Onboard PT INF to autopatch (#5068)

* build PT25 INF EC2

* build PT25 INF SM

* format json file

* build PT2.5 INF EC2

* build PT2.6 INF EC2

* format file

* build PT26 INF SM

* revert toml

---------

Co-authored-by: Yadan Wei <[email protected]>

Author: Yadan-Wei

miscellaneous_scripts/dockerfile_patch_script.sh

@@ -40,11 +40,6 @@ if [[ $LATEST_RELEASED_IMAGE_URI =~ ^763104351884\.dkr\.ecr\.us-west-2\.amazonaw
     chmod +x /usr/local/bin/dockerd-entrypoint.py
 fi
 
-# For PT 2.4, 2.5 and 2.6 inference, install openssh-client to make mpi4py working
-if [[ $LATEST_RELEASED_IMAGE_URI =~ ^763104351884\.dkr\.ecr\.us-west-2\.amazonaws\.com/pytorch-inference:2\.[4-6]\.[0-9]+-gpu ]]; then
-    apt update && apt install -y --no-install-recommends openssh-client openssh-server && echo "Installed openssh-client openssh-server"
-fi
-
 # For PT 2.6, 2.7, rerun license file
 if [[ $LATEST_RELEASED_IMAGE_URI =~ ^763104351884\.dkr\.ecr\.us-west-2\.amazonaws\.com/pytorch-(.+):2\.6 ]]; then
     curl -o /license.txt https://aws-dlc-licenses.s3.amazonaws.com/pytorch-2.6/license.txt

pytorch/inference/buildspec-2-5-ec2.yml

@@ -5,7 +5,7 @@ framework: &FRAMEWORK pytorch
 version: &VERSION 2.5.1
 short_version: &SHORT_VERSION "2.5"
 arch_type: x86
-#autopatch_build: "True"
+autopatch_build: "True"
 
 repository_info:
   inference_repository: &INFERENCE_REPOSITORY

pytorch/inference/buildspec-2-5-sm.yml

@@ -5,7 +5,7 @@ framework: &FRAMEWORK pytorch
 version: &VERSION 2.5.1
 short_version: &SHORT_VERSION "2.5"
 arch_type: x86
-# autopatch_build: "True"
+autopatch_build: "True"
 
 repository_info:
   inference_repository: &INFERENCE_REPOSITORY

pytorch/inference/buildspec-2-6-ec2.yml

@@ -5,7 +5,7 @@ framework: &FRAMEWORK pytorch
 version: &VERSION 2.6.0
 short_version: &SHORT_VERSION "2.6"
 arch_type: x86
-# autopatch_build: "True"
+autopatch_build: "True"
 
 repository_info:
   inference_repository: &INFERENCE_REPOSITORY

pytorch/inference/buildspec-2-6-sm.yml

@@ -5,7 +5,7 @@ framework: &FRAMEWORK pytorch
 version: &VERSION 2.6.0
 short_version: &SHORT_VERSION "2.6"
 arch_type: x86
-# autopatch_build: "True"
+autopatch_build: "True"
 
 repository_info:
   inference_repository: &INFERENCE_REPOSITORY

pytorch/inference/docker/2.5/py3/Dockerfile.ec2.cpu.core_packages.json

@@ -26,5 +26,8 @@
   },
   "torch-model-archiver": {
     "version_specifier": "==0.12.0"
+  },
+  "urllib3": {
+    "version_specifier": ">=2.5.0"
   }
 }

pytorch/inference/docker/2.5/py3/Dockerfile.sagemaker.cpu.core_packages.json

@@ -29,5 +29,8 @@
   },
   "sagemaker-pytorch-inference": {
     "version_specifier": "==2.0.25"
+  },
+  "urllib3": {
+    "version_specifier": ">=2.5.0"
   }
 }

pytorch/inference/docker/2.5/py3/cu124/Dockerfile.ec2.gpu.core_packages.json

@@ -26,5 +26,11 @@
   },
   "torch-model-archiver": {
     "version_specifier": "==0.12.0"
+  },
+  "urllib3": {
+    "version_specifier": ">=2.5.0"
+  },
+  "triton": {
+    "version_specifier": "==3.1.0"
   }
 }

pytorch/inference/docker/2.5/py3/cu124/Dockerfile.sagemaker.gpu.core_packages.json

@@ -29,5 +29,11 @@
   },
   "sagemaker-pytorch-inference": {
     "version_specifier": "==2.0.25"
+  },
+  "urllib3": {
+    "version_specifier": ">=2.5.0"
+  },
+  "triton": {
+    "version_specifier": "==3.1.0"
   }
 }

pytorch/inference/docker/2.6/py3/cu124/Dockerfile.ec2.gpu.core_packages.json

@@ -30,5 +30,8 @@
   },
   "torch-model-archiver": {
     "version_specifier": "==0.12.0"
+  },
+  "triton": {
+    "version_specifier": "==3.2.0"
   }
 }