Skip to content

Commit

Permalink
remove unused import
Browse files Browse the repository at this point in the history
  • Loading branch information
@romanlutz
romanlutz committed Nov 8, 2024
1 parent 15ec5f6 commit dfbdeab
Show file tree
Hide file tree
Showing 2 changed files with 196 additions and 198 deletions.
393 changes: 196 additions & 197 deletions doc/code/auxiliary_attacks/2_gcg_azure_ml.ipynb
View file
Original file line number Diff line number Diff line change
@@ -1,197 +1,196 @@
{
"cells": [
{
"cell_type": "markdown",
"id": "44a2ab29",
"metadata": {},
"source": [
"# Generating GCG Suffixes Using Azure Machine Learning"
]
},
{
"cell_type": "markdown",
"id": "4c2b3f55",
"metadata": {},
"source": [
"This notebook shows how to generate GCG suffixes using Azure Machine Learning (AML), which consists of three main steps:\n",
"1. Connect to an Azure Machine Learning (AML) workspace.\n",
"2. Create AML Environment with the Python dependencies.\n",
"3. Submit a training job to AML."
]
},
{
"cell_type": "markdown",
"id": "697c48fe",
"metadata": {},
"source": [
"## Connect to Azure Machine Learning Workspace"
]
},
{
"cell_type": "markdown",
"id": "632bef8b",
"metadata": {},
"source": [
"The [workspace](https://docs.microsoft.com/en-us/azure/machine-learning/concept-workspace) is the top-level resource for Azure Machine Learning (AML), providing a centralized place to work with all the artifacts you create when using AML. In this section, we will connect to the workspace in which the job will be run.\n",
"\n",
"To connect to a workspace, we need identifier parameters - a subscription, resource group and workspace name. We will use these details in the `MLClient` from `azure.ai.ml` to get a handle to the required AML workspace. We use the [default Azure authentication](https://docs.microsoft.com/en-us/python/api/azure-identity/azure.identity.defaultazurecredential?view=azure-python) for this tutorial."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "8645ef34",
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"from pyrit.common import default_values\n",
"\n",
"default_values.load_default_env()\n",
"\n",
"# Enter details of your AML workspace\n",
"subscription_id = os.environ.get(\"AZURE_SUBSCRIPTION_ID\")\n",
"resource_group = os.environ.get(\"AZURE_RESOURCE_GROUP\")\n",
"workspace = os.environ.get(\"AZURE_ML_WORKSPACE_NAME\")\n",
"compute_name = os.environ.get(\"AZURE_ML_COMPUTE_NAME\")\n",
"print(workspace)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "37b282e7",
"metadata": {},
"outputs": [],
"source": [
"from azure.ai.ml import MLClient\n",
"from azure.identity import DefaultAzureCredential\n",
"\n",
"# Get a handle to the workspace\n",
"ml_client = MLClient(DefaultAzureCredential(), subscription_id, resource_group, workspace)"
]
},
{
"cell_type": "markdown",
"id": "7d7c5c41",
"metadata": {},
"source": [
"## Create Compute Cluster\n",
"\n",
"Before proceeding, create a compute cluster in Azure ML. The following command may be useful:\n",
"az ml compute create --size Standard_ND96isrf_H100_v5 --type AmlCompute --name <compute-name> -g <group> -w <workspace> --min-instances 0"
]
},
{
"cell_type": "markdown",
"id": "e390439f",
"metadata": {},
"source": [
"## Create AML Environment"
]
},
{
"cell_type": "markdown",
"id": "ceedcfa7",
"metadata": {},
"source": [
"To install the dependencies needed to run GCG, we create an AML environment from a [Dockerfile](../../../pyrit/auxiliary_attacks/gcg/src/Dockerfile)."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "67783454",
"metadata": {
"lines_to_next_cell": 2
},
"outputs": [],
"source": [
"from pathlib import Path\n",
"from pyrit.common.path import HOME_PATH\n",
"from azure.ai.ml.entities import Environment, BuildContext\n",
"\n",
"# Configure the AML environment with path to Dockerfile and dependencies\n",
"env_docker_context = Environment(\n",
" build=BuildContext(path=Path(HOME_PATH) / \"pyrit\" / \"auxiliary_attacks\" / \"gcg\" / \"src\"),\n",
" name=\"pyrit\",\n",
" description=\"PyRIT environment created from a Docker context.\",\n",
")\n",
"\n",
"# Create or update the AML environment\n",
"ml_client.environments.create_or_update(env_docker_context)"
]
},
{
"cell_type": "markdown",
"id": "49d20c8e",
"metadata": {},
"source": [
"## Submit Training Job to AML"
]
},
{
"cell_type": "markdown",
"id": "ad33e623",
"metadata": {},
"source": [
"Finally, we configure the command to run the GCG algorithm. The entry file for the algorithm is [`run.py`](../../../pyrit/auxiliary_attacks/gcg/experiments/run.py), which takes several command line arguments, as shown below. We also have to specify the compute `instance_type` to run the algorithm on. In our experience, a GPU instance with at least 32GB of vRAM is required. In the example below, we use Standard_ND40rs_v2.\n",
"\n",
"Depending on the compute instance you use, you may encounter \"out of memory\" errors. In this case, we recommend training on a smaller model or lowering `n_train_data` or `batch_size`."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "938a1030",
"metadata": {},
"outputs": [],
"source": [
"from azure.ai.ml import command\n",
"from azure.ai.ml.entities import JobResourceConfiguration\n",
"\n",
"# Configure the command\n",
"job = command(\n",
" code=Path(HOME_PATH),\n",
" command=\"cd pyrit/auxiliary_attacks/gcg/experiments && python run.py --model_name ${{inputs.model_name}} --setup ${{inputs.setup}} --n_train_data ${{inputs.n_train_data}} --n_test_data ${{inputs.n_test_data}} --n_steps ${{inputs.n_steps}} --batch_size ${{inputs.batch_size}}\",\n",
" inputs={\n",
" \"model_name\": \"phi_3_mini\",\n",
" \"setup\": \"multiple\",\n",
" \"n_train_data\": 25,\n",
" \"n_test_data\": 0,\n",
" \"n_steps\": 500,\n",
" \"batch_size\": 256,\n",
" },\n",
" environment=f\"{env_docker_context.name}:{env_docker_context.version}\",\n",
" environment_variables={\"HF_TOKEN\": os.environ[\"HF_TOKEN\"]},\n",
" display_name=\"suffix_generation\",\n",
" description=\"Generate a suffix for attacking LLMs.\",\n",
" compute=compute_name,\n",
")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b40591b0",
"metadata": {},
"outputs": [],
"source": [
"# Submit the command\n",
"returned_job = ml_client.create_or_update(job)"
]
}
],
"metadata": {
"jupytext": {
"cell_metadata_filter": "-all"
},
"kernelspec": {
"display_name": "pyrit-kernel",
"language": "python",
"name": "pyrit-kernel"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
{
"cells": [
{
"cell_type": "markdown",
"id": "44a2ab29",
"metadata": {},
"source": [
"# Generating GCG Suffixes Using Azure Machine Learning"
]
},
{
"cell_type": "markdown",
"id": "4c2b3f55",
"metadata": {},
"source": [
"This notebook shows how to generate GCG suffixes using Azure Machine Learning (AML), which consists of three main steps:\n",
"1. Connect to an Azure Machine Learning (AML) workspace.\n",
"2. Create AML Environment with the Python dependencies.\n",
"3. Submit a training job to AML."
]
},
{
"cell_type": "markdown",
"id": "697c48fe",
"metadata": {},
"source": [
"## Connect to Azure Machine Learning Workspace"
]
},
{
"cell_type": "markdown",
"id": "632bef8b",
"metadata": {},
"source": [
"The [workspace](https://docs.microsoft.com/en-us/azure/machine-learning/concept-workspace) is the top-level resource for Azure Machine Learning (AML), providing a centralized place to work with all the artifacts you create when using AML. In this section, we will connect to the workspace in which the job will be run.\n",
"\n",
"To connect to a workspace, we need identifier parameters - a subscription, resource group and workspace name. We will use these details in the `MLClient` from `azure.ai.ml` to get a handle to the required AML workspace. We use the [default Azure authentication](https://docs.microsoft.com/en-us/python/api/azure-identity/azure.identity.defaultazurecredential?view=azure-python) for this tutorial."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "8645ef34",
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"from pyrit.common import default_values\n",
"\n",
"default_values.load_default_env()\n",
"\n",
"# Enter details of your AML workspace\n",
"subscription_id = os.environ.get(\"AZURE_SUBSCRIPTION_ID\")\n",
"resource_group = os.environ.get(\"AZURE_RESOURCE_GROUP\")\n",
"workspace = os.environ.get(\"AZURE_ML_WORKSPACE_NAME\")\n",
"compute_name = os.environ.get(\"AZURE_ML_COMPUTE_NAME\")\n",
"print(workspace)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "37b282e7",
"metadata": {},
"outputs": [],
"source": [
"from azure.ai.ml import MLClient\n",
"from azure.identity import DefaultAzureCredential\n",
"\n",
"# Get a handle to the workspace\n",
"ml_client = MLClient(DefaultAzureCredential(), subscription_id, resource_group, workspace)"
]
},
{
"cell_type": "markdown",
"id": "7d7c5c41",
"metadata": {},
"source": [
"## Create Compute Cluster\n",
"\n",
"Before proceeding, create a compute cluster in Azure ML. The following command may be useful:\n",
"az ml compute create --size Standard_ND96isrf_H100_v5 --type AmlCompute --name <compute-name> -g <group> -w <workspace> --min-instances 0"
]
},
{
"cell_type": "markdown",
"id": "e390439f",
"metadata": {},
"source": [
"## Create AML Environment"
]
},
{
"cell_type": "markdown",
"id": "ceedcfa7",
"metadata": {},
"source": [
"To install the dependencies needed to run GCG, we create an AML environment from a [Dockerfile](../../../pyrit/auxiliary_attacks/gcg/src/Dockerfile)."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "67783454",
"metadata": {
"lines_to_next_cell": 2
},
"outputs": [],
"source": [
"from pathlib import Path\n",
"from pyrit.common.path import HOME_PATH\n",
"from azure.ai.ml.entities import Environment, BuildContext\n",
"\n",
"# Configure the AML environment with path to Dockerfile and dependencies\n",
"env_docker_context = Environment(\n",
" build=BuildContext(path=Path(HOME_PATH) / \"pyrit\" / \"auxiliary_attacks\" / \"gcg\" / \"src\"),\n",
" name=\"pyrit\",\n",
" description=\"PyRIT environment created from a Docker context.\",\n",
")\n",
"\n",
"# Create or update the AML environment\n",
"ml_client.environments.create_or_update(env_docker_context)"
]
},
{
"cell_type": "markdown",
"id": "49d20c8e",
"metadata": {},
"source": [
"## Submit Training Job to AML"
]
},
{
"cell_type": "markdown",
"id": "ad33e623",
"metadata": {},
"source": [
"Finally, we configure the command to run the GCG algorithm. The entry file for the algorithm is [`run.py`](../../../pyrit/auxiliary_attacks/gcg/experiments/run.py), which takes several command line arguments, as shown below. We also have to specify the compute `instance_type` to run the algorithm on. In our experience, a GPU instance with at least 32GB of vRAM is required. In the example below, we use Standard_ND40rs_v2.\n",
"\n",
"Depending on the compute instance you use, you may encounter \"out of memory\" errors. In this case, we recommend training on a smaller model or lowering `n_train_data` or `batch_size`."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "938a1030",
"metadata": {},
"outputs": [],
"source": [
"from azure.ai.ml import command\n",
"\n",
"# Configure the command\n",
"job = command(\n",
" code=Path(HOME_PATH),\n",
" command=\"cd pyrit/auxiliary_attacks/gcg/experiments && python run.py --model_name ${{inputs.model_name}} --setup ${{inputs.setup}} --n_train_data ${{inputs.n_train_data}} --n_test_data ${{inputs.n_test_data}} --n_steps ${{inputs.n_steps}} --batch_size ${{inputs.batch_size}}\",\n",
" inputs={\n",
" \"model_name\": \"phi_3_mini\",\n",
" \"setup\": \"multiple\",\n",
" \"n_train_data\": 25,\n",
" \"n_test_data\": 0,\n",
" \"n_steps\": 500,\n",
" \"batch_size\": 256,\n",
" },\n",
" environment=f\"{env_docker_context.name}:{env_docker_context.version}\",\n",
" environment_variables={\"HF_TOKEN\": os.environ[\"HF_TOKEN\"]},\n",
" display_name=\"suffix_generation\",\n",
" description=\"Generate a suffix for attacking LLMs.\",\n",
" compute=compute_name,\n",
")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b40591b0",
"metadata": {},
"outputs": [],
"source": [
"# Submit the command\n",
"returned_job = ml_client.create_or_update(job)"
]
}
],
"metadata": {
"jupytext": {
"cell_metadata_filter": "-all"
},
"kernelspec": {
"display_name": "pyrit-kernel",
"language": "python",
"name": "pyrit-kernel"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
Loading

0 comments on commit dfbdeab

Please sign in to comment.