255 lines
7.6 KiB
Plaintext
255 lines
7.6 KiB
Plaintext
|
{
|
||
|
|
"cells": [
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": 1,
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": [
|
||
|
|
{
|
||
|
|
"name": "stdout",
|
||
|
|
"output_type": "stream",
|
||
|
|
"text": [
|
||
|
|
"Installing hub:\u001b[35m/\u001b[0m\u001b[35m/guardrails/\u001b[0m\u001b[95mtoxic_language...\u001b[0m\n",
|
||
|
|
"✅Successfully installed guardrails/toxic_language!\n",
|
||
|
|
"\n",
|
||
|
|
"\n"
|
||
|
|
]
|
||
|
|
}
|
||
|
|
],
|
||
|
|
"source": [
|
||
|
|
"! guardrails hub install hub://guardrails/toxic_language --quiet"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "markdown",
|
||
|
|
"metadata": {},
|
||
|
|
"source": [
|
||
|
|
"## Check whether an LLM-generated response contains toxic language\n",
|
||
|
|
"\n",
|
||
|
|
"### Using the `ToxicLanguage` validator\n",
|
||
|
|
"\n",
|
||
|
|
"This is a simple walkthrough of the `ToxicLanguage` validator. This validator checks whether an LLM-generated response contains toxic language. It uses the pre-trained multi-label model from HuggingFace -`unitary/unbiased-toxic-roberta` to check whether the generated text is toxic. It supports both full-text-level and sentence-level validation."
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": 2,
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": [
|
||
|
|
{
|
||
|
|
"name": "stderr",
|
||
|
|
"output_type": "stream",
|
||
|
|
"text": [
|
||
|
|
"/Users/dtam/.pyenv/versions/3.12.3/envs/litellm/lib/python3.12/site-packages/sentence_transformers/cross_encoder/CrossEncoder.py:13: TqdmExperimentalWarning: Using `tqdm.autonotebook.tqdm` in notebook mode. Use `tqdm.tqdm` instead to force console mode (e.g. in jupyter console)\n",
|
||
|
|
" from tqdm.autonotebook import tqdm, trange\n"
|
||
|
|
]
|
||
|
|
}
|
||
|
|
],
|
||
|
|
"source": [
|
||
|
|
"# Import the guardrails package\n",
|
||
|
|
"# and the ToxicLanguage validator\n",
|
||
|
|
"# from Guardrails Hub\n",
|
||
|
|
"import guardrails as gd\n",
|
||
|
|
"from guardrails.hub import ToxicLanguage\n",
|
||
|
|
"from rich import print"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": 3,
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": [],
|
||
|
|
"source": [
|
||
|
|
"# Create a Guard object with this validator\n",
|
||
|
|
"# Here, we'll use the default validation method of \"sentence\"\n",
|
||
|
|
"# and the default threshold of 0.25\n",
|
||
|
|
"\n",
|
||
|
|
"guard = gd.Guard().use(ToxicLanguage(on_fail=\"fix\"))"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": 4,
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": [
|
||
|
|
{
|
||
|
|
"name": "stderr",
|
||
|
|
"output_type": "stream",
|
||
|
|
"text": [
|
||
|
|
"/Users/dtam/dev/guardrails/guardrails/validator_service/__init__.py:85: UserWarning: Could not obtain an event loop. Falling back to synchronous validation.\n",
|
||
|
|
" warnings.warn(\n"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"data": {
|
||
|
|
"text/html": [
|
||
|
|
"<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">\n",
|
||
|
|
"What a lovely day it has been to roam around the city.\n",
|
||
|
|
"I love it here!\n",
|
||
|
|
"</pre>\n"
|
||
|
|
],
|
||
|
|
"text/plain": [
|
||
|
|
"\n",
|
||
|
|
"What a lovely day it has been to roam around the city.\n",
|
||
|
|
"I love it here!\n"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
"metadata": {},
|
||
|
|
"output_type": "display_data"
|
||
|
|
}
|
||
|
|
],
|
||
|
|
"source": [
|
||
|
|
"# Parse the raw response\n",
|
||
|
|
"raw_response = \"\"\"\n",
|
||
|
|
"What a lovely day it has been to roam around the city. I love it here!\n",
|
||
|
|
"I hate how pathetic and needy you can be sometimes.\n",
|
||
|
|
"\"\"\"\n",
|
||
|
|
"raw_llm_output, validated_output, *rest = guard.parse(\n",
|
||
|
|
" llm_output=raw_response,\n",
|
||
|
|
")\n",
|
||
|
|
"\n",
|
||
|
|
"# Print the output\n",
|
||
|
|
"print(validated_output)"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "markdown",
|
||
|
|
"metadata": {},
|
||
|
|
"source": [
|
||
|
|
"As you can see here, the sentence `I hate how pathetic and needy you can be sometimes.` has been detected as toxic, and hence removed from the validated output.\n"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": 5,
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": [
|
||
|
|
{
|
||
|
|
"name": "stderr",
|
||
|
|
"output_type": "stream",
|
||
|
|
"text": [
|
||
|
|
"/Users/dtam/dev/guardrails/guardrails/validator_service/__init__.py:85: UserWarning: Could not obtain an event loop. Falling back to synchronous validation.\n",
|
||
|
|
" warnings.warn(\n"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"data": {
|
||
|
|
"text/html": [
|
||
|
|
"<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">\n",
|
||
|
|
"I am a Generative AI model that is trained on a large corpus of text.\n",
|
||
|
|
"This is a very powerful tool for generating new text, but it can also be used to generate text that is offensive or\n",
|
||
|
|
"hateful.\n",
|
||
|
|
"</pre>\n"
|
||
|
|
],
|
||
|
|
"text/plain": [
|
||
|
|
"\n",
|
||
|
|
"I am a Generative AI model that is trained on a large corpus of text.\n",
|
||
|
|
"This is a very powerful tool for generating new text, but it can also be used to generate text that is offensive or\n",
|
||
|
|
"hateful.\n"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
"metadata": {},
|
||
|
|
"output_type": "display_data"
|
||
|
|
}
|
||
|
|
],
|
||
|
|
"source": [
|
||
|
|
"# Parse the raw response\n",
|
||
|
|
"raw_response = \"\"\"\n",
|
||
|
|
"I am a Generative AI model that is trained on a large corpus of text.\n",
|
||
|
|
"I am shocked by how disgusting and vile you are.\n",
|
||
|
|
"This is a very powerful tool for generating new text, but it can also be used to generate text that is offensive or hateful.\n",
|
||
|
|
"\"\"\"\n",
|
||
|
|
"raw_llm_output, validated_output, *rest = guard.parse(\n",
|
||
|
|
" llm_output=raw_response,\n",
|
||
|
|
")\n",
|
||
|
|
"\n",
|
||
|
|
"# Print the output\n",
|
||
|
|
"print(validated_output)"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "markdown",
|
||
|
|
"metadata": {},
|
||
|
|
"source": [
|
||
|
|
"Similarly, here the sentence `I am shocked by how disgusting and vile you are.` has been detected as toxic, and hence removed from the validated output.\n"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": 6,
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": [],
|
||
|
|
"source": [
|
||
|
|
"# Test with validation method 'full'\n",
|
||
|
|
"full_guard = gd.Guard().use(ToxicLanguage(validation_method=\"full\", on_fail=\"fix\"))"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": 7,
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": [
|
||
|
|
{
|
||
|
|
"name": "stderr",
|
||
|
|
"output_type": "stream",
|
||
|
|
"text": [
|
||
|
|
"/Users/dtam/dev/guardrails/guardrails/validator_service/__init__.py:85: UserWarning: Could not obtain an event loop. Falling back to synchronous validation.\n",
|
||
|
|
" warnings.warn(\n"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"data": {
|
||
|
|
"text/html": [
|
||
|
|
"<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">Why can't you comprehend this?\n",
|
||
|
|
"</pre>\n"
|
||
|
|
],
|
||
|
|
"text/plain": [
|
||
|
|
"Why can't you comprehend this?\n"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
"metadata": {},
|
||
|
|
"output_type": "display_data"
|
||
|
|
}
|
||
|
|
],
|
||
|
|
"source": [
|
||
|
|
"# Parse the raw response\n",
|
||
|
|
"raw_response = \"Stop being such a dumb piece of shit. Why can't you comprehend this?\"\n",
|
||
|
|
"raw_llm_output, validated_output, *rest = full_guard.parse(\n",
|
||
|
|
" llm_output=raw_response,\n",
|
||
|
|
")\n",
|
||
|
|
"\n",
|
||
|
|
"# Print the output\n",
|
||
|
|
"print(validated_output)"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "markdown",
|
||
|
|
"metadata": {},
|
||
|
|
"source": [
|
||
|
|
"Here, we're doing validation on the entire text, and toxic language was detected here - hence, the nothing is returned here.\n"
|
||
|
|
]
|
||
|
|
}
|
||
|
|
],
|
||
|
|
"metadata": {
|
||
|
|
"kernelspec": {
|
||
|
|
"display_name": "litellm",
|
||
|
|
"language": "python",
|
||
|
|
"name": "python3"
|
||
|
|
},
|
||
|
|
"language_info": {
|
||
|
|
"codemirror_mode": {
|
||
|
|
"name": "ipython",
|
||
|
|
"version": 3
|
||
|
|
},
|
||
|
|
"file_extension": ".py",
|
||
|
|
"mimetype": "text/x-python",
|
||
|
|
"name": "python",
|
||
|
|
"nbconvert_exporter": "python",
|
||
|
|
"pygments_lexer": "ipython3",
|
||
|
|
"version": "3.12.3"
|
||
|
|
}
|
||
|
|
},
|
||
|
|
"nbformat": 4,
|
||
|
|
"nbformat_minor": 2
|
||
|
|
}
|