init

pirocheto · pirocheto · commit 150315590222 · 2025-01-12T19:36:11.000+01:00
diff --git a/.DS_Store b/.DS_Store
diff --git a/.github/workflows/model-to-s3-from-hf-hub.yaml b/.github/workflows/model-to-s3-from-hf-hub.yaml
@@ -0,0 +1,49 @@
+name: Deploy HuggingFace Model to S3 from HuggingFace Hub
+
+on:
+  push:
+    branches:
+      - test
+      - main
+  workflow_dispatch:
+
+permissions:
+  id-token: write
+  contents: read
+
+env:
+  AWS_REGION: "us-east-1"
+  HF_MODEL_NAME: "tabularisai/multilingual-sentiment-analysis"
+  ENVIRONMENT: ${{ github.ref_name == 'main' && 'prod' || github.ref_name == 'test' && 'test' }}
+
+jobs:
+  deploy_model:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout Code
+        uses: actions/checkout@v3
+
+      - name: Deploy Hugging Face Model
+        run: |
+          git clone ${{ env.HF_MODEL_NAME }}
+          cd ${{ env.HF_MODEL_NAME }}
+
+      - name: Create Archive
+        run: |
+          cd ${{ env.HF_MODEL_NAME }}
+          tar -czvf model.tar.gz *
+
+      - name: Configure AWS Credentials
+        uses: aws-actions/configure-aws-credentials@v4
+        with:
+          audience: sts.amazonaws.com
+          aws-region: ${{ env.AWS_REGION }}
+          role-to-assume: arn:aws:iam::639269844451:role/github-actions
+          role-session-name: testing-deployment
+
+      - name: Upload Model to S3
+        env:
+          S3_BUCKET_NAME: "dev-sagemaker-us-east-1-639269844451"
+          S3_MODEL_PATH: "${{ env.ENVIRONMENT }}/${{ env.HF_MODEL_NAME }}/model.tar.gz"
+        run: s3 cp model.tar.gz s3://${{ env.S3_BUCKET_NAME }}/${{ env.S3_MODEL_PATH }}
diff --git a/.gitignore b/.gitignore
@@ -0,0 +1,18 @@
+# Python-generated files
+__pycache__/
+*.py[oc]
+build/
+dist/
+wheels/
+*.egg-info
+
+# Virtual environments
+.venv
+
+.pytest_cache/
+hf_model/
+model.tar.gz
+
+
+.terraform/
+.terraform.lock.hcl
diff --git a/.python-version b/.python-version
@@ -0,0 +1 @@
+3.12
diff --git a/Makefile b/Makefile
@@ -0,0 +1,25 @@
+ENV = dev
+
+.PHONY: init
+init:
+	terraform -chdir=terraform init
+
+.PHONY: validate
+validate:
+	terraform -chdir=terraform validate
+
+.PHONY: plan
+plan:
+	terraform -chdir=terraform plan
+
+.PHONY: apply
+apply:
+	terraform -chdir=terraform apply -auto-approve
+
+.PHONY: destroy
+destroy:
+	terraform -chdir=terraform destroy
+
+.PHONY: workspace
+workspace:
+	terraform -chdir=terraform workspace select -or-create ${ENV}
diff --git a/README.md b/README.md
diff --git a/pyproject.toml b/pyproject.toml
@@ -0,0 +1,34 @@
+[project]
+name = "aws-hf-model-deployment"
+version = "0.1.0"
+description = "Add your description here"
+readme = "README.md"
+requires-python = ">=3.12"
+dependencies = [
+    "huggingface-hub[cli]>=0.27.1",
+    "humanize>=4.11.0",
+    "transformers[torch]>=4.48.0",
+]
+
+[dependency-groups]
+dev = [
+    "boto3>=1.35.97",
+    "pytest>=8.3.4",
+    "sagemaker-huggingface-inference-toolkit>=2.4.1",
+]
+
+[tool.mypy]
+disallow_untyped_calls = true
+
+[tool.ruff]
+target-version = "py312"
+
+[tool.ruff.lint]
+select = [
+    "F",    # Pyflakes
+    "I",    # isort
+    "E",    # pycodestyle
+    "S",    # flake8-bandit
+    "N",    # pep8-naming
+    "PERF",
+]
diff --git a/terraform/main.tf b/terraform/main.tf
@@ -0,0 +1,20 @@
+locals {
+  env = terraform.workspace
+}
+
+module "model" {
+  source               = "./modules/aws-sagemaker-hf-pytorch-inference-model-deployment"
+  aws_region           = "us-east-1"
+  model_author_name    = "tabularisai"
+  model_name           = "multilingual-sentiment-analysis"
+  model_src            = abspath("../model.tar.gz")
+  python_version       = "py310"
+  ubuntu_version       = "22.04"
+  transformers_version = "4.37.0"
+  pytorch_version      = "2.1.0"
+  model_bucket_name    = "sagemaker-us-east-1-639269844451"
+  model_bucket_key     = "${local.env}/tabularisai/multilingual-sentiment-analysis/model.tar.gz"
+  instance_type        = "ml.m5.large"
+  memory_size_in_mb    = 2048
+  max_concurrency      = 1
+}
diff --git a/terraform/modules/aws-sagemaker-hf-pytorch-inference-model-deployment/main.tf b/terraform/modules/aws-sagemaker-hf-pytorch-inference-model-deployment/main.tf
@@ -0,0 +1,127 @@
+data "aws_caller_identity" "current" {}
+
+
+locals {
+  env            = terraform.workspace
+  account_id     = data.aws_caller_identity.current.account_id
+  ecr_image_name = "763104351884.dkr.ecr.${var.aws_region}.amazonaws.com/huggingface-pytorch-inference"
+  ecr_image_tag  = "${var.pytorch_version}-transformers${var.transformers_version}-cpu-${var.python_version}-ubuntu${var.ubuntu_version}"
+}
+
+resource "aws_s3_object" "model_upload" {
+  bucket = var.model_bucket_name
+  key    = var.model_bucket_key
+  source = var.model_src
+  etag   = filemd5(var.model_src)
+}
+
+resource "aws_iam_role" "model_execution_role" {
+  name = "${local.env}-sagemaker-${var.model_name}-execution-role"
+
+  assume_role_policy = jsonencode({
+    Version = "2012-10-17"
+    Statement = [
+      {
+        Effect = "Allow"
+        Principal = {
+          Service = "sagemaker.amazonaws.com"
+        }
+        Action = "sts:AssumeRole"
+      },
+    ]
+  })
+}
+
+resource "aws_iam_policy" "s3_policy" {
+  name = "${local.env}-sagemaker-${var.model_name}-s3-policy"
+
+  policy = jsonencode({
+    "Version" : "2012-10-17",
+    "Statement" : [
+      {
+        "Effect" : "Allow",
+        "Action" : [
+          "s3:GetObject",
+          "s3:ListBucket",
+        ],
+        "Resource" : "arn:aws:s3:::${var.model_bucket_name}/*"
+      }
+    ]
+  })
+}
+
+resource "aws_iam_policy" "ecr_policy" {
+  name = "${local.env}-sagemaker-${var.model_name}-ecr-policy"
+
+  policy = jsonencode({
+    "Version" : "2012-10-17",
+    "Statement" : [
+      {
+        "Effect" : "Allow",
+        "Action" : [
+          "ecr:GetAuthorizationToken",
+          "ecr:BatchCheckLayerAvailability",
+          "ecr:GetDownloadUrlForLayer",
+          "ecr:GetRepositoryPolicy",
+          "ecr:DescribeRepositories",
+          "ecr:ListImages",
+          "ecr:DescribeImages",
+          "ecr:BatchGetImage",
+          "ecr:GetLifecyclePolicy",
+          "ecr:GetLifecyclePolicyPreview",
+          "ecr:ListTagsForResource",
+          "ecr:DescribeImageScanFindings",
+        ],
+        "Resource" : "arn:aws:ecr:${var.aws_region}:*:repository/*"
+      }
+    ]
+  })
+}
+
+resource "aws_iam_role_policy_attachment" "s3_policy_attachment" {
+  policy_arn = aws_iam_policy.s3_policy.arn
+  role       = aws_iam_role.model_execution_role.name
+}
+
+resource "aws_iam_role_policy_attachment" "ecr_policy_attachment" {
+  policy_arn = aws_iam_policy.ecr_policy.arn
+  role       = aws_iam_role.model_execution_role.name
+}
+
+
+resource "aws_sagemaker_model" "model" {
+  name               = var.model_name
+  execution_role_arn = aws_iam_role.model_execution_role.arn
+  # primary_container {
+  #   image          = "${local.ecr_image_name}:${local.ecr_image_tag}"
+  #   model_data_url = "s3://${aws_s3_object.model_upload.bucket}/${aws_s3_object.model_upload.key}"
+  # }
+
+  primary_container {
+    image = "763104351884.dkr.ecr.us-east-1.amazonaws.com/huggingface-pytorch-inference:2.1.0-transformers4.37.0-cpu-py310-ubuntu22.04"
+    environment = {
+      HF_MODEL_ID = "distilbert-base-uncased"
+      HF_TASK     = "text-classification"
+    }
+  }
+}
+
+resource "aws_sagemaker_endpoint_configuration" "model_endpoint_configuration" {
+  name = "${local.env}-sagemaker-${var.model_name}-endpoint-config"
+  production_variants {
+    initial_variant_weight = 1
+    model_name             = aws_sagemaker_model.model.name
+    variant_name           = "Version1"
+
+    serverless_config {
+      memory_size_in_mb = var.memory_size_in_mb
+      max_concurrency   = var.max_concurrency
+    }
+  }
+}
+
+resource "aws_sagemaker_endpoint" "model_endpoint" {
+  endpoint_config_name = aws_sagemaker_endpoint_configuration.model_endpoint_configuration.name
+  name                 = "${local.env}-sagemaker-${var.model_name}-endpoint"
+}
+
diff --git a/terraform/modules/aws-sagemaker-hf-pytorch-inference-model-deployment/outputs.tf b/terraform/modules/aws-sagemaker-hf-pytorch-inference-model-deployment/outputs.tf
@@ -0,0 +1,4 @@
+
+output "sagemaker_endpoint_name" {
+  value = aws_sagemaker_endpoint.model_endpoint.name
+}
diff --git a/terraform/modules/aws-sagemaker-hf-pytorch-inference-model-deployment/variables.tf b/terraform/modules/aws-sagemaker-hf-pytorch-inference-model-deployment/variables.tf
@@ -0,0 +1,58 @@
+
+
+variable "aws_region" {
+  description = "AWS region"
+  default     = "us-east-1"
+}
+
+variable "model_author_name" {
+  description = "Author name of the model"
+}
+
+variable "model_name" {
+  description = "Name of the model"
+}
+
+variable "model_src" {
+  description = "Path to the model.tar.gz file"
+}
+
+variable "python_version" {
+  description = "Python version"
+}
+
+variable "ubuntu_version" {
+  description = "Ubuntu version"
+}
+
+variable "transformers_version" {
+  description = "Transformers version"
+}
+
+variable "pytorch_version" {
+  description = "PyTorch version"
+}
+
+
+variable "model_bucket_name" {
+  description = "Name of the S3 bucket to store the model"
+}
+
+variable "model_bucket_key" {
+  description = "Key of the S3 bucket to store the model"
+}
+
+variable "instance_type" {
+  description = "Instance type for the endpoint"
+  default     = "ml.m5.large"
+}
+
+variable "memory_size_in_mb" {
+  description = "Memory size in MB for the endpoint"
+  default     = 2048
+}
+
+variable "max_concurrency" {
+  description = "Maximum concurrency for the endpoint"
+  default     = 1
+}
diff --git a/terraform/outputs.tf b/terraform/outputs.tf
@@ -0,0 +1,3 @@
+output "sagemaker_endpoint_name" {
+  value = module.model.sagemaker_endpoint_name
+}
diff --git a/terraform/provider.tf b/terraform/provider.tf
@@ -0,0 +1,20 @@
+terraform {
+  required_providers {
+    aws = {
+      source  = "hashicorp/aws"
+      version = "~> 5.0"
+    }
+  }
+
+  required_version = ">= 1.3.0"
+
+  backend "s3" {
+    bucket = "terraform-states-us-east-1-639269844451"
+    key    = "sagemaker-hf-model-deployment/terraform.tfstate"
+    region = "us-east-1"
+  }
+}
+
+provider "aws" {
+  region = var.aws_region
+}
diff --git a/terraform/variables.tf b/terraform/variables.tf
@@ -0,0 +1,4 @@
+variable "aws_region" {
+  description = "AWS region"
+  default     = "us-east-1"
+}
diff --git a/test_endpoint.py b/test_endpoint.py
diff --git a/uv.lock b/uv.lock

-Original file line number
+Diff line change
@@ @@ -0,0 +1,4 @@ @@
++
 +output "sagemaker_endpoint_name" {
 +  value = aws_sagemaker_endpoint.model_endpoint.name
 +}