Chronos: train tcn model on gpu and speed up inference on cpu

docs/readthedocs/source/_static/js/chronos_tutorial.js

@@ -51,7 +51,8 @@ $(".checkboxes").click(function(){
             var ids = ["ChronosForecaster","TuneaForecasting","AutoTSEstimator","AutoWIDE",
             "MultvarWIDE","MultstepWIDE","LSTMForecaster","AutoProphet","AnomalyDetection",
             "DeepARmodel","TFTmodel","hyperparameter","taxiDataset","distributedFashion",
-            "ONNX","Quantize","TCMFForecaster","PenalizeUnderestimation"];
+            "ONNX","Quantize","TCMFForecaster","PenalizeUnderestimation",
+            "GPUtrainingCPUacceleration"];
             showTutorials(ids);
             var disIds = ["simulation"];
             disCheck(disIds);
@@ -94,7 +95,7 @@ $(".checkboxes").click(function(){
             disCheck(disIds);
         }
         else if(vals.includes("customized_model")){
-            var ids = ["AutoTSEstimator","DeepARmodel","TFTmodel"];
+            var ids = ["AutoTSEstimator","DeepARmodel","TFTmodel", "GPUtrainingCPUacceleration"];
             showTutorials(ids);
             var disIds = ["anomaly_detection","simulation","onnxruntime","quantization","distributed"];
             disCheck(disIds);
@@ -114,7 +115,7 @@ $(".checkboxes").click(function(){
             disCheck(disIds);
         }
         else if(vals.includes("forecast") && vals.includes("customized_model")){
-            var ids = ["DeepARmodel","TFTmodel","AutoTSEstimator"];
+            var ids = ["DeepARmodel","TFTmodel","AutoTSEstimator","GPUtrainingCPUacceleration"];
             showTutorials(ids);
             var disIds = ["anomaly_detection","simulation","onnxruntime","quantization","distributed"];
             disCheck(disIds);

docs/readthedocs/source/doc/Chronos/QuickStart/index.md

@@ -244,6 +244,16 @@
         </details>
         <hr>
 
+        <details id="GPUtrainingCPUacceleration">
+            <summary>
+                <a href="https://github.com/intel-analytics/BigDL/tree/main/python/chronos/example/inference-acceleration">Accelerate the inference speed of model trained on other platform</a>
+                <p>Tag: <button value="forecast">forecast</button>&nbsp;<button value="customized_model">customized model</button></p>
+            </summary>
+            <img src="../../../_images/GitHub-Mark-32px.png"><a href="https://github.com/intel-analytics/BigDL/tree/main/python/chronos/example/inference-acceleration">View source on GitHub</a>
+            <p>In this example, we show an example to train the model on GPU and accelerate the model by using onnxruntime on CPU.</p>
+        </details>
+        <hr>
+
     </div>
 
     <script src="../../../_static/js/chronos_tutorial.js"></script> 

python/chronos/example/inference-acceleration/cpu_inference_acceleration.py

@@ -0,0 +1,90 @@
+#
+# Copyright 2016 The BigDL Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+import torch
+from bigdl.chronos.pytorch import TSTrainer as Trainer
+from bigdl.chronos.model.tcn import model_creator
+from bigdl.chronos.metric.forecast_metrics import Evaluator
+from bigdl.chronos.data.repo_dataset import get_public_dataset
+from sklearn.preprocessing import StandardScaler
+
+def gen_dataloader():
+    tsdata_train, tsdata_val,\
+        tsdata_test = get_public_dataset(name='nyc_taxi',
+                                         with_split=True,
+                                         val_ratio=0.1,
+                                         test_ratio=0.1)
+
+    stand = StandardScaler()
+    for tsdata in [tsdata_train, tsdata_val, tsdata_test]:
+        tsdata.deduplicate()\
+            .impute()\
+            .gen_dt_feature()\
+            .scale(stand, fit=tsdata is tsdata_train)\
+            .roll(lookback=48,horizon=1)
+
+    tsdata_traindataloader = tsdata_train.to_torch_data_loader(batch_size=32)
+    tsdata_valdataloader = tsdata_val.to_torch_data_loader(batch_size=32, shuffle=False)
+    tsdata_testdataloader = tsdata_test.to_torch_data_loader(batch_size=32, shuffle=False)
+
+    return tsdata_traindataloader, tsdata_valdataloader, tsdata_testdataloader
+
+def predict_wraper(model, input_sample):
+    model(input_sample)
+
+if __name__ == '__main__':
+
+    # create data loaders for train/valid/test
+    tsdata_traindataloader,\
+    tsdata_valdataloader,\
+    tsdata_testdataloader = gen_dataloader()
+
+    # create a model
+    # This could be an arbitrary model, we choose to use a built-in model TCN here
+    config = {'input_feature_num':8,
+              'output_feature_num':1,
+              'past_seq_len':48,
+              'future_seq_len':1,
+              'kernel_size':3,
+              'repo_initialization':True,
+              'dropout':0.1,
+              'seed': 0,
+              'num_channels':[30]*7
+              }
+    model = model_creator(config)
+    loss = torch.nn.MSELoss()
+    optimizer = torch.optim.Adam(lr=0.001, params=model.parameters())
+    lit_model = Trainer.compile(model, loss, optimizer)
+
+    # train the model
+    # You may use any method to train the model either on gpu or cpu
+    trainer = Trainer(max_epochs=3,
+                      accelerator='gpu', 
+                      devices=1,
+                     )
+    trainer.fit(lit_model, tsdata_traindataloader, tsdata_testdataloader)
+
+    # get an input sample
+    x = None
+    for x, _ in tsdata_traindataloader:
+        break
+    input_sample = x[0].unsqueeze(0)
+
+    # speed up the model using Chronos TSTrainer
+    speed_model = Trainer.trace(lit_model, accelerator="onnxruntime", input_sample=input_sample)
+
+    # evaluate the model's latency
+    print("original pytorch latency (ms):", Evaluator.get_latency(predict_wraper, lit_model, input_sample))
+    print("onnxruntime latency (ms):", Evaluator.get_latency(predict_wraper, speed_model, input_sample))

python/chronos/example/inference-acceleration/readme.md

@@ -0,0 +1,20 @@
+# Accelerate the inference speed of model trained on other platform
+
+## Introduction
+Chronos has many built-in models wrapped in forecasters, detectors and simulators optimized on CPU (especially intel CPU) platform.
+
+While users may want to use their own model or built-in models trained on another platform (e.g. GPU) but prefer to carry out the inferencing process on CPU platform. Chronos can also help users to accelerate their model for inferencing.
+
+In this example, we show an example to train the model on GPU and accelerate the model by using onnxruntime on CPU.
+
+## How to run this example
+```bash
+python cpu_inference_acceleration.py
+```
+
+## Sample output
+```bash
+Epoch 2: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████| 288/288
+original pytorch latency (ms): {'p50': 1.236, 'p90': 1.472, 'p95': 1.612, 'p99': 32.989}
+onnxruntime latency (ms): {'p50': 0.124, 'p90': 0.129, 'p95': 0.148, 'p99': 0.363}
+```