File size: 5,761 Bytes
5b45d10 297437a e760939 5b45d10 297437a e93a46f fa0faa4 e93a46f fa0faa4 e760939 fa0faa4 e760939 88f7b26 dc99d7c e93a46f e760939 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 |
import gradio as gr
import requests
import os
hf_token = os.getenv("HuggingFace_Token")
vapi_url = "https://api-inference.huggingface.co/models/vectara/hallucination_evaluation_model"
headers = {"Authorization": f"Bearer {hf_token}"}
# Function to query the API
def query(payload):
response = requests.post(vapi_url, headers=headers, json=payload)
return response.json()
def check_hallucination(assertion, citation):
api_url = "https://api-inference.huggingface.co/models/vectara/hallucination_evaluation_model"
header = {"Authorization": f"Bearer {hf_token}"}
payload = {"inputs": f"{assertion} [SEP] {citation}"}
response = requests.post(api_url, headers=header, json=payload, timeout=120)
output = response.json()
output = output[0][0]["score"]
return f"**hallucination score:** {output}"
def query_vectara(text):
user_message = text
customer_id = os.getenv('CUSTOMER_ID')
corpus_id = os.getenv('CORPUS_ID')
api_key = os.getenv('API_KEY')
api_key_header = {
"customer-id": customer_id,
"x-api-key": api_key
}
request_body = {
"query": [
{
"query": user_message,
"queryContext": "",
"start": 1,
"numResults": 25,
"contextConfig": {
"charsBefore": 0,
"charsAfter": 0,
"sentencesBefore": 2,
"sentencesAfter": 2,
"startTag": "%START_SNIPPET%",
"endTag": "%END_SNIPPET%",
},
"rerankingConfig": {
"rerankerId": 272725718,
"mmrConfig": {
"diversityBias": 0.35
}
},
"corpusKey": [
{
"customerId": customer_id,
"corpusId": corpus_id,
"semantics": 0,
"metadataFilter": "",
"lexicalInterpolationConfig": {
"lambda": 0
},
"dim": []
}
],
"summary": [
{
"maxSummarizedResults": 5,
"responseLang": "auto",
"summarizerPromptName": "vectara-summary-ext-v1.2.0"
}
]
}
]
}
response = requests.post(
"https://api.vectara.io/v1/query",
json=request_body,
verify=True,
headers=api_key_header
)
if response.status_code == 200:
query_data = response.json()
if query_data:
sources_info = []
# Extract the summary.
summary = query_data['responseSet'][0]['summary'][0]['text']
# Iterate over all response sets
for response_set in query_data.get('responseSet', []):
# Extract sources
# Limit to top 5 sources.
for source in response_set.get('response', [])[:5]:
source_metadata = source.get('metadata', [])
source_info = {}
for metadata in source_metadata:
metadata_name = metadata.get('name', '')
metadata_value = metadata.get('value', '')
if metadata_name == 'title':
source_info['title'] = metadata_value
elif metadata_name == 'author':
source_info['author'] = metadata_value
elif metadata_name == 'pageNumber':
source_info['page number'] = metadata_value
if source_info:
sources_info.append(source_info)
result = {"summary": summary, "sources": sources_info}
return f"{json.dumps(result, indent=2)}"
else:
return "No data found in the response."
else:
return f"Error: {response.status_code}"
# Create the Gradio interface
iface = gr.Interface(
fn=evaluate_hallucination,
inputs=[gr.Textbox(label="Assertion"), gr.Textbox(label="Citation")],
outputs=[gr.Label(), gr.Textbox(label="Explanation")],
live=False,
title="👋🏻Welcome to 🌟Team Tonic's 🧠🌈SureRAG🔴🟢",
description="Nothing is more important than reputation. However you can create automated content pipelines for public facing content. How can businesses grow their reputation while mitigating risks due to AI? How it works : vectara rag retrieval reranking and summarization is used to return content. then an LLM generates content based on these returns. this content is checked for hallucination before being validated for publishing on twitter. SureRAG is fixed on Tonic-AI's README files as a Demo, provide input to generate a response. This response is checked by Vectara's HHME. Check out the model [vectara/hallucination_evaluation_model](https://huggingface.co/vectara/hallucination_evaluation_model) Join us : 🌟TeamTonic🌟 is always making cool demos! Join our active builder's🛠️community 👻 [](https://discord.gg/GWpVpekp) On 🤗Huggingface: [TeamTonic](https://huggingface.co/TeamTonic) & [MultiTransformer](https://huggingface.co/MultiTransformer) On 🌐Github: [Tonic-AI](https://github.com/tonic-ai) & contribute to 🌟 [DataTonic](https://github.com/Tonic-AI/DataTonic)",
)
# Launch the interface
iface.launch() |