Deprecate torchscript frontend

Author: narendasan

.pre-commit-config.yaml

@@ -1,7 +1,7 @@
 exclude: ^.github/actions/assigner/dist
 repos:
     - repo: https://github.com/pre-commit/pre-commit-hooks
-      rev: v4.5.0
+      rev: v5.0.0
       hooks:
           - id: check-yaml
           - id: trailing-whitespace
@@ -32,7 +32,7 @@ repos:
       hooks:
           - id: validate-pyproject
     - repo: https://github.com/pycqa/isort
-      rev: 5.13.2
+      rev: 6.0.0
       hooks:
           - id: isort
             name: isort (python)
@@ -52,7 +52,7 @@ repos:
           - id: black
             exclude: ^examples/custom_converters/elu_converter/setup.py|^docs
     - repo: https://github.com/crate-ci/typos
-      rev: v1.22.9
+      rev: typos-dict-v0.12.4
       hooks:
           - id: typos
     - repo: https://github.com/astral-sh/uv-pre-commit

cpp/include/torch_tensorrt/macros.h

@@ -30,6 +30,7 @@
   STR(TORCH_TENSORRT_MAJOR_VERSION) \
   "." STR(TORCH_TENSORRT_MINOR_VERSION) "." STR(TORCH_TENSORRT_PATCH_VERSION)
 
+#define TORCH_TENSORRT_PTQ_DEPRECATION [[deprecated("Int8 PTQ Calibrator has been deprecated by TensorRT, please plan on porting to a NVIDIA Model Optimizer Toolkit based workflow. See: https://pytorch.org/TensorRT/tutorials/_rendered_examples/dynamo/vgg16_ptq.html for more details")]]
 // Setup namespace aliases for ease of use
 namespace torch_tensorrt {
 namespace torchscript {}

cpp/include/torch_tensorrt/ptq.h

@@ -59,7 +59,7 @@ class Int8Calibrator : Algorithm {
    * calibration cache
    * @param use_cache : bool - Whether to use the cache (if it exists)
    */
-  Int8Calibrator(DataLoaderUniquePtr dataloader, const std::string& cache_file_path, bool use_cache)
+   Int8Calibrator(DataLoaderUniquePtr dataloader, const std::string& cache_file_path, bool use_cache)
       : dataloader_(dataloader.get()), cache_file_path_(cache_file_path), use_cache_(use_cache) {
     for (auto batch : *dataloader_) {
       batched_data_.push_back(batch.data);
@@ -308,9 +308,8 @@ class Int8CacheCalibrator : Algorithm {
  * @param use_cache: bool - use calibration cache
  * @return Int8Calibrator<Algorithm, DataLoader>
  */
-
 template <typename Algorithm = nvinfer1::IInt8EntropyCalibrator2, typename DataLoader>
-inline Int8Calibrator<Algorithm, DataLoader> make_int8_calibrator(
+TORCH_TENSORRT_PTQ_DEPRECATION inline Int8Calibrator<Algorithm, DataLoader> make_int8_calibrator(
     DataLoader dataloader,
     const std::string& cache_file_path,
     bool use_cache) {
@@ -344,7 +343,7 @@ inline Int8Calibrator<Algorithm, DataLoader> make_int8_calibrator(
  * @return Int8CacheCalibrator<Algorithm>
  */
 template <typename Algorithm = nvinfer1::IInt8EntropyCalibrator2>
-inline Int8CacheCalibrator<Algorithm> make_int8_cache_calibrator(const std::string& cache_file_path) {
+TORCH_TENSORRT_PTQ_DEPRECATION inline Int8CacheCalibrator<Algorithm> make_int8_cache_calibrator(const std::string& cache_file_path) {
   return Int8CacheCalibrator<Algorithm>(cache_file_path);
 }
 

py/torch_tensorrt/ts/_compiler.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import warnings
 from typing import Any, List, Optional, Sequence, Set, Tuple
 
 import torch
@@ -102,6 +103,12 @@ def compile(
         torch.jit.ScriptModule: Compiled TorchScript Module, when run it will execute via TensorRT
     """
 
+    warnings.warn(
+        'The torchscript frontend for Torch-TensorRT has been deprecated, please plan on porting to the dynamo frontend (torch_tensorrt.compile(..., ir="dynamo"). Torchscript will continue to be a supported deployment format via post compilation torchscript tracing, see: https://pytorch.org/TensorRT/user_guide/saving_models.html for more details',
+        DeprecationWarning,
+        stacklevel=2
+    )
+
     input_list = list(inputs) if inputs is not None else []
     enabled_precisions_set = (
         enabled_precisions if enabled_precisions is not None else set()
@@ -240,6 +247,12 @@ def convert_method_to_trt_engine(
     Returns:
         bytes: Serialized TensorRT engine, can either be saved to a file or deserialized via TensorRT APIs
     """
+    warnings.warn(
+        'The torchscript frontend for Torch-TensorRT has been deprecated, please plan on porting to the dynamo frontend (torch_tensorrt.convert_method_to_trt_engine(..., ir="dynamo"). Torchscript will continue to be a supported deployment format via post compilation torchscript tracing, see: https://pytorch.org/TensorRT/user_guide/saving_models.html for more details',
+        DeprecationWarning,
+        stacklevel=2
+    )
+
     input_list = list(inputs) if inputs is not None else []
     enabled_precisions_set = (
         enabled_precisions if enabled_precisions is not None else {torch.float}

py/torch_tensorrt/ts/ptq.py

@@ -13,6 +13,8 @@
 from torch_tensorrt import _C
 from torch_tensorrt.ts.logging import Level, log
 
+import warnings
+
 
 class CalibrationAlgo(Enum):
     ENTROPY_CALIBRATION = _C.CalibrationAlgo.ENTROPY_CALIBRATION
@@ -88,6 +90,11 @@ def __init__(self, **kwargs: Any):
         pass
 
     def __new__(cls, *args: Any, **kwargs: Any) -> Self:
+        warnings.warn(
+            "Int8 PTQ Calibrator has been deprecated by TensorRT, please plan on porting to a NVIDIA Model Optimizer Toolkit based workflow. See: https://pytorch.org/TensorRT/tutorials/_rendered_examples/dynamo/vgg16_ptq.html for more details",
+            DeprecationWarning,
+            stacklevel=2
+        )
         dataloader = args[0]
         algo_type = kwargs.get("algo_type", CalibrationAlgo.ENTROPY_CALIBRATION_2)
         cache_file = kwargs.get("cache_file", None)
@@ -175,6 +182,11 @@ def __init__(self, **kwargs: Any):
         pass
 
     def __new__(cls, *args: Any, **kwargs: Any) -> Self:
+        warnings.warn(
+            "Int8 PTQ Calibrator has been deprecated by TensorRT, please plan on porting to a NVIDIA Model Optimizer Toolkit based workflow. See: https://pytorch.org/TensorRT/tutorials/_rendered_examples/dynamo/vgg16_ptq.html for more details",
+            DeprecationWarning,
+            stacklevel=2
+        )
         cache_file = args[0]
         algo_type = kwargs.get("algo_type", CalibrationAlgo.ENTROPY_CALIBRATION_2)