,\n",
+    "    mode=\"prediction\",\n",
+    "    trainer_kwargs={\"accelerator\": \"cpu\", \"enable_progress_bar\": False},\n",
+    ")\n",
     "print(\"DLinear prediction shape:\", preds_dl[\"prediction\"].shape)\n",
     "print(\"DLinear prediction:\", preds_dl[\"prediction\"])"
    ]

From c2f98daffc63fe21b0fca9b9f22ad0bf07f3ca4a Mon Sep 17 00:00:00 2001
From: Echo Xiao <echoxiao666@gmail.com>
Date: Sun, 2 Aug 2026 11:01:01 -0700
Subject: [PATCH 3/4] ...


From 34f4e53beb888ea129cef212971f76c83340f9fc Mon Sep 17 00:00:00 2001
From: Echo Xiao <echoxiao666@gmail.com>
Date: Mon, 3 Aug 2026 10:58:19 -0700
Subject: [PATCH 4/4] [DOC] address review: keep low-level pipeline, use
 load_toydata, add thuml/enc-dec note and highlight unstable-API warning

---
 docs/source/tutorials/tslib_v2_example.ipynb | 1898 +++++++++++++-----
 1 file changed, 1390 insertions(+), 508 deletions(-)

diff --git a/docs/source/tutorials/tslib_v2_example.ipynb b/docs/source/tutorials/tslib_v2_example.ipynb
index bf6b6fd21..0f3c8394a 100644
--- a/docs/source/tutorials/tslib_v2_example.ipynb
+++ b/docs/source/tutorials/tslib_v2_example.ipynb
@@ -2,7 +2,7 @@
  "cells": [
   {
    "cell_type": "markdown",
-   "id": "662cccd2",
+   "id": "66a17c35",
    "metadata": {},
    "source": [
     "# TSLib for v2 - Example notebook for full pipeline"
@@ -10,69 +10,48 @@
   },
   {
    "cell_type": "markdown",
-   "id": "5b43e1d5",
+   "id": "2e986d11",
    "metadata": {},
    "source": [
-    "## Basic imports for getting started\n",
+    "<div class=\"alert alert-block alert-warning\">\n",
     "\n",
-    "This notebook is an end-to-end vignette for the `tslib` models in v2 of PyTorch Forecasting, using the high-level **package (`pkg`) classes**. You give the package class three config dicts — `model_cfg`, `datamodule_cfg`, `trainer_cfg` — and call `.fit()` / `.predict()`; it wires up the datamodule, model and `Trainer` for you.\n",
+    ":warning: <b>Experimental, unstable API.</b> The <code>tslib</code> models and their data layer are part of an experimental rework of <code>pytorch-forecasting</code>, planned for the v2.0.0 release. The API is unstable and may change without prior notice, and is <b>not recommended for production use</b>. Feedback is very welcome on <a href=\"https://github.com/sktime/pytorch-forecasting/pull/1836\">issue&nbsp;#1836</a>.\n",
     "\n",
-    "This is an experimental, unstable API. Feedback welcome on issue [#1836](https://github.com/sktime/pytorch-forecasting/pull/1836)."
+    "</div>"
    ]
   },
   {
-   "cell_type": "code",
-   "execution_count": 1,
-   "id": "18c542de",
-   "metadata": {
-    "execution": {
-     "iopub.execute_input": "2026-08-01T23:05:31.044514Z",
-     "iopub.status.busy": "2026-08-01T23:05:31.044420Z",
-     "iopub.status.idle": "2026-08-01T23:05:34.235414Z",
-     "shell.execute_reply": "2026-08-01T23:05:34.234976Z"
-    },
-    "jupyter": {
-     "is_executing": true
-    }
-   },
-   "outputs": [],
+   "cell_type": "markdown",
+   "id": "405d0acc",
+   "metadata": {},
    "source": [
-    "import numpy as np\n",
-    "import pandas as pd\n",
-    "from sklearn.preprocessing import StandardScaler\n",
+    "## About the `tslib` models\n",
     "\n",
-    "from pytorch_forecasting.data.data_module import (\n",
-    "    TslibDataModule,\n",
-    ")  # only for the metadata peek below\n",
-    "from pytorch_forecasting.data.encoders import TorchNormalizer\n",
-    "from pytorch_forecasting.data.timeseries import TimeSeries\n",
-    "from pytorch_forecasting.metrics import MAE, SMAPE, QuantileLoss\n",
-    "from pytorch_forecasting.models.timexer import TimeXer_pkg_v2\n",
-    "from pytorch_forecasting.models.dlinear import DLinear_pkg_v2"
+    "The `tslib` models in v2 (e.g. `TimeXer`, `DLinear`) are adapted from the [Time-Series-Library (`thuml`)](https://github.com/thuml/Time-Series-Library). They are **different from the encoder–decoder models** in v2: `tslib` models are built on `TslibBaseModel` and consume the batch layout produced by `TslibDataModule`, whereas the encoder–decoder models (e.g. `TFT`) are built for `EncoderDecoderTimeSeriesDataModule`. The two families are **not interchangeable** — a `tslib` model expects the context/target tensors emitted by `TslibDataModule`, not those from the encoder–decoder datamodule.\n",
+    "\n",
+    "You can train a `tslib` model in **two ways**, and this notebook shows both:\n",
+    "\n",
+    "1. **High-level package (`pkg`) API** — hand a package class (e.g. `TimeXer_pkg_v2`) three config dicts (`model_cfg`, `datamodule_cfg`, `trainer_cfg`) and call `.fit()` / `.predict()`. It builds the datamodule, model and Lightning `Trainer` for you.\n",
+    "2. **Low-level 3-stage pipeline** — build the `TslibDataModule`, model and `Trainer` yourself. More verbose, but gives full control over the trainer, callbacks and preprocessing.\n",
+    "\n",
+    "Each section below is **self-contained**: it creates its own `TimeSeries` dataset and imports what it needs, so you can follow either one on its own. We cover the high-level API first."
    ]
   },
   {
    "cell_type": "markdown",
-   "id": "07365bf1",
+   "id": "25a96c0b",
    "metadata": {},
    "source": [
-    "## Construct a time series dataset\n",
+    "## Create the synthetic dataset\n",
     "\n",
-    "This step requires us to build a `TimeSeries` object for creating a time series dataset, which identifies the features from a raw time series dataset. As you can see below, we are initialising a sample time series dataset."
+    "Both sections share the same raw dataframe. We generate it with the built-in `load_toydata` helper, which returns a `pandas` DataFrame of `num_series` noisy sine-wave series with numeric, categorical, known-future and static columns."
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 2,
-   "id": "1c932107",
-   "metadata": {
-    "execution": {
-     "iopub.execute_input": "2026-08-01T23:05:34.237862Z",
-     "iopub.status.busy": "2026-08-01T23:05:34.237703Z",
-     "iopub.status.idle": "2026-08-01T23:05:34.264007Z",
-     "shell.execute_reply": "2026-08-01T23:05:34.263659Z"
-    }
-   },
+   "execution_count": 1,
+   "id": "a97c1750",
+   "metadata": {},
    "outputs": [
     {
      "data": {
@@ -110,55 +89,55 @@
        "      <th>0</th>\n",
        "      <td>0</td>\n",
        "      <td>0</td>\n",
-       "      <td>-0.010135</td>\n",
-       "      <td>0.211194</td>\n",
+       "      <td>-0.100873</td>\n",
+       "      <td>0.361583</td>\n",
        "      <td>0</td>\n",
        "      <td>1.000000</td>\n",
-       "      <td>0.065485</td>\n",
+       "      <td>0.029542</td>\n",
        "      <td>0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>1</th>\n",
        "      <td>0</td>\n",
        "      <td>1</td>\n",
-       "      <td>0.211194</td>\n",
-       "      <td>0.570354</td>\n",
+       "      <td>0.361583</td>\n",
+       "      <td>0.429515</td>\n",
        "      <td>0</td>\n",
        "      <td>0.995004</td>\n",
-       "      <td>0.065485</td>\n",
+       "      <td>0.029542</td>\n",
        "      <td>0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>2</th>\n",
        "      <td>0</td>\n",
        "      <td>2</td>\n",
-       "      <td>0.570354</td>\n",
-       "      <td>0.433613</td>\n",
+       "      <td>0.429515</td>\n",
+       "      <td>0.672608</td>\n",
        "      <td>0</td>\n",
        "      <td>0.980067</td>\n",
-       "      <td>0.065485</td>\n",
+       "      <td>0.029542</td>\n",
        "      <td>0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>3</th>\n",
        "      <td>0</td>\n",
        "      <td>3</td>\n",
-       "      <td>0.433613</td>\n",
-       "      <td>0.615071</td>\n",
+       "      <td>0.672608</td>\n",
+       "      <td>0.898369</td>\n",
        "      <td>0</td>\n",
        "      <td>0.955336</td>\n",
-       "      <td>0.065485</td>\n",
+       "      <td>0.029542</td>\n",
        "      <td>0</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>4</th>\n",
        "      <td>0</td>\n",
        "      <td>4</td>\n",
-       "      <td>0.615071</td>\n",
-       "      <td>0.858715</td>\n",
+       "      <td>0.898369</td>\n",
+       "      <td>1.002088</td>\n",
        "      <td>0</td>\n",
        "      <td>0.921061</td>\n",
-       "      <td>0.065485</td>\n",
+       "      <td>0.029542</td>\n",
        "      <td>0</td>\n",
        "    </tr>\n",
        "  </tbody>\n",
@@ -167,54 +146,37 @@
       ],
       "text/plain": [
        "   series_id  time_idx         x         y  category  future_known_feature  \\\n",
-       "0          0         0 -0.010135  0.211194         0              1.000000   \n",
-       "1          0         1  0.211194  0.570354         0              0.995004   \n",
-       "2          0         2  0.570354  0.433613         0              0.980067   \n",
-       "3          0         3  0.433613  0.615071         0              0.955336   \n",
-       "4          0         4  0.615071  0.858715         0              0.921061   \n",
+       "0          0         0 -0.100873  0.361583         0              1.000000   \n",
+       "1          0         1  0.361583  0.429515         0              0.995004   \n",
+       "2          0         2  0.429515  0.672608         0              0.980067   \n",
+       "3          0         3  0.672608  0.898369         0              0.955336   \n",
+       "4          0         4  0.898369  1.002088         0              0.921061   \n",
        "\n",
        "   static_feature  static_feature_cat  \n",
-       "0        0.065485                   0  \n",
-       "1        0.065485                   0  \n",
-       "2        0.065485                   0  \n",
-       "3        0.065485                   0  \n",
-       "4        0.065485                   0  "
+       "0        0.029542                   0  \n",
+       "1        0.029542                   0  \n",
+       "2        0.029542                   0  \n",
+       "3        0.029542                   0  \n",
+       "4        0.029542                   0  "
       ]
      },
-     "execution_count": 2,
+     "execution_count": 1,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
-    "num_series = 100\n",
-    "seq_length = 50\n",
-    "data_list = []\n",
-    "for i in range(num_series):\n",
-    "    x = np.arange(seq_length)\n",
-    "    y = np.sin(x / 5.0) + np.random.normal(scale=0.1, size=seq_length)\n",
-    "    category = i % 5\n",
-    "    static_value = np.random.rand()\n",
-    "    for t in range(seq_length - 1):\n",
-    "        data_list.append(\n",
-    "            {\n",
-    "                \"series_id\": i,\n",
-    "                \"time_idx\": t,\n",
-    "                \"x\": y[t],\n",
-    "                \"y\": y[t + 1],\n",
-    "                \"category\": category,\n",
-    "                \"future_known_feature\": np.cos(t / 10),\n",
-    "                \"static_feature\": static_value,\n",
-    "                \"static_feature_cat\": i % 3,\n",
-    "            }\n",
-    "        )\n",
-    "data_df = pd.DataFrame(data_list)\n",
+    "from pytorch_forecasting.data.examples import load_toydata\n",
+    "\n",
+    "num_series = 100  # number of individual time series\n",
+    "seq_length = 50  # length of each series\n",
+    "data_df = load_toydata(num_series, seq_length)\n",
     "data_df.head()"
    ]
   },
   {
    "cell_type": "markdown",
-   "id": "db37eeee",
+   "id": "bd371520",
    "metadata": {},
    "source": [
     "## Feature Categories and Definitions\n",
@@ -260,18 +222,69 @@
     "- **Usage**: Entity-specific characteristics that don't change over time"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "id": "ed836f71",
+   "metadata": {},
+   "source": [
+    "> **Note on categorical features.** The `cat` columns above are declared and carried through the `TslibDataModule` (as `history_cat` / `future_cat` / `static_categorical_features`), but **TimeXer v2 currently uses only the continuous features in the context window** — categorical variables are not yet consumed by the model (`enc_in` is set to the continuous dimension only).\n",
+    "\n",
+    "Also note the datamodule does **not** encode categoricals: every feature is cast to `float32` and sliced by index, so categorical columns must already be numeric (string categories would fail). There is no categorical-encoder step in this pipeline yet. Support for proper categorical handling is planned for a future release. We keep the `cat` declaration here to show the full data definition, not because it affects the forecast today."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "b0fff5e5",
+   "metadata": {},
+   "source": [
+    "## 1. High-level package (`pkg`) API"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "b7f87b84",
+   "metadata": {},
+   "source": [
+    "### Steps\n",
+    "* Create the `TimeSeries` dataset\n",
+    "* Create the config dicts (`datamodule_cfg`, `trainer_cfg`, and a per-model `model_cfg`)\n",
+    "* Create the package object (e.g. `TimeXer_pkg_v2`)\n",
+    "* Call `pkg.fit(dataset)` and `pkg.predict(dataset)`\n",
+    "\n",
+    "The package class is a thin wrapper that manages the model, its `TslibDataModule` and the Lightning `Trainer`. Checkpoints are saved automatically under `checkpoints/` unless you pass `save_ckpt=False`."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "1debca4a",
+   "metadata": {},
+   "source": [
+    "### Create the `TimeSeries` dataset\n",
+    "\n",
+    "`TimeSeries` turns the raw dataframe into tensors and records which columns are the target, groups, numeric/categorical, known/unknown and static features (see the reference above)."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "id": "4f0f37de",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from sklearn.preprocessing import StandardScaler\n",
+    "\n",
+    "from pytorch_forecasting.data.encoders import TorchNormalizer\n",
+    "from pytorch_forecasting.data.timeseries import TimeSeries\n",
+    "from pytorch_forecasting.metrics import MAE, SMAPE, QuantileLoss\n",
+    "from pytorch_forecasting.models.timexer import TimeXer_pkg_v2\n",
+    "from pytorch_forecasting.models.dlinear import DLinear_pkg_v2"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 3,
-   "id": "7b9016cb",
-   "metadata": {
-    "execution": {
-     "iopub.execute_input": "2026-08-01T23:05:34.265375Z",
-     "iopub.status.busy": "2026-08-01T23:05:34.265299Z",
-     "iopub.status.idle": "2026-08-01T23:05:34.271403Z",
-     "shell.execute_reply": "2026-08-01T23:05:34.271057Z"
-    }
-   },
+   "id": "7fb824fc",
+   "metadata": {},
    "outputs": [
     {
      "name": "stderr",
@@ -298,36 +311,19 @@
   },
   {
    "cell_type": "markdown",
-   "id": "99d80bd2",
-   "metadata": {},
-   "source": [
-    "> **Note on categorical features.** The `cat` columns above are declared and carried through the `TslibDataModule` (as `history_cat` / `future_cat` / `static_categorical_features`), but **TimeXer v2 currently uses only the continuous features in the context window** — categorical variables are not yet consumed by the model (`enc_in` is set to the continuous dimension only).\n",
-    "\n",
-    "Also note the datamodule does **not** encode categoricals: every feature is cast to `float32` and sliced by index, so categorical columns must already be numeric (string categories would fail). There is no categorical-encoder step in this pipeline yet. Support for proper categorical handling is planned for a future release. We keep the `cat` declaration here to show the full data definition, not because it affects the forecast today."
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "5ab5e1e9",
+   "id": "1692afdf",
    "metadata": {},
    "source": [
-    "## The high-level package (`pkg`) API\n",
+    "### Create the configs\n",
     "\n",
-    "Every v2 model has a package class (e.g. `TimeXer_pkg_v2`, `DLinear_pkg_v2`). It is a thin wrapper that manages the model, its `TslibDataModule` and the Lightning `Trainer`. You configure it with three dicts and drive it with `.fit(data)` and `.predict(data)` — no manual object wiring, and checkpoints are saved automatically under `checkpoints/`."
+    "`datamodule_cfg` and `trainer_cfg` are shared across the models below; each model gets its own `model_cfg` inline in the package call."
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 4,
-   "id": "36fe8aba",
-   "metadata": {
-    "execution": {
-     "iopub.execute_input": "2026-08-01T23:05:34.273984Z",
-     "iopub.status.busy": "2026-08-01T23:05:34.273895Z",
-     "iopub.status.idle": "2026-08-01T23:05:34.275739Z",
-     "shell.execute_reply": "2026-08-01T23:05:34.275436Z"
-    }
-   },
+   "id": "d585fd06",
+   "metadata": {},
    "outputs": [],
    "source": [
     "datamodule_cfg = dict(\n",
@@ -348,111 +344,25 @@
     "    accelerator=\"cpu\",\n",
     "    devices=1,\n",
     "    enable_progress_bar=False,\n",
-    "    enable_model_summary=True,\n",
+    "    enable_model_summary=False,\n",
     ")"
    ]
   },
   {
    "cell_type": "markdown",
-   "id": "a6081661",
+   "id": "6cdfe14d",
    "metadata": {},
    "source": [
-    "## Peek at the metadata\n",
+    "### TimeXer with MAE (point forecast) via the package class\n",
     "\n",
-    "The `pkg` classes build the `TslibDataModule` for you internally. If you want to inspect the metadata that gets passed to the model, you can build one explicitly:"
+    "Instead of wiring up the model, datamodule and `Trainer` by hand, we hand three config dicts to `TimeXer_pkg_v2` and call `.fit()` / `.predict()`."
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 5,
-   "id": "8da49230",
-   "metadata": {
-    "execution": {
-     "iopub.execute_input": "2026-08-01T23:05:34.276848Z",
-     "iopub.status.busy": "2026-08-01T23:05:34.276784Z",
-     "iopub.status.idle": "2026-08-01T23:05:34.345367Z",
-     "shell.execute_reply": "2026-08-01T23:05:34.344958Z"
-    }
-   },
-   "outputs": [
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/data/data_module/_tslib_data_module.py:331: UserWarning: TslibDataModule is experimental and subject to change. The API is not stable and may change without prior warning.\n",
-      "  warnings.warn(\n"
-     ]
-    },
-    {
-     "data": {
-      "text/plain": [
-       "{'feature_names': {'categorical': ['category', 'static_feature_cat'],\n",
-       "  'continuous': ['x', 'future_known_feature', 'static_feature'],\n",
-       "  'static': ['static_feature', 'static_feature_cat'],\n",
-       "  'known': ['future_known_feature'],\n",
-       "  'unknown': ['x', 'category', 'static_feature', 'static_feature_cat'],\n",
-       "  'target': ['y'],\n",
-       "  'all': ['x',\n",
-       "   'category',\n",
-       "   'future_known_feature',\n",
-       "   'static_feature',\n",
-       "   'static_feature_cat'],\n",
-       "  'static_categorical': ['static_feature_cat'],\n",
-       "  'static_continuous': ['static_feature']},\n",
-       " 'feature_indices': {'categorical': [1, 4],\n",
-       "  'continuous': [0, 2, 3],\n",
-       "  'static': [],\n",
-       "  'known': [2],\n",
-       "  'unknown': [0, 1, 3, 4],\n",
-       "  'target': [0]},\n",
-       " 'n_features': {'categorical': 2,\n",
-       "  'continuous': 3,\n",
-       "  'static': 2,\n",
-       "  'known': 1,\n",
-       "  'unknown': 4,\n",
-       "  'target': 1,\n",
-       "  'all': 5,\n",
-       "  'static_categorical': 1,\n",
-       "  'static_continuous': 1},\n",
-       " 'context_length': 32,\n",
-       " 'prediction_length': 1,\n",
-       " 'freq': 'h',\n",
-       " 'features': 'MS'}"
-      ]
-     },
-     "execution_count": 5,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "_dm_preview = TslibDataModule(time_series_dataset=dataset, **datamodule_cfg)\n",
-    "_dm_preview.setup(stage=\"fit\")\n",
-    "_dm_preview.metadata"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "b19ce3e3",
+   "id": "48c3a097",
    "metadata": {},
-   "source": [
-    "## TimeXer with MAE (point forecast) via the package class\n",
-    "\n",
-    "Instead of wiring up the model, datamodule and `Trainer` by hand, we hand three config dicts to `TimeXer_pkg_v2` and call `.fit()` / `.predict()`."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 6,
-   "id": "92a9d2a6",
-   "metadata": {
-    "execution": {
-     "iopub.execute_input": "2026-08-01T23:05:34.346619Z",
-     "iopub.status.busy": "2026-08-01T23:05:34.346541Z",
-     "iopub.status.idle": "2026-08-01T23:05:34.349410Z",
-     "shell.execute_reply": "2026-08-01T23:05:34.348947Z"
-    }
-   },
    "outputs": [
     {
      "name": "stdout",
@@ -485,16 +395,9 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 7,
-   "id": "a051e0bf",
-   "metadata": {
-    "execution": {
-     "iopub.execute_input": "2026-08-01T23:05:34.350621Z",
-     "iopub.status.busy": "2026-08-01T23:05:34.350544Z",
-     "iopub.status.idle": "2026-08-01T23:05:44.127364Z",
-     "shell.execute_reply": "2026-08-01T23:05:44.126961Z"
-    }
-   },
+   "execution_count": 6,
+   "id": "94e0f688",
+   "metadata": {},
    "outputs": [
     {
      "name": "stderr",
@@ -522,55 +425,18 @@
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/setup.py:175: GPU available but not used. You can set it by doing `Trainer(accelerator='gpu')`.\n",
-      "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n"
+      "/Users/echoooooo/miniconda3/lib/python3.14/site-packages/lightning/pytorch/trainer/setup.py:175: GPU available but not used. You can set it by doing `Trainer(accelerator='gpu')`.\n",
+      ""
      ]
     },
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "\n",
-      "  | Name            | Type                   | Params | Mode  | FLOPs\n",
-      "---------------------------------------------------------------------------\n",
-      "0 | loss            | MAE                    | 0      | train | 0    \n",
-      "1 | logging_metrics | ModuleList             | 0      | train | 0    \n",
-      "2 | en_embedding    | EnEmbedding            | 320    | train | 0    \n",
-      "3 | ex_embedding    | DataEmbedding_inverted | 2.1 K  | train | 0    \n",
-      "4 | encoder         | Encoder                | 133 K  | train | 0    \n",
-      "5 | head            | FlattenHead            | 577    | train | 0    \n",
-      "---------------------------------------------------------------------------\n",
-      "136 K     Trainable params\n",
-      "0         Non-trainable params\n",
-      "136 K     Total params\n",
-      "0.547     Total estimated model params size (MB)\n",
-      "60        Modules in train mode\n",
-      "0         Modules in eval mode\n",
-      "0         Total Flops\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'val_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'train_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n",
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/loops/fit_loop.py:321: The number of training batches (38) is smaller than the logging interval Trainer(log_every_n_steps=50). Set a lower value for log_every_n_steps if you want to see logs for the training epoch.\n"
+      "/Users/echoooooo/miniconda3/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
+      "/Users/echoooooo/miniconda3/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'val_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n",
+      "/Users/echoooooo/miniconda3/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'train_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n",
+      "/Users/echoooooo/miniconda3/lib/python3.14/site-packages/lightning/pytorch/loops/fit_loop.py:321: The number of training batches (38) is smaller than the logging interval Trainer(log_every_n_steps=50). Set a lower value for log_every_n_steps if you want to see logs for the training epoch.\n"
      ]
     },
     {
@@ -587,16 +453,9 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 8,
-   "id": "cb5aa00f",
-   "metadata": {
-    "execution": {
-     "iopub.execute_input": "2026-08-01T23:05:44.128653Z",
-     "iopub.status.busy": "2026-08-01T23:05:44.128567Z",
-     "iopub.status.idle": "2026-08-01T23:05:45.858557Z",
-     "shell.execute_reply": "2026-08-01T23:05:45.858157Z"
-    }
-   },
+   "execution_count": 7,
+   "id": "3ace66cc",
+   "metadata": {},
    "outputs": [
     {
      "name": "stderr",
@@ -618,23 +477,15 @@
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/setup.py:175: GPU available but not used. You can set it by doing `Trainer(accelerator='gpu')`.\n",
-      "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n"
+      "/Users/echoooooo/miniconda3/lib/python3.14/site-packages/lightning/pytorch/trainer/setup.py:175: GPU available but not used. You can set it by doing `Trainer(accelerator='gpu')`.\n",
+      ""
      ]
     },
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'predict_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n"
+      "/Users/echoooooo/miniconda3/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'predict_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n"
      ]
     },
     {
@@ -642,13 +493,13 @@
      "output_type": "stream",
      "text": [
       "Point prediction shape: torch.Size([1700, 1])\n",
-      "Point prediction: tensor([[ 0.2108],\n",
-      "        [ 0.3451],\n",
-      "        [ 0.4759],\n",
+      "Point prediction: tensor([[ 0.2879],\n",
+      "        [ 0.4023],\n",
+      "        [ 0.5390],\n",
       "        ...,\n",
-      "        [-0.0950],\n",
-      "        [-0.3101],\n",
-      "        [-0.4429]])\n"
+      "        [-0.0669],\n",
+      "        [-0.2310],\n",
+      "        [-0.4071]])\n"
      ]
     }
    ],
@@ -664,26 +515,19 @@
   },
   {
    "cell_type": "markdown",
-   "id": "4bc8ddd8",
+   "id": "099c5cb2",
    "metadata": {},
    "source": [
-    "## TimeXer with QuantileLoss (interval forecast)\n",
+    "### TimeXer with QuantileLoss (interval forecast)\n",
     "\n",
     "Swap the loss to `QuantileLoss` to get quantile / prediction-interval outputs."
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 9,
-   "id": "27a5689f",
-   "metadata": {
-    "execution": {
-     "iopub.execute_input": "2026-08-01T23:05:45.859695Z",
-     "iopub.status.busy": "2026-08-01T23:05:45.859624Z",
-     "iopub.status.idle": "2026-08-01T23:05:45.862143Z",
-     "shell.execute_reply": "2026-08-01T23:05:45.861730Z"
-    }
-   },
+   "execution_count": 8,
+   "id": "b355ba52",
+   "metadata": {},
    "outputs": [
     {
      "name": "stdout",
@@ -718,23 +562,14 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 10,
-   "id": "c89820fc",
-   "metadata": {
-    "execution": {
-     "iopub.execute_input": "2026-08-01T23:05:45.863145Z",
-     "iopub.status.busy": "2026-08-01T23:05:45.863067Z",
-     "iopub.status.idle": "2026-08-01T23:05:55.640465Z",
-     "shell.execute_reply": "2026-08-01T23:05:55.639906Z"
-    }
-   },
+   "execution_count": 9,
+   "id": "e20c7e0a",
+   "metadata": {},
    "outputs": [
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/data/data_module/_tslib_data_module.py:331: UserWarning: TslibDataModule is experimental and subject to change. The API is not stable and may change without prior warning.\n",
-      "  warnings.warn(\n",
       "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/models/base/_base_model_v2.py:85: UserWarning: The Model 'TimeXer' is part of an experimental reworkof the pytorch-forecasting model layer, scheduled for release with v2.0.0. The API is not stable and may change without prior warning. This class is intended for beta testing and as a basic skeleton, but not for stable production use. Feedback and suggestions are very welcome in pytorch-forecasting issue 1736, https://github.com/sktime/pytorch-forecasting/issues/1736\n",
       "  warn(\n",
       "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/models/base/_tslib_base_model_v2.py:63: UserWarning: The Model 'TimeXer' is part of an experimental implementationof the pytorch-forecasting model layer for Time Series Library, scheduledfor release with v2.0.0. The API is not stableand may change without prior warning. This class is intended for betatesting, not for stable production use.\n",
@@ -751,55 +586,6 @@
       "TPU available: False, using: 0 TPU cores\n"
      ]
     },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/setup.py:175: GPU available but not used. You can set it by doing `Trainer(accelerator='gpu')`.\n",
-      "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "\n",
-      "  | Name            | Type                   | Params | Mode  | FLOPs\n",
-      "---------------------------------------------------------------------------\n",
-      "0 | loss            | QuantileLoss           | 0      | train | 0    \n",
-      "1 | logging_metrics | ModuleList             | 0      | train | 0    \n",
-      "2 | en_embedding    | EnEmbedding            | 320    | train | 0    \n",
-      "3 | ex_embedding    | DataEmbedding_inverted | 2.1 K  | train | 0    \n",
-      "4 | encoder         | Encoder                | 133 K  | train | 0    \n",
-      "5 | head            | FlattenHead            | 1.7 K  | train | 0    \n",
-      "---------------------------------------------------------------------------\n",
-      "137 K     Trainable params\n",
-      "0         Non-trainable params\n",
-      "137 K     Total params\n",
-      "0.551     Total estimated model params size (MB)\n",
-      "58        Modules in train mode\n",
-      "0         Modules in eval mode\n",
-      "0         Total Flops\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'val_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n",
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'train_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n",
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/loops/fit_loop.py:321: The number of training batches (38) is smaller than the logging interval Trainer(log_every_n_steps=50). Set a lower value for log_every_n_steps if you want to see logs for the training epoch.\n"
-     ]
-    },
     {
      "name": "stderr",
      "output_type": "stream",
@@ -814,23 +600,14 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 11,
-   "id": "8240e1fb",
-   "metadata": {
-    "execution": {
-     "iopub.execute_input": "2026-08-01T23:05:55.641813Z",
-     "iopub.status.busy": "2026-08-01T23:05:55.641734Z",
-     "iopub.status.idle": "2026-08-01T23:05:57.364930Z",
-     "shell.execute_reply": "2026-08-01T23:05:57.364434Z"
-    }
-   },
+   "execution_count": 10,
+   "id": "53c8c3b8",
+   "metadata": {},
    "outputs": [
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/data/data_module/_tslib_data_module.py:331: UserWarning: TslibDataModule is experimental and subject to change. The API is not stable and may change without prior warning.\n",
-      "  warnings.warn(\n",
       "GPU available: True (mps), used: False\n"
      ]
     },
@@ -842,46 +619,23 @@
      ]
     },
     {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/setup.py:175: GPU available but not used. You can set it by doing `Trainer(accelerator='gpu')`.\n",
-      "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'predict_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n"
-     ]
-    },
-    {
-     "name": "stdout",
+     "name": "stdout",
      "output_type": "stream",
      "text": [
       "Quantile prediction shape (batch, pred_len, n_quantiles): torch.Size([1700, 1, 3])\n",
-      "Quantile prediction: tensor([[[ 0.1863,  0.2792,  0.4827]],\n",
+      "Quantile prediction: tensor([[[ 0.1936,  0.4580,  0.5065]],\n",
       "\n",
-      "        [[ 0.3067,  0.4198,  0.6217]],\n",
+      "        [[ 0.3599,  0.6098,  0.6686]],\n",
       "\n",
-      "        [[ 0.4261,  0.5496,  0.7150]],\n",
+      "        [[ 0.4852,  0.7296,  0.7853]],\n",
       "\n",
       "        ...,\n",
       "\n",
-      "        [[-0.1460,  0.1544,  0.1308]],\n",
+      "        [[-0.2017,  0.1089,  0.2374]],\n",
       "\n",
-      "        [[-0.3319, -0.0281, -0.0082]],\n",
+      "        [[-0.3642, -0.0622,  0.0831]],\n",
       "\n",
-      "        [[-0.5041, -0.2163, -0.1586]]])\n"
+      "        [[-0.5479, -0.2107, -0.0850]]])\n"
      ]
     }
    ],
@@ -900,33 +654,24 @@
   },
   {
    "cell_type": "markdown",
-   "id": "cb9492cf",
+   "id": "831949eb",
    "metadata": {},
    "source": [
-    "## Same pipeline, different model: DLinear\n",
+    "### Same pipeline, different model: DLinear\n",
     "\n",
     "`DLinear` is the other model built on the `tslib` base, so it runs through the exact same pipeline — only the package class and its `model_cfg` change."
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 12,
-   "id": "98090c16",
-   "metadata": {
-    "execution": {
-     "iopub.execute_input": "2026-08-01T23:05:57.366294Z",
-     "iopub.status.busy": "2026-08-01T23:05:57.366198Z",
-     "iopub.status.idle": "2026-08-01T23:06:03.812697Z",
-     "shell.execute_reply": "2026-08-01T23:06:03.812248Z"
-    }
-   },
+   "execution_count": 11,
+   "id": "e3f0201a",
+   "metadata": {},
    "outputs": [
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/data/data_module/_tslib_data_module.py:331: UserWarning: TslibDataModule is experimental and subject to change. The API is not stable and may change without prior warning.\n",
-      "  warnings.warn(\n",
       "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/models/base/_base_model_v2.py:85: UserWarning: The Model 'DLinear' is part of an experimental reworkof the pytorch-forecasting model layer, scheduled for release with v2.0.0. The API is not stable and may change without prior warning. This class is intended for beta testing and as a basic skeleton, but not for stable production use. Feedback and suggestions are very welcome in pytorch-forecasting issue 1736, https://github.com/sktime/pytorch-forecasting/issues/1736\n",
       "  warn(\n",
       "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/models/base/_tslib_base_model_v2.py:63: UserWarning: The Model 'DLinear' is part of an experimental implementationof the pytorch-forecasting model layer for Time Series Library, scheduledfor release with v2.0.0. The API is not stableand may change without prior warning. This class is intended for betatesting, not for stable production use.\n",
@@ -943,43 +688,6 @@
       "TPU available: False, using: 0 TPU cores\n"
      ]
     },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/setup.py:175: GPU available but not used. You can set it by doing `Trainer(accelerator='gpu')`.\n",
-      "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "\n",
-      "  | Name            | Type                | Params | Mode  | FLOPs\n",
-      "------------------------------------------------------------------------\n",
-      "0 | loss            | MAE                 | 0      | train | 0    \n",
-      "1 | logging_metrics | ModuleList          | 0      | train | 0    \n",
-      "2 | decomposition   | SeriesDecomposition | 0      | train | 0    \n",
-      "3 | linear_seasonal | Linear              | 33     | train | 0    \n",
-      "4 | linear_trend    | Linear              | 33     | train | 0    \n",
-      "------------------------------------------------------------------------\n",
-      "66        Trainable params\n",
-      "0         Non-trainable params\n",
-      "66        Total params\n",
-      "0.000     Total estimated model params size (MB)\n",
-      "9         Modules in train mode\n",
-      "0         Modules in eval mode\n",
-      "0         Total Flops\n"
-     ]
-    },
     {
      "name": "stdout",
      "output_type": "stream",
@@ -987,17 +695,6 @@
       "{'loss': MAE(), 'moving_avg': 25, 'individual': False, 'logging_metrics': [MAE(), SMAPE()], 'optimizer': 'adam', 'optimizer_params': {'lr': 0.001}}\n"
      ]
     },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'val_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n",
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'train_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n",
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/loops/fit_loop.py:321: The number of training batches (38) is smaller than the logging interval Trainer(log_every_n_steps=50). Set a lower value for log_every_n_steps if you want to see logs for the training epoch.\n"
-     ]
-    },
     {
      "name": "stderr",
      "output_type": "stream",
@@ -1009,8 +706,6 @@
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/data/data_module/_tslib_data_module.py:331: UserWarning: TslibDataModule is experimental and subject to change. The API is not stable and may change without prior warning.\n",
-      "  warnings.warn(\n",
       "GPU available: True (mps), used: False\n"
      ]
     },
@@ -1021,41 +716,18 @@
       "TPU available: False, using: 0 TPU cores\n"
      ]
     },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/setup.py:175: GPU available but not used. You can set it by doing `Trainer(accelerator='gpu')`.\n",
-      "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/utilities/_pytree.py:21: `isinstance(treespec, LeafSpec)` is deprecated, use `isinstance(treespec, TreeSpec) and treespec.is_leaf()` instead.\n",
-      "/Users/echoooooo/Desktop/code/pytorch-forecasting/.venv/lib/python3.14/site-packages/lightning/pytorch/trainer/connectors/data_connector.py:434: The 'predict_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=7` in the `DataLoader` to improve performance.\n"
-     ]
-    },
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "DLinear prediction shape: torch.Size([1700, 1])\n",
-      "DLinear prediction: tensor([[ 0.2997],\n",
-      "        [ 0.4742],\n",
-      "        [ 0.6185],\n",
+      "DLinear prediction: tensor([[ 0.3987],\n",
+      "        [ 0.5763],\n",
+      "        [ 0.7305],\n",
       "        ...,\n",
-      "        [ 0.0846],\n",
-      "        [-0.1148],\n",
-      "        [-0.2949]])\n"
+      "        [ 0.0261],\n",
+      "        [-0.1774],\n",
+      "        [-0.3777]])\n"
      ]
     }
    ],
@@ -1082,19 +754,1229 @@
     "print(\"DLinear prediction:\", preds_dl[\"prediction\"])"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "id": "0fbf38fc",
+   "metadata": {},
+   "source": [
+    "## 2. Low-level 3-stage pipeline"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "11d1710e",
+   "metadata": {},
+   "source": [
+    "### Steps\n",
+    "1. Create the `TimeSeries` dataset\n",
+    "2. Create the `TslibDataModule`\n",
+    "3. Initialise, train and run inference with the model\n",
+    "\n",
+    "This section does by hand what the package class did for you above. It is fully self-contained — it rebuilds the `TimeSeries` dataset from `data_df` and re-imports what it needs — so you can follow it without running section 1."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "5bfbdd89",
+   "metadata": {},
+   "source": [
+    "### 1. Create the dataset\n",
+    "\n",
+    "Same `TimeSeries` construction as above; the `data_module` will consume this and handle the dataloaders and preprocessing."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 12,
+   "id": "988ff219",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import torch\n",
+    "from sklearn.preprocessing import StandardScaler\n",
+    "\n",
+    "from pytorch_forecasting.data.data_module import TslibDataModule\n",
+    "from pytorch_forecasting.data.encoders import TorchNormalizer\n",
+    "from pytorch_forecasting.data.timeseries import TimeSeries\n",
+    "from pytorch_forecasting.metrics import MAE, SMAPE, QuantileLoss\n",
+    "from pytorch_forecasting.models.timexer._timexer_v2 import TimeXer"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 13,
+   "id": "32e01d37",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/echoooooo/Desktop/code/pytorch-forecasting/pytorch_forecasting/data/timeseries/_timeseries_v2.py:104: UserWarning: TimeSeries is part of an experimental rework of the pytorch-forecasting data layer, scheduled for release with v2.0.0. The API is not stable and may change without prior warning. For beta testing, but not for stable production use. Feedback and suggestions are very welcome in pytorch-forecasting issue 1736, https://github.com/sktime/pytorch-forecasting/issues/1736\n",
+      "  warn(\n"
+     ]
+    }
+   ],
+   "source": [
+    "dataset = TimeSeries(\n",
+    "    data=data_df,\n",
+    "    time=\"time_idx\",\n",
+    "    target=\"y\",\n",
+    "    group=[\"series_id\"],\n",
+    "    num=[\"x\", \"future_known_feature\", \"static_feature\"],\n",
+    "    cat=[\"category\", \"static_feature_cat\"],\n",
+    "    known=[\"future_known_feature\"],\n",
+    "    unknown=[\"x\", \"category\"],\n",
+    "    static=[\"static_feature\", \"static_feature_cat\"],\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "b641fb61",
+   "metadata": {},
+   "source": [
+    "### 2. Create the datamodule\n",
+    "\n",
+    "We build the `TslibDataModule` explicitly. We do not pass `categorical_encoders`: the `tslib` datamodule casts every feature to `float32` and does not run a categorical-encoder step (see the note on categorical features above). `setup(stage=\"fit\")` prepares the train/val splits and exposes `metadata`, which the model needs for its input/output shapes."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 14,
+   "id": "50915926",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "{'feature_names': {'categorical': ['category', 'static_feature_cat'],\n",
+       "  'continuous': ['x', 'future_known_feature', 'static_feature'],\n",
+       "  'static': ['static_feature', 'static_feature_cat'],\n",
+       "  'known': ['future_known_feature'],\n",
+       "  'unknown': ['x', 'category', 'static_feature', 'static_feature_cat'],\n",
+       "  'target': ['y'],\n",
+       "  'all': ['x',\n",
+       "   'category',\n",
+       "   'future_known_feature',\n",
+       "   'static_feature',\n",
+       "   'static_feature_cat'],\n",
+       "  'static_categorical': ['static_feature_cat'],\n",
+       "  'static_continuous': ['static_feature']},\n",
+       " 'feature_indices': {'categorical': [1, 4],\n",
+       "  'continuous': [0, 2, 3],\n",
+       "  'static': [],\n",
+       "  'known': [2],\n",
+       "  'unknown': [0, 1, 3, 4],\n",
+       "  'target': [0]},\n",
+       " 'n_features': {'categorical': 2,\n",
+       "  'continuous': 3,\n",
+       "  'static': 2,\n",
+       "  'known': 1,\n",
+       "  'unknown': 4,\n",
+       "  'target': 1,\n",
+       "  'all': 5,\n",
+       "  'static_categorical': 1,\n",
+       "  'static_continuous': 1},\n",
+       " 'context_length': 32,\n",
+       " 'prediction_length': 1,\n",
+       " 'freq': 'h',\n",
+       " 'features': 'MS'}"
+      ]
+     },
+     "execution_count": 14,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "data_module = TslibDataModule(\n",
+    "    time_series_dataset=dataset,\n",
+    "    context_length=32,  # divisible by patch_length(4) -> no dropped-timestep warning\n",
+    "    prediction_length=1,\n",
+    "    add_relative_time_idx=True,\n",
+    "    target_normalizer=TorchNormalizer(),\n",
+    "    scalers={\n",
+    "        \"x\": StandardScaler(),\n",
+    "        \"future_known_feature\": StandardScaler(),\n",
+    "        \"static_feature\": StandardScaler(),\n",
+    "    },\n",
+    "    batch_size=32,\n",
+    ")\n",
+    "data_module.setup(stage=\"fit\")\n",
+    "data_module.metadata"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "1c6d5a44",
+   "metadata": {},
+   "source": [
+    "### 3. Initialise and train the model\n",
+    "\n",
+    "We build two `TimeXer` models — one with `MAE()` (point forecast) and one with `QuantileLoss()` (interval forecast) — passing `metadata=data_module.metadata` so each model knows its input/output shapes. The `QuantileLoss` model omits `logging_metrics`: point metrics (MAE/SMAPE) cannot be computed on quantile output."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 15,
+   "id": "e480232d",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "model_mae = TimeXer(\n",
+    "    loss=MAE(),\n",
+    "    hidden_size=64,\n",
+    "    n_heads=4,\n",
+    "    e_layers=2,\n",
+    "    d_ff=256,\n",
+    "    dropout=0.1,\n",
+    "    patch_length=4,\n",
+    "    logging_metrics=[MAE(), SMAPE()],\n",
+    "    optimizer=\"adam\",\n",
+    "    optimizer_params={\"lr\": 1e-3},\n",
+    "    lr_scheduler=\"reduce_lr_on_plateau\",\n",
+    "    lr_scheduler_params={\"mode\": \"min\", \"factor\": 0.5, \"patience\": 5},\n",
+    "    metadata=data_module.metadata,\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "6f3bc535",
+   "metadata": {},
+   "source": [
+    "Inspect the model architecture with a plain-text `ModelSummary` (we disabled the trainer's automatic summary above to keep the training logs clean):"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 16,
+   "id": "25e0de8f",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "  | Name            | Type                   | Params | Mode  | FLOPs\n",
+      "---------------------------------------------------------------------------\n",
+      "0 | loss            | MAE                    | 0      | train | 0    \n",
+      "1 | logging_metrics | ModuleList             | 0      | train | 0    \n",
+      "2 | en_embedding    | EnEmbedding            | 320    | train | 0    \n",
+      "3 | ex_embedding    | DataEmbedding_inverted | 2.1 K  | train | 0    \n",
+      "4 | encoder         | Encoder                | 133 K  | train | 0    \n",
+      "5 | head            | FlattenHead            | 577    | train | 0    \n",
+      "---------------------------------------------------------------------------\n",
+      "136 K     Trainable params\n",
+      "0         Non-trainable params\n",
+      "136 K     Total params\n",
+      "0.547     Total estimated model params size (MB)\n",
+      "60        Modules in train mode\n",
+      "0         Modules in eval mode\n",
+      "0         Total Flops\n"
+     ]
+    }
+   ],
+   "source": [
+    "from lightning.pytorch.utilities.model_summary import ModelSummary\n",
+    "\n",
+    "print(ModelSummary(model_mae, max_depth=1))"
+   ]
+  },
   {
    "cell_type": "code",
-   "execution_count": null,
-   "id": "9d06d4ce8d057e44",
+   "execution_count": 17,
+   "id": "f052068b",
    "metadata": {},
    "outputs": [],
-   "source": []
+   "source": [
+    "model_q = TimeXer(\n",
+    "    loss=QuantileLoss(quantiles=[0.1, 0.5, 0.9]),\n",
+    "    hidden_size=64,\n",
+    "    n_heads=4,\n",
+    "    e_layers=2,\n",
+    "    d_ff=256,\n",
+    "    dropout=0.1,\n",
+    "    patch_length=4,\n",
+    "    # no logging_metrics: point metrics can't be computed on QuantileLoss output\n",
+    "    optimizer=\"adam\",\n",
+    "    optimizer_params={\"lr\": 1e-3},\n",
+    "    lr_scheduler=\"reduce_lr_on_plateau\",\n",
+    "    lr_scheduler_params={\"mode\": \"min\", \"factor\": 0.5, \"patience\": 5},\n",
+    "    metadata=data_module.metadata,\n",
+    ")"
+   ]
   },
   {
    "cell_type": "markdown",
-   "id": "9fe81e3e1d43d1f2",
+   "id": "2a49bcc4",
    "metadata": {},
-   "source": []
+   "source": [
+    "We use a PyTorch Lightning `Trainer` to train each model. With the low-level API you create the `Trainer` yourself and call `trainer.fit(model, data_module)`."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 18,
+   "id": "dac60055",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "GPU available: True (mps), used: False\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "TPU available: False, using: 0 TPU cores\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "GPU available: True (mps), used: False\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "TPU available: False, using: 0 TPU cores\n"
+     ]
+    }
+   ],
+   "source": [
+    "from lightning.pytorch import Trainer\n",
+    "\n",
+    "trainer_mae = Trainer(\n",
+    "    max_epochs=5,\n",
+    "    accelerator=\"cpu\",\n",
+    "    devices=1,\n",
+    "    enable_progress_bar=False,\n",
+    "    enable_model_summary=False,\n",
+    ")\n",
+    "trainer_q = Trainer(\n",
+    "    max_epochs=5,\n",
+    "    accelerator=\"cpu\",\n",
+    "    devices=1,\n",
+    "    enable_progress_bar=False,\n",
+    "    enable_model_summary=False,\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 19,
+   "id": "c06b9943",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "`Trainer.fit` stopped: `max_epochs=5` reached.\n"
+     ]
+    }
+   ],
+   "source": [
+    "trainer_mae.fit(model_mae, data_module)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "711f2f1a",
+   "metadata": {},
+   "source": [
+    "Now train the `QuantileLoss` model."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 20,
+   "id": "e0c7e151",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "`Trainer.fit` stopped: `max_epochs=5` reached.\n"
+     ]
+    }
+   ],
+   "source": [
+    "trainer_q.fit(model_q, data_module)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "d70ac0dd",
+   "metadata": {},
+   "source": [
+    "#### Output\n",
+    "\n",
+    "For inference we set up the test split and call `model.predict(test_dataloader, mode=...)`, which runs a Lightning `Trainer` internally and returns a dict with a `prediction` key. Use `mode=\"prediction\"` for the point model and `mode=\"quantiles\"` for the quantile model."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 21,
+   "id": "d947b3a5",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "GPU available: True (mps), used: False\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "TPU available: False, using: 0 TPU cores\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Point prediction shape: torch.Size([255, 1])\n",
+      "Point prediction: tensor([[ 0.3488],\n",
+      "        [ 0.5059],\n",
+      "        [ 0.6343],\n",
+      "        [ 0.7578],\n",
+      "        [ 0.8611],\n",
+      "        [ 0.9248],\n",
+      "        [ 0.9299],\n",
+      "        [ 0.9100],\n",
+      "        [ 0.8661],\n",
+      "        [ 0.7891],\n",
+      "        [ 0.6814],\n",
+      "        [ 0.5535],\n",
+      "        [ 0.3827],\n",
+      "        [ 0.2044],\n",
+      "        [ 0.0192],\n",
+      "        [-0.1669],\n",
+      "        [-0.3161],\n",
+      "        [ 0.3698],\n",
+      "        [ 0.5284],\n",
+      "        [ 0.6664],\n",
+      "        [ 0.7810],\n",
+      "        [ 0.8704],\n",
+      "        [ 0.9094],\n",
+      "        [ 0.9174],\n",
+      "        [ 0.9129],\n",
+      "        [ 0.8754],\n",
+      "        [ 0.7907],\n",
+      "        [ 0.6852],\n",
+      "        [ 0.5433],\n",
+      "        [ 0.3623],\n",
+      "        [ 0.1919],\n",
+      "        [ 0.0011],\n",
+      "        [-0.1583],\n",
+      "        [-0.3293],\n",
+      "        [ 0.3221],\n",
+      "        [ 0.4982],\n",
+      "        [ 0.6518],\n",
+      "        [ 0.7669],\n",
+      "        [ 0.8737],\n",
+      "        [ 0.9386],\n",
+      "        [ 0.9625],\n",
+      "        [ 0.9699],\n",
+      "        [ 0.9225],\n",
+      "        [ 0.8329],\n",
+      "        [ 0.7288],\n",
+      "        [ 0.5883],\n",
+      "        [ 0.4025],\n",
+      "        [ 0.2260],\n",
+      "        [ 0.0270],\n",
+      "        [-0.1689],\n",
+      "        [-0.3510],\n",
+      "        [ 0.3709],\n",
+      "        [ 0.5245],\n",
+      "        [ 0.6626],\n",
+      "        [ 0.7782],\n",
+      "        [ 0.8707],\n",
+      "        [ 0.9114],\n",
+      "        [ 0.9378],\n",
+      "        [ 0.9064],\n",
+      "        [ 0.8537],\n",
+      "        [ 0.7720],\n",
+      "        [ 0.6473],\n",
+      "        [ 0.5014],\n",
+      "        [ 0.3294],\n",
+      "        [ 0.1255],\n",
+      "        [-0.0468],\n",
+      "        [-0.2053],\n",
+      "        [-0.3445],\n",
+      "        [ 0.3461],\n",
+      "        [ 0.5025],\n",
+      "        [ 0.6340],\n",
+      "        [ 0.7541],\n",
+      "        [ 0.8552],\n",
+      "        [ 0.8974],\n",
+      "        [ 0.9192],\n",
+      "        [ 0.9146],\n",
+      "        [ 0.8652],\n",
+      "        [ 0.7836],\n",
+      "        [ 0.6785],\n",
+      "        [ 0.5342],\n",
+      "        [ 0.3724],\n",
+      "        [ 0.1925],\n",
+      "        [ 0.0052],\n",
+      "        [-0.1656],\n",
+      "        [-0.3205],\n",
+      "        [ 0.3227],\n",
+      "        [ 0.4775],\n",
+      "        [ 0.6339],\n",
+      "        [ 0.7542],\n",
+      "        [ 0.8405],\n",
+      "        [ 0.9109],\n",
+      "        [ 0.9486],\n",
+      "        [ 0.9309],\n",
+      "        [ 0.8976],\n",
+      "        [ 0.8153],\n",
+      "        [ 0.7016],\n",
+      "        [ 0.5509],\n",
+      "        [ 0.3996],\n",
+      "        [ 0.2076],\n",
+      "        [ 0.0354],\n",
+      "        [-0.1221],\n",
+      "        [-0.3016],\n",
+      "        [ 0.3404],\n",
+      "        [ 0.4948],\n",
+      "        [ 0.6283],\n",
+      "        [ 0.7628],\n",
+      "        [ 0.8589],\n",
+      "        [ 0.9118],\n",
+      "        [ 0.9261],\n",
+      "        [ 0.9265],\n",
+      "        [ 0.8945],\n",
+      "        [ 0.8029],\n",
+      "        [ 0.6820],\n",
+      "        [ 0.5675],\n",
+      "        [ 0.3966],\n",
+      "        [ 0.1972],\n",
+      "        [ 0.0257],\n",
+      "        [-0.1518],\n",
+      "        [-0.3015],\n",
+      "        [ 0.3244],\n",
+      "        [ 0.4852],\n",
+      "        [ 0.6275],\n",
+      "        [ 0.7462],\n",
+      "        [ 0.8522],\n",
+      "        [ 0.9115],\n",
+      "        [ 0.9424],\n",
+      "        [ 0.9502],\n",
+      "        [ 0.9168],\n",
+      "        [ 0.8320],\n",
+      "        [ 0.7119],\n",
+      "        [ 0.5590],\n",
+      "        [ 0.4020],\n",
+      "        [ 0.2250],\n",
+      "        [ 0.0349],\n",
+      "        [-0.1281],\n",
+      "        [-0.2896],\n",
+      "        [ 0.3270],\n",
+      "        [ 0.4817],\n",
+      "        [ 0.6173],\n",
+      "        [ 0.7281],\n",
+      "        [ 0.8225],\n",
+      "        [ 0.8909],\n",
+      "        [ 0.9342],\n",
+      "        [ 0.9350],\n",
+      "        [ 0.8892],\n",
+      "        [ 0.8050],\n",
+      "        [ 0.6944],\n",
+      "        [ 0.5455],\n",
+      "        [ 0.3916],\n",
+      "        [ 0.2246],\n",
+      "        [ 0.0224],\n",
+      "        [-0.1533],\n",
+      "        [-0.3149],\n",
+      "        [ 0.3825],\n",
+      "        [ 0.5227],\n",
+      "        [ 0.6552],\n",
+      "        [ 0.7667],\n",
+      "        [ 0.8442],\n",
+      "        [ 0.9101],\n",
+      "        [ 0.9470],\n",
+      "        [ 0.9203],\n",
+      "        [ 0.8463],\n",
+      "        [ 0.7771],\n",
+      "        [ 0.6784],\n",
+      "        [ 0.5306],\n",
+      "        [ 0.3680],\n",
+      "        [ 0.1760],\n",
+      "        [-0.0184],\n",
+      "        [-0.1717],\n",
+      "        [-0.3260],\n",
+      "        [ 0.3328],\n",
+      "        [ 0.5026],\n",
+      "        [ 0.6456],\n",
+      "        [ 0.7637],\n",
+      "        [ 0.8734],\n",
+      "        [ 0.9205],\n",
+      "        [ 0.9507],\n",
+      "        [ 0.9297],\n",
+      "        [ 0.8943],\n",
+      "        [ 0.7927],\n",
+      "        [ 0.6658],\n",
+      "        [ 0.5326],\n",
+      "        [ 0.3568],\n",
+      "        [ 0.1712],\n",
+      "        [-0.0269],\n",
+      "        [-0.2027],\n",
+      "        [-0.3453],\n",
+      "        [ 0.3091],\n",
+      "        [ 0.4582],\n",
+      "        [ 0.5828],\n",
+      "        [ 0.6981],\n",
+      "        [ 0.8046],\n",
+      "        [ 0.8765],\n",
+      "        [ 0.9052],\n",
+      "        [ 0.9249],\n",
+      "        [ 0.8764],\n",
+      "        [ 0.8027],\n",
+      "        [ 0.7078],\n",
+      "        [ 0.5764],\n",
+      "        [ 0.4075],\n",
+      "        [ 0.2525],\n",
+      "        [ 0.0670],\n",
+      "        [-0.1189],\n",
+      "        [-0.2594],\n",
+      "        [ 0.3604],\n",
+      "        [ 0.5369],\n",
+      "        [ 0.6708],\n",
+      "        [ 0.7692],\n",
+      "        [ 0.8614],\n",
+      "        [ 0.9426],\n",
+      "        [ 0.9516],\n",
+      "        [ 0.9290],\n",
+      "        [ 0.8794],\n",
+      "        [ 0.7970],\n",
+      "        [ 0.6993],\n",
+      "        [ 0.5636],\n",
+      "        [ 0.3785],\n",
+      "        [ 0.1931],\n",
+      "        [ 0.0211],\n",
+      "        [-0.1545],\n",
+      "        [-0.3122],\n",
+      "        [ 0.3462],\n",
+      "        [ 0.5141],\n",
+      "        [ 0.6610],\n",
+      "        [ 0.7852],\n",
+      "        [ 0.8467],\n",
+      "        [ 0.8948],\n",
+      "        [ 0.9374],\n",
+      "        [ 0.9398],\n",
+      "        [ 0.8970],\n",
+      "        [ 0.8152],\n",
+      "        [ 0.6901],\n",
+      "        [ 0.5336],\n",
+      "        [ 0.3569],\n",
+      "        [ 0.1655],\n",
+      "        [-0.0151],\n",
+      "        [-0.1643],\n",
+      "        [-0.3127],\n",
+      "        [ 0.3611],\n",
+      "        [ 0.5123],\n",
+      "        [ 0.6283],\n",
+      "        [ 0.7385],\n",
+      "        [ 0.8354],\n",
+      "        [ 0.8904],\n",
+      "        [ 0.9257],\n",
+      "        [ 0.9215],\n",
+      "        [ 0.8892],\n",
+      "        [ 0.8094],\n",
+      "        [ 0.6892],\n",
+      "        [ 0.5396],\n",
+      "        [ 0.3761],\n",
+      "        [ 0.2028],\n",
+      "        [ 0.0313],\n",
+      "        [-0.1376],\n",
+      "        [-0.2999]])\n"
+     ]
+    }
+   ],
+   "source": [
+    "data_module.setup(stage=\"test\")\n",
+    "test_dataloader = data_module.test_dataloader()\n",
+    "\n",
+    "preds_mae = model_mae.predict(\n",
+    "    test_dataloader,\n",
+    "    mode=\"prediction\",\n",
+    "    trainer_kwargs={\"accelerator\": \"cpu\", \"enable_progress_bar\": False},\n",
+    ")\n",
+    "print(\"Point prediction shape:\", preds_mae[\"prediction\"].shape)\n",
+    "print(\"Point prediction:\", preds_mae[\"prediction\"])"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "a17e8f8b",
+   "metadata": {},
+   "source": [
+    "Same for the `QuantileLoss` model."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 22,
+   "id": "aaa6dc00",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "GPU available: True (mps), used: False\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "TPU available: False, using: 0 TPU cores\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Quantile prediction shape (batch, pred_len, n_quantiles): torch.Size([255, 1, 3])\n",
+      "Quantile prediction: tensor([[[ 9.9807e-02,  2.7499e-01,  4.4650e-01]],\n",
+      "\n",
+      "        [[ 2.0286e-01,  4.5577e-01,  5.5673e-01]],\n",
+      "\n",
+      "        [[ 3.7001e-01,  5.8030e-01,  6.6626e-01]],\n",
+      "\n",
+      "        [[ 4.9127e-01,  7.1805e-01,  7.8477e-01]],\n",
+      "\n",
+      "        [[ 5.7864e-01,  7.9167e-01,  8.6310e-01]],\n",
+      "\n",
+      "        [[ 6.0197e-01,  8.7421e-01,  9.0665e-01]],\n",
+      "\n",
+      "        [[ 6.4056e-01,  9.0229e-01,  9.1842e-01]],\n",
+      "\n",
+      "        [[ 6.0831e-01,  8.7775e-01,  9.1337e-01]],\n",
+      "\n",
+      "        [[ 5.7948e-01,  8.5235e-01,  8.3199e-01]],\n",
+      "\n",
+      "        [[ 4.7244e-01,  7.5956e-01,  7.3842e-01]],\n",
+      "\n",
+      "        [[ 3.8413e-01,  6.6504e-01,  6.0270e-01]],\n",
+      "\n",
+      "        [[ 2.2860e-01,  5.3190e-01,  4.7483e-01]],\n",
+      "\n",
+      "        [[ 8.0860e-02,  3.8528e-01,  3.1630e-01]],\n",
+      "\n",
+      "        [[-6.6180e-02,  2.2654e-01,  1.6566e-01]],\n",
+      "\n",
+      "        [[-2.2870e-01,  3.9990e-02,  1.0635e-02]],\n",
+      "\n",
+      "        [[-4.5350e-01, -1.6615e-01, -1.6245e-01]],\n",
+      "\n",
+      "        [[-5.6181e-01, -3.0119e-01, -2.7940e-01]],\n",
+      "\n",
+      "        [[ 1.1397e-01,  3.1443e-01,  4.5570e-01]],\n",
+      "\n",
+      "        [[ 2.3875e-01,  4.5722e-01,  5.6592e-01]],\n",
+      "\n",
+      "        [[ 3.9594e-01,  6.1202e-01,  6.8070e-01]],\n",
+      "\n",
+      "        [[ 5.0909e-01,  7.2343e-01,  7.9554e-01]],\n",
+      "\n",
+      "        [[ 5.6820e-01,  8.2549e-01,  8.8526e-01]],\n",
+      "\n",
+      "        [[ 6.0274e-01,  8.7167e-01,  9.0742e-01]],\n",
+      "\n",
+      "        [[ 6.3060e-01,  9.1073e-01,  9.2737e-01]],\n",
+      "\n",
+      "        [[ 6.2148e-01,  8.6380e-01,  9.0087e-01]],\n",
+      "\n",
+      "        [[ 5.6531e-01,  8.4161e-01,  8.2112e-01]],\n",
+      "\n",
+      "        [[ 4.7440e-01,  7.6080e-01,  7.2122e-01]],\n",
+      "\n",
+      "        [[ 3.5833e-01,  6.6437e-01,  6.0321e-01]],\n",
+      "\n",
+      "        [[ 2.2504e-01,  5.3037e-01,  4.6037e-01]],\n",
+      "\n",
+      "        [[ 5.6557e-02,  3.6581e-01,  3.1029e-01]],\n",
+      "\n",
+      "        [[-6.2241e-02,  2.1161e-01,  1.4741e-01]],\n",
+      "\n",
+      "        [[-2.6160e-01,  9.4035e-04, -1.0257e-02]],\n",
+      "\n",
+      "        [[-4.4650e-01, -1.5157e-01, -1.6410e-01]],\n",
+      "\n",
+      "        [[-5.8261e-01, -3.1896e-01, -2.8520e-01]],\n",
+      "\n",
+      "        [[ 6.8224e-02,  2.6715e-01,  4.2531e-01]],\n",
+      "\n",
+      "        [[ 2.3004e-01,  4.4067e-01,  5.5171e-01]],\n",
+      "\n",
+      "        [[ 3.6883e-01,  5.7923e-01,  6.7432e-01]],\n",
+      "\n",
+      "        [[ 4.9196e-01,  7.2738e-01,  7.8510e-01]],\n",
+      "\n",
+      "        [[ 5.8808e-01,  8.1275e-01,  8.7540e-01]],\n",
+      "\n",
+      "        [[ 6.5047e-01,  8.7580e-01,  9.2384e-01]],\n",
+      "\n",
+      "        [[ 6.5986e-01,  9.3328e-01,  9.6060e-01]],\n",
+      "\n",
+      "        [[ 6.7055e-01,  9.2779e-01,  9.4879e-01]],\n",
+      "\n",
+      "        [[ 6.1443e-01,  8.8840e-01,  8.7499e-01]],\n",
+      "\n",
+      "        [[ 5.2145e-01,  8.1166e-01,  7.7975e-01]],\n",
+      "\n",
+      "        [[ 4.1959e-01,  7.0152e-01,  6.5948e-01]],\n",
+      "\n",
+      "        [[ 2.7865e-01,  5.7722e-01,  5.0482e-01]],\n",
+      "\n",
+      "        [[ 1.1372e-01,  3.9628e-01,  3.4282e-01]],\n",
+      "\n",
+      "        [[-5.8308e-02,  2.4688e-01,  1.7393e-01]],\n",
+      "\n",
+      "        [[-2.3804e-01,  5.6334e-02,  1.1986e-02]],\n",
+      "\n",
+      "        [[-4.3431e-01, -1.1985e-01, -1.5714e-01]],\n",
+      "\n",
+      "        [[-5.8965e-01, -3.2342e-01, -2.9989e-01]],\n",
+      "\n",
+      "        [[ 1.0963e-01,  3.1168e-01,  4.6216e-01]],\n",
+      "\n",
+      "        [[ 2.7842e-01,  4.8819e-01,  5.7088e-01]],\n",
+      "\n",
+      "        [[ 3.8399e-01,  6.1198e-01,  6.9901e-01]],\n",
+      "\n",
+      "        [[ 4.9422e-01,  7.1578e-01,  7.7966e-01]],\n",
+      "\n",
+      "        [[ 5.9026e-01,  8.3217e-01,  8.7925e-01]],\n",
+      "\n",
+      "        [[ 6.2625e-01,  8.6161e-01,  9.1631e-01]],\n",
+      "\n",
+      "        [[ 6.3366e-01,  8.7784e-01,  9.1430e-01]],\n",
+      "\n",
+      "        [[ 5.9915e-01,  8.8720e-01,  8.7842e-01]],\n",
+      "\n",
+      "        [[ 5.3163e-01,  8.0650e-01,  7.9515e-01]],\n",
+      "\n",
+      "        [[ 4.6380e-01,  7.5668e-01,  7.0034e-01]],\n",
+      "\n",
+      "        [[ 3.3820e-01,  6.1961e-01,  5.8047e-01]],\n",
+      "\n",
+      "        [[ 1.9239e-01,  4.9909e-01,  4.2081e-01]],\n",
+      "\n",
+      "        [[ 4.3815e-02,  3.3371e-01,  2.7643e-01]],\n",
+      "\n",
+      "        [[-1.5598e-01,  1.4883e-01,  1.0373e-01]],\n",
+      "\n",
+      "        [[-2.9957e-01, -1.2554e-02, -4.8678e-02]],\n",
+      "\n",
+      "        [[-4.4921e-01, -1.8426e-01, -1.7194e-01]],\n",
+      "\n",
+      "        [[-5.7228e-01, -3.4601e-01, -2.8218e-01]],\n",
+      "\n",
+      "        [[ 8.3240e-02,  2.8180e-01,  4.4251e-01]],\n",
+      "\n",
+      "        [[ 2.2670e-01,  4.5275e-01,  5.4464e-01]],\n",
+      "\n",
+      "        [[ 3.5842e-01,  5.7559e-01,  6.6990e-01]],\n",
+      "\n",
+      "        [[ 4.9171e-01,  7.0034e-01,  7.7952e-01]],\n",
+      "\n",
+      "        [[ 5.4633e-01,  7.8802e-01,  8.5538e-01]],\n",
+      "\n",
+      "        [[ 6.0242e-01,  8.6036e-01,  8.9241e-01]],\n",
+      "\n",
+      "        [[ 6.2993e-01,  8.9038e-01,  9.1905e-01]],\n",
+      "\n",
+      "        [[ 6.1258e-01,  8.7093e-01,  8.8347e-01]],\n",
+      "\n",
+      "        [[ 5.5138e-01,  8.2841e-01,  8.2372e-01]],\n",
+      "\n",
+      "        [[ 4.7361e-01,  7.6866e-01,  7.3051e-01]],\n",
+      "\n",
+      "        [[ 3.7874e-01,  6.5683e-01,  6.1570e-01]],\n",
+      "\n",
+      "        [[ 2.2295e-01,  5.2539e-01,  4.5199e-01]],\n",
+      "\n",
+      "        [[ 9.5070e-02,  3.7595e-01,  3.1711e-01]],\n",
+      "\n",
+      "        [[-8.5887e-02,  2.0173e-01,  1.4964e-01]],\n",
+      "\n",
+      "        [[-2.5123e-01,  2.3295e-02, -9.8590e-03]],\n",
+      "\n",
+      "        [[-4.3143e-01, -1.4560e-01, -1.4391e-01]],\n",
+      "\n",
+      "        [[-5.5544e-01, -3.0226e-01, -2.6597e-01]],\n",
+      "\n",
+      "        [[ 5.9465e-02,  2.5712e-01,  4.1841e-01]],\n",
+      "\n",
+      "        [[ 2.2534e-01,  4.2906e-01,  5.3150e-01]],\n",
+      "\n",
+      "        [[ 3.3970e-01,  5.6810e-01,  6.6276e-01]],\n",
+      "\n",
+      "        [[ 4.6560e-01,  6.7612e-01,  7.5248e-01]],\n",
+      "\n",
+      "        [[ 5.5287e-01,  7.9469e-01,  8.5845e-01]],\n",
+      "\n",
+      "        [[ 6.1629e-01,  8.6074e-01,  9.1417e-01]],\n",
+      "\n",
+      "        [[ 6.2908e-01,  8.9467e-01,  9.3152e-01]],\n",
+      "\n",
+      "        [[ 6.3811e-01,  9.0170e-01,  8.9450e-01]],\n",
+      "\n",
+      "        [[ 5.9622e-01,  8.4047e-01,  8.6211e-01]],\n",
+      "\n",
+      "        [[ 4.9510e-01,  7.9654e-01,  7.4223e-01]],\n",
+      "\n",
+      "        [[ 3.9885e-01,  6.7851e-01,  6.3447e-01]],\n",
+      "\n",
+      "        [[ 2.4988e-01,  5.6526e-01,  4.9061e-01]],\n",
+      "\n",
+      "        [[ 1.2074e-01,  3.7796e-01,  3.3539e-01]],\n",
+      "\n",
+      "        [[-7.4694e-02,  2.3034e-01,  1.7265e-01]],\n",
+      "\n",
+      "        [[-2.1436e-01,  6.2538e-02,  2.4097e-02]],\n",
+      "\n",
+      "        [[-3.9239e-01, -1.0717e-01, -1.1646e-01]],\n",
+      "\n",
+      "        [[-5.5500e-01, -2.7147e-01, -2.5749e-01]],\n",
+      "\n",
+      "        [[ 8.5695e-02,  2.6148e-01,  4.2282e-01]],\n",
+      "\n",
+      "        [[ 2.3940e-01,  4.5270e-01,  5.5483e-01]],\n",
+      "\n",
+      "        [[ 3.6641e-01,  5.7146e-01,  6.6453e-01]],\n",
+      "\n",
+      "        [[ 4.7790e-01,  7.2527e-01,  7.7900e-01]],\n",
+      "\n",
+      "        [[ 5.6712e-01,  8.0482e-01,  8.6703e-01]],\n",
+      "\n",
+      "        [[ 6.0549e-01,  8.8421e-01,  9.2169e-01]],\n",
+      "\n",
+      "        [[ 6.5790e-01,  8.9285e-01,  9.2020e-01]],\n",
+      "\n",
+      "        [[ 6.1083e-01,  8.9196e-01,  9.2065e-01]],\n",
+      "\n",
+      "        [[ 5.6070e-01,  8.4824e-01,  8.2934e-01]],\n",
+      "\n",
+      "        [[ 4.7873e-01,  7.7305e-01,  7.3288e-01]],\n",
+      "\n",
+      "        [[ 3.7758e-01,  6.8113e-01,  6.1767e-01]],\n",
+      "\n",
+      "        [[ 2.4868e-01,  5.4074e-01,  4.9440e-01]],\n",
+      "\n",
+      "        [[ 9.3219e-02,  3.8444e-01,  3.1803e-01]],\n",
+      "\n",
+      "        [[-8.1127e-02,  2.0306e-01,  1.6303e-01]],\n",
+      "\n",
+      "        [[-2.2893e-01,  4.1839e-02,  7.8391e-03]],\n",
+      "\n",
+      "        [[-4.2516e-01, -1.3712e-01, -1.4634e-01]],\n",
+      "\n",
+      "        [[-5.4870e-01, -2.8975e-01, -2.7055e-01]],\n",
+      "\n",
+      "        [[ 8.9124e-02,  2.7928e-01,  4.3621e-01]],\n",
+      "\n",
+      "        [[ 2.3139e-01,  4.4494e-01,  5.5406e-01]],\n",
+      "\n",
+      "        [[ 3.5274e-01,  5.8739e-01,  6.5556e-01]],\n",
+      "\n",
+      "        [[ 4.4339e-01,  7.2095e-01,  7.6852e-01]],\n",
+      "\n",
+      "        [[ 5.6372e-01,  8.1255e-01,  8.6896e-01]],\n",
+      "\n",
+      "        [[ 6.1433e-01,  8.7696e-01,  9.2724e-01]],\n",
+      "\n",
+      "        [[ 6.5213e-01,  8.9790e-01,  9.4222e-01]],\n",
+      "\n",
+      "        [[ 6.4048e-01,  9.0298e-01,  9.1548e-01]],\n",
+      "\n",
+      "        [[ 5.7732e-01,  8.5814e-01,  8.4854e-01]],\n",
+      "\n",
+      "        [[ 4.9995e-01,  7.9472e-01,  7.5559e-01]],\n",
+      "\n",
+      "        [[ 4.1071e-01,  7.1170e-01,  6.3694e-01]],\n",
+      "\n",
+      "        [[ 2.7086e-01,  5.4928e-01,  4.9757e-01]],\n",
+      "\n",
+      "        [[ 1.0377e-01,  3.9813e-01,  3.2809e-01]],\n",
+      "\n",
+      "        [[-7.8528e-02,  2.3026e-01,  1.6900e-01]],\n",
+      "\n",
+      "        [[-2.3633e-01,  5.8970e-02,  1.2067e-02]],\n",
+      "\n",
+      "        [[-3.5787e-01, -9.6255e-02, -1.1465e-01]],\n",
+      "\n",
+      "        [[-5.3583e-01, -2.8372e-01, -2.4692e-01]],\n",
+      "\n",
+      "        [[ 6.4040e-02,  2.6271e-01,  4.1785e-01]],\n",
+      "\n",
+      "        [[ 2.1890e-01,  4.0835e-01,  5.3605e-01]],\n",
+      "\n",
+      "        [[ 3.3648e-01,  5.6810e-01,  6.6188e-01]],\n",
+      "\n",
+      "        [[ 4.5166e-01,  6.7471e-01,  7.6441e-01]],\n",
+      "\n",
+      "        [[ 5.3114e-01,  7.8730e-01,  8.4444e-01]],\n",
+      "\n",
+      "        [[ 6.2036e-01,  8.5816e-01,  8.9598e-01]],\n",
+      "\n",
+      "        [[ 6.2312e-01,  8.8417e-01,  9.2671e-01]],\n",
+      "\n",
+      "        [[ 6.3499e-01,  8.9551e-01,  8.9948e-01]],\n",
+      "\n",
+      "        [[ 5.8168e-01,  8.5119e-01,  8.5915e-01]],\n",
+      "\n",
+      "        [[ 4.9857e-01,  7.8948e-01,  7.5264e-01]],\n",
+      "\n",
+      "        [[ 3.9344e-01,  6.6870e-01,  6.3241e-01]],\n",
+      "\n",
+      "        [[ 2.5457e-01,  5.4290e-01,  4.7884e-01]],\n",
+      "\n",
+      "        [[ 1.0606e-01,  3.8402e-01,  3.2660e-01]],\n",
+      "\n",
+      "        [[-5.6515e-02,  2.2615e-01,  1.6695e-01]],\n",
+      "\n",
+      "        [[-2.4996e-01,  4.9497e-02,  4.9640e-03]],\n",
+      "\n",
+      "        [[-4.2720e-01, -1.1386e-01, -1.3620e-01]],\n",
+      "\n",
+      "        [[-5.6947e-01, -2.9031e-01, -2.6498e-01]],\n",
+      "\n",
+      "        [[ 9.5746e-02,  3.0031e-01,  4.5950e-01]],\n",
+      "\n",
+      "        [[ 2.5609e-01,  4.8182e-01,  5.6761e-01]],\n",
+      "\n",
+      "        [[ 3.8856e-01,  6.1652e-01,  7.1072e-01]],\n",
+      "\n",
+      "        [[ 4.9245e-01,  7.1326e-01,  7.8025e-01]],\n",
+      "\n",
+      "        [[ 5.6165e-01,  8.0840e-01,  8.5138e-01]],\n",
+      "\n",
+      "        [[ 6.3295e-01,  8.7799e-01,  9.2276e-01]],\n",
+      "\n",
+      "        [[ 6.2531e-01,  8.9312e-01,  9.3346e-01]],\n",
+      "\n",
+      "        [[ 6.0895e-01,  8.8396e-01,  8.8771e-01]],\n",
+      "\n",
+      "        [[ 5.5706e-01,  8.3674e-01,  8.2919e-01]],\n",
+      "\n",
+      "        [[ 4.7386e-01,  7.5375e-01,  7.2319e-01]],\n",
+      "\n",
+      "        [[ 3.5052e-01,  6.3204e-01,  5.8431e-01]],\n",
+      "\n",
+      "        [[ 2.0327e-01,  5.3813e-01,  4.4271e-01]],\n",
+      "\n",
+      "        [[ 8.6660e-02,  3.5091e-01,  3.0531e-01]],\n",
+      "\n",
+      "        [[-1.0494e-01,  1.9784e-01,  1.3496e-01]],\n",
+      "\n",
+      "        [[-2.8547e-01,  6.0597e-03, -2.6737e-02]],\n",
+      "\n",
+      "        [[-4.4731e-01, -1.6901e-01, -1.7081e-01]],\n",
+      "\n",
+      "        [[-5.6197e-01, -3.3158e-01, -2.8447e-01]],\n",
+      "\n",
+      "        [[ 1.0085e-01,  2.8361e-01,  4.3741e-01]],\n",
+      "\n",
+      "        [[ 2.5792e-01,  4.5220e-01,  5.5045e-01]],\n",
+      "\n",
+      "        [[ 3.4085e-01,  5.7949e-01,  6.6339e-01]],\n",
+      "\n",
+      "        [[ 5.0523e-01,  7.1858e-01,  7.6561e-01]],\n",
+      "\n",
+      "        [[ 5.8831e-01,  8.1191e-01,  8.9661e-01]],\n",
+      "\n",
+      "        [[ 6.2844e-01,  8.6103e-01,  9.0313e-01]],\n",
+      "\n",
+      "        [[ 6.4458e-01,  9.0609e-01,  9.4090e-01]],\n",
+      "\n",
+      "        [[ 6.3490e-01,  8.9844e-01,  9.0340e-01]],\n",
+      "\n",
+      "        [[ 5.5762e-01,  8.3458e-01,  8.2804e-01]],\n",
+      "\n",
+      "        [[ 4.8105e-01,  7.7549e-01,  7.2353e-01]],\n",
+      "\n",
+      "        [[ 3.7026e-01,  6.7048e-01,  6.1581e-01]],\n",
+      "\n",
+      "        [[ 2.2807e-01,  5.1181e-01,  4.4185e-01]],\n",
+      "\n",
+      "        [[ 3.7930e-02,  3.6447e-01,  2.9856e-01]],\n",
+      "\n",
+      "        [[-1.1246e-01,  2.0081e-01,  1.3358e-01]],\n",
+      "\n",
+      "        [[-2.6385e-01,  6.7742e-03, -1.1569e-02]],\n",
+      "\n",
+      "        [[-4.5742e-01, -1.8700e-01, -1.8088e-01]],\n",
+      "\n",
+      "        [[-5.9037e-01, -3.5330e-01, -2.8659e-01]],\n",
+      "\n",
+      "        [[ 6.5153e-02,  2.3592e-01,  4.0362e-01]],\n",
+      "\n",
+      "        [[ 1.7578e-01,  3.8005e-01,  5.1560e-01]],\n",
+      "\n",
+      "        [[ 2.9644e-01,  5.2635e-01,  6.1373e-01]],\n",
+      "\n",
+      "        [[ 4.3944e-01,  6.6463e-01,  7.4343e-01]],\n",
+      "\n",
+      "        [[ 5.0876e-01,  7.3946e-01,  8.0405e-01]],\n",
+      "\n",
+      "        [[ 5.9904e-01,  8.2615e-01,  8.8129e-01]],\n",
+      "\n",
+      "        [[ 6.1663e-01,  8.7560e-01,  9.1641e-01]],\n",
+      "\n",
+      "        [[ 6.2345e-01,  8.7698e-01,  9.0942e-01]],\n",
+      "\n",
+      "        [[ 5.5821e-01,  8.4739e-01,  8.3576e-01]],\n",
+      "\n",
+      "        [[ 5.1653e-01,  7.9141e-01,  7.6143e-01]],\n",
+      "\n",
+      "        [[ 4.1469e-01,  6.7007e-01,  6.4260e-01]],\n",
+      "\n",
+      "        [[ 2.7160e-01,  5.6566e-01,  4.9475e-01]],\n",
+      "\n",
+      "        [[ 1.1492e-01,  4.1242e-01,  3.5776e-01]],\n",
+      "\n",
+      "        [[-3.8015e-02,  2.5949e-01,  2.0511e-01]],\n",
+      "\n",
+      "        [[-2.0979e-01,  8.4884e-02,  4.2242e-02]],\n",
+      "\n",
+      "        [[-3.6997e-01, -8.6048e-02, -1.0172e-01]],\n",
+      "\n",
+      "        [[-5.1088e-01, -2.4791e-01, -2.2313e-01]],\n",
+      "\n",
+      "        [[ 9.8262e-02,  3.1037e-01,  4.5498e-01]],\n",
+      "\n",
+      "        [[ 2.5056e-01,  4.9253e-01,  6.0234e-01]],\n",
+      "\n",
+      "        [[ 3.7600e-01,  6.4038e-01,  6.9735e-01]],\n",
+      "\n",
+      "        [[ 5.0658e-01,  7.4425e-01,  7.9832e-01]],\n",
+      "\n",
+      "        [[ 5.9397e-01,  8.4450e-01,  8.9360e-01]],\n",
+      "\n",
+      "        [[ 6.3720e-01,  8.8409e-01,  9.2933e-01]],\n",
+      "\n",
+      "        [[ 6.3565e-01,  9.1066e-01,  9.4339e-01]],\n",
+      "\n",
+      "        [[ 6.3572e-01,  9.1659e-01,  9.2350e-01]],\n",
+      "\n",
+      "        [[ 5.8792e-01,  8.4999e-01,  8.4288e-01]],\n",
+      "\n",
+      "        [[ 4.8319e-01,  7.7887e-01,  7.3816e-01]],\n",
+      "\n",
+      "        [[ 3.8246e-01,  6.7503e-01,  6.1371e-01]],\n",
+      "\n",
+      "        [[ 2.3944e-01,  5.3841e-01,  4.7206e-01]],\n",
+      "\n",
+      "        [[ 7.9835e-02,  3.8465e-01,  3.1268e-01]],\n",
+      "\n",
+      "        [[-7.1891e-02,  2.2273e-01,  1.6437e-01]],\n",
+      "\n",
+      "        [[-2.5790e-01,  2.9014e-02, -1.1495e-02]],\n",
+      "\n",
+      "        [[-4.2010e-01, -1.5221e-01, -1.6836e-01]],\n",
+      "\n",
+      "        [[-5.5266e-01, -2.9643e-01, -2.7069e-01]],\n",
+      "\n",
+      "        [[ 6.5341e-02,  2.7252e-01,  4.2122e-01]],\n",
+      "\n",
+      "        [[ 2.5403e-01,  4.7456e-01,  5.6344e-01]],\n",
+      "\n",
+      "        [[ 3.9813e-01,  6.0208e-01,  6.9627e-01]],\n",
+      "\n",
+      "        [[ 4.8692e-01,  7.1395e-01,  7.9074e-01]],\n",
+      "\n",
+      "        [[ 5.5250e-01,  8.2135e-01,  8.6205e-01]],\n",
+      "\n",
+      "        [[ 6.1700e-01,  8.7382e-01,  9.1996e-01]],\n",
+      "\n",
+      "        [[ 6.3738e-01,  8.9887e-01,  9.3028e-01]],\n",
+      "\n",
+      "        [[ 6.2956e-01,  8.8642e-01,  8.9588e-01]],\n",
+      "\n",
+      "        [[ 5.8343e-01,  8.5662e-01,  8.4109e-01]],\n",
+      "\n",
+      "        [[ 5.0555e-01,  7.7396e-01,  7.5080e-01]],\n",
+      "\n",
+      "        [[ 3.5991e-01,  6.6985e-01,  6.0834e-01]],\n",
+      "\n",
+      "        [[ 2.2618e-01,  5.4520e-01,  4.6742e-01]],\n",
+      "\n",
+      "        [[ 8.8920e-02,  3.5737e-01,  3.0228e-01]],\n",
+      "\n",
+      "        [[-9.1688e-02,  1.7764e-01,  1.2848e-01]],\n",
+      "\n",
+      "        [[-2.9603e-01,  1.1382e-02, -2.4436e-02]],\n",
+      "\n",
+      "        [[-4.3416e-01, -1.2885e-01, -1.4868e-01]],\n",
+      "\n",
+      "        [[-5.6478e-01, -2.8707e-01, -2.5098e-01]],\n",
+      "\n",
+      "        [[ 8.6806e-02,  2.9878e-01,  4.3477e-01]],\n",
+      "\n",
+      "        [[ 2.4188e-01,  4.4121e-01,  5.5636e-01]],\n",
+      "\n",
+      "        [[ 3.6820e-01,  6.0269e-01,  6.6721e-01]],\n",
+      "\n",
+      "        [[ 4.6699e-01,  6.7692e-01,  7.7065e-01]],\n",
+      "\n",
+      "        [[ 5.4361e-01,  8.1212e-01,  8.5542e-01]],\n",
+      "\n",
+      "        [[ 5.9484e-01,  8.5596e-01,  9.0012e-01]],\n",
+      "\n",
+      "        [[ 6.3045e-01,  8.8823e-01,  9.2780e-01]],\n",
+      "\n",
+      "        [[ 6.2737e-01,  8.7979e-01,  8.9424e-01]],\n",
+      "\n",
+      "        [[ 5.6011e-01,  8.4219e-01,  8.4310e-01]],\n",
+      "\n",
+      "        [[ 4.9571e-01,  7.8150e-01,  7.3383e-01]],\n",
+      "\n",
+      "        [[ 3.7415e-01,  6.6597e-01,  6.1190e-01]],\n",
+      "\n",
+      "        [[ 2.3752e-01,  5.3129e-01,  4.6802e-01]],\n",
+      "\n",
+      "        [[ 7.8355e-02,  3.8520e-01,  3.1256e-01]],\n",
+      "\n",
+      "        [[-5.8871e-02,  2.0440e-01,  1.6026e-01]],\n",
+      "\n",
+      "        [[-2.5583e-01,  2.8542e-02,  3.4320e-03]],\n",
+      "\n",
+      "        [[-3.9615e-01, -1.2049e-01, -1.3581e-01]],\n",
+      "\n",
+      "        [[-5.4790e-01, -2.8266e-01, -2.4942e-01]]])\n"
+     ]
+    }
+   ],
+   "source": [
+    "preds_q = model_q.predict(\n",
+    "    test_dataloader,\n",
+    "    mode=\"quantiles\",\n",
+    "    trainer_kwargs={\"accelerator\": \"cpu\", \"enable_progress_bar\": False},\n",
+    ")\n",
+    "print(\n",
+    "    \"Quantile prediction shape (batch, pred_len, n_quantiles):\",\n",
+    "    preds_q[\"prediction\"].shape,\n",
+    ")\n",
+    "print(\"Quantile prediction:\", preds_q[\"prediction\"])"
+   ]
   }
  ],
  "metadata": {