Update to PyTorch 1.12 (nv-morpheus#523)

- Update to PyTorch 1.12.
- Latest PyTorch is v1.13 but is built for CUDA 11.6/11.7. Morpheus and RAPIDS officially support up to CUDA 11.5.
- Fixes errors seen in training scripts/notebooks when using large batch sizes.
- Updates `dfencoder` from `22.09` to `22.11`
- Updated `phishing` and `root cause` inference scripts to move `cudf` imports before `torch` imports. Allows to get past this error which also occurs with PyTorch 1.10:
```
Traceback (most recent call last):
  File "root-cause-inference.py", line 33, in <module>
    import cudf
  File "/opt/conda/envs/morpheus/lib/python3.8/site-packages/cudf/__init__.py", line 5, in <module>
    validate_setup()
  File "/opt/conda/envs/morpheus/lib/python3.8/site-packages/cudf/utils/gpu_utils.py", line 18, in validate_setup
    from cuda.cudart import cudaDeviceAttr, cudaError_t
ImportError: /usr/lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.29' not found (required by /opt/conda/envs/morpheus/lib/python3.8/site-packages/cuda/cudart.cpython-38-x86_64-linux-gnu.so)
````

Close nv-morpheus#491

Authors:
  - Eli Fajardo (https://github.com/efajardo-nv)

Approvers:
  - Michael Demoret (https://github.com/mdemoret-nv)

URL: nv-morpheus#523

docker/conda/environments/requirements.txt

@@ -4,12 +4,12 @@
 
 ####### Pip-only runtime dependencies (keep sorted!) #######
 # Packages listed here should also be listed in setup.py
-git+https://github.com/nv-morpheus/dfencoder.git@branch-22.09#egg=dfencoder
+git+https://github.com/nv-morpheus/dfencoder.git@branch-23.01#egg=dfencoder
 ipywidgets
 jupyter-core>=4.11.2,<5.0
 jupyterlab
 nvidia-pyindex
 pytest-benchmark>=4.0
-torch==1.10.2+cu113
+torch==1.12.0+cu113
 tritonclient[all]==2.17.*
 websockets

models/validation-inference-scripts/dfp-models/requirements.txt

@@ -6,4 +6,4 @@ matplotlib==3.4.2
 numpy==1.22.0
 pandas==1.0.1
 scikit_learn==1.0.2
-torch==1.7.1
+torch==1.12.0+cu113

models/validation-inference-scripts/log-parsing-models/log-parsing-inference.py

@@ -110,8 +110,7 @@ def preprocess(self, raw_data_col, stride_len=64, max_seq_len=256):
         for symbol in string.punctuation:
             raw_data_col = raw_data_col.str.replace(symbol, ' ' + symbol + ' ')
 
-        byte_count = raw_data_col.str.byte_count()
-        max_rows_tensor = int((byte_count / 120).ceil().sum())
+        max_rows_tensor = len(raw_data_col) * 2
 
         tokenizer = SubwordTokenizer(self._hashpath, do_lower_case=False)
         tokenizer_output = tokenizer(raw_data_col,

models/validation-inference-scripts/phishing-models/phish-bert-inference-script.py

@@ -21,6 +21,13 @@
     --output phishing-email-validation-output.jsonlines
 """
 
+###########################################################################################
+# cudf imports moved before torch import to avoid the following error:
+# ImportError: /usr/lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.29' not found
+import cudf
+from cudf.core.subword_tokenizer import SubwordTokenizer
+###########################################################################################
+
 import argparse
 import json
 
@@ -29,9 +36,6 @@
 import torch
 from scipy.special import expit
 
-import cudf
-from cudf.core.subword_tokenizer import SubwordTokenizer
-
 
 def infer(validationdata, vocab, model, output):
 

models/validation-inference-scripts/root-cause-models/root-cause-inference.py

@@ -22,6 +22,13 @@
     --output root-cause-validation-output.jsonlines
 """
 
+###########################################################################################
+# cudf imports moved before torch import to avoid the following error:
+# ImportError: /usr/lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.29' not found
+import cudf
+from cudf.core.subword_tokenizer import SubwordTokenizer
+###########################################################################################
+
 import argparse
 import json
 
@@ -30,9 +37,6 @@
 import torch
 from scipy.special import expit
 
-import cudf
-from cudf.core.subword_tokenizer import SubwordTokenizer
-
 
 def infer(
     validationdata,

setup.py

@@ -42,8 +42,8 @@
     install_requires=[
         # Only list the packages which cannot be installed via conda here. Should mach the requirements in
         # docker/conda/environments/requirements.txt
-        "dfencoder @ git+https://github.com/nv-morpheus/dfencoder.git@branch-22.09#egg=dfencoder",
-        "torch==1.10.2+cu113",
+        "dfencoder @ git+https://github.com/nv-morpheus/dfencoder.git@branch-23.01#egg=dfencoder",
+        "torch==1.12.0+cu113",
         "tritonclient[all]==2.17.*",  # Force to 2.17 since they require grpcio==1.41 for newer versions
     ],
     license="Apache",