Upload 7 files
Browse files- .streamlit/config.toml +22 -0
- Dockerfile +31 -0
- LLMInsights.py +534 -0
- pages/DocIndex.py +61 -0
- pages/InsightTrace.py +28 -0
- requirements.txt +13 -0
- test-logo.png +0 -0
.streamlit/config.toml
ADDED
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
[theme] # You have to add this line
|
2 |
+
|
3 |
+
#primaryColor = '#FF8C02' # Bright Orange
|
4 |
+
|
5 |
+
#secondaryColor = '#FF8C02' # Bright Orange
|
6 |
+
|
7 |
+
#backgroundColor = '#00325B' # Dark Blue
|
8 |
+
|
9 |
+
#secondaryBackgroundColor = '#55B2FF' # Lighter Blue
|
10 |
+
|
11 |
+
|
12 |
+
#primaryColor="#ff4b4b"
|
13 |
+
#backgroundColor="#00325B"
|
14 |
+
#secondaryBackgroundColor="#262730"
|
15 |
+
#textColor="#fafafa"
|
16 |
+
#font="monospace"
|
17 |
+
|
18 |
+
|
19 |
+
|
20 |
+
base="light"
|
21 |
+
primaryColor="#efa729"
|
22 |
+
textColor="#3a0aa6"
|
Dockerfile
ADDED
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Use the official Python 3.9 image
|
2 |
+
FROM python:3.9
|
3 |
+
|
4 |
+
# Set the working directory to /code
|
5 |
+
WORKDIR /code
|
6 |
+
|
7 |
+
# Copy the current directory contents into the container at /code
|
8 |
+
COPY ./requirements.txt /code/requirements.txt
|
9 |
+
|
10 |
+
|
11 |
+
|
12 |
+
# Set up a new user named "user" with user ID 1000
|
13 |
+
RUN useradd -m -u 1000 user
|
14 |
+
# Switch to the "user" user
|
15 |
+
USER user
|
16 |
+
# Set home to the user's home directory
|
17 |
+
ENV HOME=/home/user \
|
18 |
+
PATH=/home/user/.local/bin:$PATH
|
19 |
+
|
20 |
+
# Install requirements.txt
|
21 |
+
RUN pip install --no-cache-dir --upgrade --user -r /code/requirements.txt
|
22 |
+
|
23 |
+
# Set the working directory to the user's home directory
|
24 |
+
WORKDIR $HOME/app
|
25 |
+
|
26 |
+
EXPOSE 6060
|
27 |
+
|
28 |
+
# Copy the current directory contents into the container at $HOME/app setting the owner to the user
|
29 |
+
COPY --chown=user . $HOME/app
|
30 |
+
|
31 |
+
CMD ["streamlit", "run", "LLMInsights.py", "--server.port", "7860"]
|
LLMInsights.py
ADDED
@@ -0,0 +1,534 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import json
|
3 |
+
import pandas as pd
|
4 |
+
import time
|
5 |
+
|
6 |
+
import phoenix as px
|
7 |
+
from phoenix.trace.langchain import OpenInferenceTracer, LangChainInstrumentor
|
8 |
+
|
9 |
+
#from hallucinator import HallucinatonEvaluater
|
10 |
+
|
11 |
+
from langchain.embeddings import HuggingFaceEmbeddings #for using HugginFace models
|
12 |
+
from langchain.chains.question_answering import load_qa_chain
|
13 |
+
from langchain import HuggingFaceHub
|
14 |
+
from langchain.prompts import PromptTemplate
|
15 |
+
|
16 |
+
from langchain.chains import RetrievalQA
|
17 |
+
from langchain.callbacks import StdOutCallbackHandler
|
18 |
+
|
19 |
+
#from langchain.retrievers import KNNRetriever
|
20 |
+
from langchain.storage import LocalFileStore
|
21 |
+
from langchain.embeddings import CacheBackedEmbeddings
|
22 |
+
from langchain.vectorstores import FAISS
|
23 |
+
|
24 |
+
|
25 |
+
from langchain.document_loaders import WebBaseLoader
|
26 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
27 |
+
|
28 |
+
|
29 |
+
import numpy as np
|
30 |
+
import streamlit as st
|
31 |
+
import pandas as pd
|
32 |
+
# from sklearn import datasets
|
33 |
+
# from sklearn.ensemble import RandomForestClassifier
|
34 |
+
|
35 |
+
from PIL import Image
|
36 |
+
|
37 |
+
|
38 |
+
global trace_df
|
39 |
+
|
40 |
+
# Page config
|
41 |
+
st.set_page_config(page_title="RAG PoC", layout="wide")
|
42 |
+
st.sidebar.image(Image.open("./test-logo.png"), use_column_width=True)
|
43 |
+
|
44 |
+
@st.cache_resource
|
45 |
+
def tracer_config():
|
46 |
+
#phoenix setup
|
47 |
+
session = px.launch_app()
|
48 |
+
# If no exporter is specified, the tracer will export to the locally running Phoenix server
|
49 |
+
tracer = OpenInferenceTracer()
|
50 |
+
# If no tracer is specified, a tracer is constructed for you
|
51 |
+
LangChainInstrumentor(tracer).instrument()
|
52 |
+
time.sleep(3)
|
53 |
+
print(session.url)
|
54 |
+
|
55 |
+
tracer_config()
|
56 |
+
|
57 |
+
|
58 |
+
|
59 |
+
|
60 |
+
tab1, tab2, tab3 = st.tabs(["📈 **RAG**", "🗃 FactVsHallucinate", "🤖 **RAG Scoring** " ])
|
61 |
+
|
62 |
+
|
63 |
+
|
64 |
+
os.environ["HUGGINGFACEHUB_API_TOKEN"] = "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"
|
65 |
+
|
66 |
+
# embedding cache
|
67 |
+
#store = LocalFileStore("./cache/")
|
68 |
+
|
69 |
+
# define embedder
|
70 |
+
embedder = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")
|
71 |
+
#embedder=HuggingFaceHub(repo_id="sentence-transformers/all-mpnet-base-v2")
|
72 |
+
#embedder = CacheBackedEmbeddings.from_bytes_store(core_embeddings_model, store)
|
73 |
+
|
74 |
+
# define llm
|
75 |
+
llm=HuggingFaceHub(repo_id="google/flan-t5-xxl", model_kwargs={"temperature":1, "max_length":1000000})
|
76 |
+
#llm=HuggingFaceHub(repo_id="gpt2", model_kwargs={"temperature":1, "max_length":1000000})
|
77 |
+
handler = StdOutCallbackHandler()
|
78 |
+
|
79 |
+
# set global variable
|
80 |
+
# vectorstore = None
|
81 |
+
# retriever = None
|
82 |
+
|
83 |
+
|
84 |
+
class HallucinatePromptContext:
|
85 |
+
def __init__(self):
|
86 |
+
self.variables_list = ["query","answer","context"]
|
87 |
+
self.base_template = """In this task, you will be presented with a query, a reference text and an answer. The answer is
|
88 |
+
generated to the question based on the reference text. The answer may contain false information, you
|
89 |
+
must use the reference text to determine if the answer to the question contains false information,
|
90 |
+
if the answer is a hallucination of facts. Your objective is to determine whether the reference text
|
91 |
+
contains factual information and is not a hallucination. A 'hallucination' in this context refers to
|
92 |
+
an answer that is not based on the reference text or assumes information that is not available in
|
93 |
+
the reference text. Your response should be a single word: either "factual" or "hallucinated", and
|
94 |
+
it should not include any other text or characters. "hallucinated" indicates that the answer
|
95 |
+
provides factually inaccurate information to the query based on the reference text. "factual"
|
96 |
+
indicates that the answer to the question is correct relative to the reference text, and does not
|
97 |
+
contain made up information. Please read the query and reference text carefully before determining
|
98 |
+
your response.
|
99 |
+
|
100 |
+
# Query: {query}
|
101 |
+
# Reference text: {context}
|
102 |
+
# Answer: {answer}
|
103 |
+
Is the answer above factual or hallucinated based on the query and reference text?"""
|
104 |
+
|
105 |
+
|
106 |
+
|
107 |
+
class HallucinatonEvaluater:
|
108 |
+
def __init__(self, item):
|
109 |
+
self.question = item["question"]
|
110 |
+
self.answer = item["answer"]
|
111 |
+
#self.domain = item["domain"]
|
112 |
+
self.context = item["context"]
|
113 |
+
self.llm=HuggingFaceHub(repo_id="google/flan-t5-xxl", model_kwargs={"temperature":1, "max_length":1000000})
|
114 |
+
|
115 |
+
def get_prompt_template(self):
|
116 |
+
prompt = HallucinatePromptContext()
|
117 |
+
template = prompt.base_template
|
118 |
+
varialbles = prompt.variables_list
|
119 |
+
eval_template = PromptTemplate(input_variables=varialbles, template=template)
|
120 |
+
return eval_template
|
121 |
+
|
122 |
+
def evaluate(self):
|
123 |
+
prompt = self.get_prompt_template().format(query = self.question, answer = self.answer, context = self.context)
|
124 |
+
score = self.llm(prompt)
|
125 |
+
return score
|
126 |
+
|
127 |
+
|
128 |
+
|
129 |
+
|
130 |
+
@st.cache_resource
|
131 |
+
def initialize_vectorstore():
|
132 |
+
|
133 |
+
webpage_loader = WebBaseLoader("https://www.tredence.com/case-studies/forecasting-app-installs-for-a-large-retailer-in-the-us").load()
|
134 |
+
webpage_chunks = _text_splitter(webpage_loader)
|
135 |
+
|
136 |
+
global vectorstore
|
137 |
+
global retriever
|
138 |
+
|
139 |
+
# store embeddings in vector store
|
140 |
+
vectorstore = FAISS.from_documents(webpage_chunks, embedder)
|
141 |
+
print("vector store initialized with sample doc")
|
142 |
+
|
143 |
+
# instantiate a retriever
|
144 |
+
retriever = vectorstore.as_retriever()
|
145 |
+
st.session_state['vectorstore'] = vectorstore
|
146 |
+
st.session_state['docadd'] = 0
|
147 |
+
|
148 |
+
return retriever
|
149 |
+
|
150 |
+
|
151 |
+
def _text_splitter(doc):
|
152 |
+
text_splitter = RecursiveCharacterTextSplitter(
|
153 |
+
chunk_size=600,
|
154 |
+
chunk_overlap=50,
|
155 |
+
length_function=len,
|
156 |
+
)
|
157 |
+
return text_splitter.transform_documents(doc)
|
158 |
+
|
159 |
+
def _load_docs(path: str):
|
160 |
+
load_doc = WebBaseLoader(path).load()
|
161 |
+
doc = _text_splitter(load_doc)
|
162 |
+
return doc
|
163 |
+
|
164 |
+
|
165 |
+
|
166 |
+
|
167 |
+
|
168 |
+
|
169 |
+
def rag_response(response):
|
170 |
+
#st.markdown("""<hr style="height:10px;border:none;color:#333;background-color:#333;" /> """, unsafe_allow_html=True)
|
171 |
+
|
172 |
+
#st.markdown(".stTextInput > label {font-size:105%; font-weight:bold; color:blue;} ",unsafe_allow_html=True) #for all text-input label sections
|
173 |
+
|
174 |
+
question_title = '<h1 style="color:#33ff33;font-size:24px;">Question</h1>'
|
175 |
+
|
176 |
+
|
177 |
+
|
178 |
+
st.markdown('<h1 style="color:#100170;font-size:48px;text-align:center;">RAG Response</h1>', unsafe_allow_html=True)
|
179 |
+
st.markdown('<h1 style="color:#100170;font-size:24px;">Question</h1>', unsafe_allow_html=True)
|
180 |
+
st.text_area(label="", value=response["query"], height=30)
|
181 |
+
st.markdown('<h1 style="color:#100170;font-size:24px;">RAG Output</h1>', unsafe_allow_html=True)
|
182 |
+
st.text_area(label="", value=response["result"])
|
183 |
+
# st.markdown('<h1 style="color:#100170;font-size:24px;">Augmented knowledge</h1>', unsafe_allow_html=True)
|
184 |
+
# st.text_area(label="", value=response["source_documents"])
|
185 |
+
|
186 |
+
#st.button("Check Hallucination")
|
187 |
+
|
188 |
+
|
189 |
+
|
190 |
+
|
191 |
+
|
192 |
+
# Create extractor instance
|
193 |
+
def _create_hallucination_scenario(item):
|
194 |
+
score = HallucinatonEvaluater(item).evaluate()
|
195 |
+
return score
|
196 |
+
|
197 |
+
def hallu_eval(question: str, answer: str, context: str):
|
198 |
+
print("in hallu eval")
|
199 |
+
hallucination_score = _create_hallucination_scenario({
|
200 |
+
"question": question,
|
201 |
+
"answer": answer,
|
202 |
+
"context": context
|
203 |
+
}
|
204 |
+
)
|
205 |
+
print("got hallu score")
|
206 |
+
st.markdown('<h1 style="color:#100170;font-size:24px;">Hallucinated?</h1>', unsafe_allow_html=True)
|
207 |
+
st.text_area(label=" ", value=hallucination_score, height=30)
|
208 |
+
#return {"hallucination_score": hallucination_score}
|
209 |
+
#time.sleep(10)
|
210 |
+
|
211 |
+
|
212 |
+
def scoring_eval(question: str, answer: str, context: str):
|
213 |
+
print("in scoring eval")
|
214 |
+
score = _create_evaluation_scenario({
|
215 |
+
"question": question,
|
216 |
+
"answer": answer,
|
217 |
+
"context": context
|
218 |
+
}
|
219 |
+
)
|
220 |
+
print("got score")
|
221 |
+
st.markdown('<h1 style="color:#100170;font-size:24px;">Score</h1>', unsafe_allow_html=True)
|
222 |
+
st.text_area(label=" ", value=score, height=30)
|
223 |
+
#return {"hallucination_score": hallucination_score}
|
224 |
+
#time.sleep(10)
|
225 |
+
|
226 |
+
|
227 |
+
|
228 |
+
# if 'clicked' not in st.session_state:
|
229 |
+
# print("set state to False")
|
230 |
+
# st.session_state.clicked = False
|
231 |
+
|
232 |
+
|
233 |
+
def click_button(response):
|
234 |
+
# print("set state to True")
|
235 |
+
# st.session_state.clicked = True
|
236 |
+
|
237 |
+
hallu_eval(response["query"], response["result"], "blah blah")
|
238 |
+
|
239 |
+
|
240 |
+
class BasePromptContext:
|
241 |
+
def __init__(self):
|
242 |
+
self.variables_list = ["question","answer","context"]
|
243 |
+
self.base_template = """Please act as an impartial judge and evaluate the quality of the provided answer which attempts to answer the provided question based on a provided context.
|
244 |
+
And you'll need to submit your grading for the correctness, comprehensiveness and readability of the answer, using JSON format with the 2 items in parenthesis:
|
245 |
+
("score": [your score number for the correctness of the answer], "reasoning": [your one line step by step reasoning about the correctness of the answer])
|
246 |
+
Below is your grading rubric:
|
247 |
+
- Correctness: If the answer correctly answer the question, below are the details for different scores:
|
248 |
+
- Score 0: the answer is completely incorrect, doesn’t mention anything about the question or is completely contrary to the correct answer.
|
249 |
+
- For example, when asked “How to terminate a databricks cluster”, the answer is empty string, or content that’s completely irrelevant, or sorry I don’t know the answer.
|
250 |
+
- Score 4: the answer provides some relevance to the question and answer one aspect of the question correctly.
|
251 |
+
- Example:
|
252 |
+
- Question: How to terminate a databricks cluster
|
253 |
+
- Answer: Databricks cluster is a cloud-based computing environment that allows users to process big data and run distributed data processing tasks efficiently.
|
254 |
+
- Or answer: In the Databricks workspace, navigate to the "Clusters" tab. And then this is a hard question that I need to think more about it
|
255 |
+
- Score 7: the answer mostly answer the question but is missing or hallucinating on one critical aspect.
|
256 |
+
- Example:
|
257 |
+
- Question: How to terminate a databricks cluster”
|
258 |
+
- Answer: “In the Databricks workspace, navigate to the "Clusters" tab.
|
259 |
+
Find the cluster you want to terminate from the list of active clusters.
|
260 |
+
And then you’ll find a button to terminate all clusters at once”
|
261 |
+
- Score 10: the answer correctly answer the question and not missing any major aspect
|
262 |
+
- Example:
|
263 |
+
- Question: How to terminate a databricks cluster
|
264 |
+
- Answer: In the Databricks workspace, navigate to the "Clusters" tab.
|
265 |
+
Find the cluster you want to terminate from the list of active clusters.
|
266 |
+
Click on the down-arrow next to the cluster name to open the cluster details.
|
267 |
+
Click on the "Terminate" button. A confirmation dialog will appear. Click "Terminate" again to confirm the action.”
|
268 |
+
Provided question:
|
269 |
+
{question}
|
270 |
+
Provided answer:
|
271 |
+
{answer}
|
272 |
+
Provided context:
|
273 |
+
{context}
|
274 |
+
Please provide your grading for the correctness and explain you gave the particular grading"""
|
275 |
+
|
276 |
+
class Evaluater:
|
277 |
+
def __init__(self, item):
|
278 |
+
self.question = item["question"]
|
279 |
+
self.answer = item["answer"]
|
280 |
+
#self.domain = item["domain"]
|
281 |
+
self.context = item["context"]
|
282 |
+
self.llm=HuggingFaceHub(repo_id="google/flan-t5-xxl", model_kwargs={"temperature":1, "max_length":1000000})
|
283 |
+
|
284 |
+
def get_prompt_template(self):
|
285 |
+
prompt = BasePromptContext()
|
286 |
+
template = prompt.base_template
|
287 |
+
varialbles = prompt.variables_list
|
288 |
+
eval_template = PromptTemplate(input_variables=varialbles, template=template)
|
289 |
+
return eval_template
|
290 |
+
|
291 |
+
def evaluate(self):
|
292 |
+
prompt = self.get_prompt_template().format(question = self.question, answer = self.answer, context = self.context)
|
293 |
+
score = self.llm(prompt)
|
294 |
+
return score
|
295 |
+
|
296 |
+
# Create extractor instance
|
297 |
+
def _create_evaluation_scenario(item):
|
298 |
+
score = Evaluater(item).evaluate()
|
299 |
+
return score
|
300 |
+
|
301 |
+
|
302 |
+
# Create extractor instance
|
303 |
+
def _create_hallucination_scenario(item):
|
304 |
+
score = HallucinatonEvaluater(item).evaluate()
|
305 |
+
return score
|
306 |
+
|
307 |
+
#st.write(''' # RAG App''')
|
308 |
+
|
309 |
+
with tab1:
|
310 |
+
|
311 |
+
with st.form(" RAG with evaluation - scoring & hallucination "):
|
312 |
+
#tab1.subheader(''' # RAG App''')
|
313 |
+
initialize_vectorstore()
|
314 |
+
if st.session_state['docadd'] == 1:
|
315 |
+
retriever = st.session_state['retriever']
|
316 |
+
else:
|
317 |
+
retriever = initialize_vectorstore()
|
318 |
+
|
319 |
+
#print("lenght in tab1, ", len(vectorstore.serialize_to_bytes()))
|
320 |
+
options = ["true", "false"]
|
321 |
+
|
322 |
+
st.markdown('<h1 style="color:#100170;font-size:24px;">User Query</h1>', unsafe_allow_html=True)
|
323 |
+
|
324 |
+
question = st.text_input(label="", value="", placeholder="Type in question",label_visibility="visible", disabled=False)
|
325 |
+
#st.markdown('<h2 style="color:#3a0aa6;font-size:24px;">Evaluation</h2>', unsafe_allow_html=True)
|
326 |
+
evaluate = st.selectbox(label="***Perform Evaluation?***",options=options, index=1, placeholder="Choose an option", disabled=False, label_visibility="visible")
|
327 |
+
|
328 |
+
m = st.markdown("""
|
329 |
+
<style>
|
330 |
+
div.stButton > button:first-child {
|
331 |
+
background-color: #100170;
|
332 |
+
color:#ffffff;
|
333 |
+
}
|
334 |
+
div.stButton > button:hover {
|
335 |
+
background-color: #00ff00;
|
336 |
+
color:#ff0000;
|
337 |
+
}
|
338 |
+
</style>""", unsafe_allow_html=True)
|
339 |
+
|
340 |
+
#st.markdown("----", unsafe_allow_html=True)
|
341 |
+
columns = st.columns([2,1,2])
|
342 |
+
|
343 |
+
if columns[1].form_submit_button(" Start RAG "):
|
344 |
+
|
345 |
+
st.markdown("""<hr style="height:10px;border:none;color:#333;background-color: #100170;" /> """, unsafe_allow_html=True)
|
346 |
+
|
347 |
+
print("retrie ,", retriever)
|
348 |
+
chain = RetrievalQA.from_chain_type(
|
349 |
+
llm=llm,
|
350 |
+
retriever=retriever,
|
351 |
+
callbacks=[handler],
|
352 |
+
return_source_documents=True
|
353 |
+
)
|
354 |
+
|
355 |
+
#response = chain("how tredence brought good insight?")
|
356 |
+
response = chain(question)
|
357 |
+
print(response["result"])
|
358 |
+
|
359 |
+
|
360 |
+
rag_response(response)
|
361 |
+
#click_button(response)
|
362 |
+
|
363 |
+
|
364 |
+
time.sleep(4)
|
365 |
+
|
366 |
+
df = px.active_session().get_spans_dataframe()
|
367 |
+
#print(px.active_session())
|
368 |
+
#print(px.active_session().get_spans_dataframe())
|
369 |
+
print(df.count())
|
370 |
+
df_sorted = df.sort_values(by='end_time',ascending=False)
|
371 |
+
|
372 |
+
model_input = json.loads(df_sorted[df_sorted["name"] == "LLMChain"]["attributes.input.value"][0])
|
373 |
+
context = model_input["context"]
|
374 |
+
|
375 |
+
print(context)
|
376 |
+
|
377 |
+
if evaluate:
|
378 |
+
score = _create_evaluation_scenario({
|
379 |
+
"question": question,
|
380 |
+
"answer": response['result'],
|
381 |
+
"context": context
|
382 |
+
})
|
383 |
+
hallucination_score = _create_hallucination_scenario({
|
384 |
+
"question": question,
|
385 |
+
"answer": response['result'],
|
386 |
+
"context": context
|
387 |
+
}
|
388 |
+
)
|
389 |
+
else:
|
390 |
+
score = "Evaluation is Turned OFF"
|
391 |
+
st.markdown('<h1 style="color:#100170;font-size:24px;">Completeness Score</h1>', unsafe_allow_html=True)
|
392 |
+
st.text_area(label=" ", value=score, height=30)
|
393 |
+
st.markdown('<h1 style="color:#100170;font-size:24px;">Hallucinated?</h1>', unsafe_allow_html=True)
|
394 |
+
st.text_area(label=" ", value=hallucination_score, height=30)
|
395 |
+
st.markdown('<h1 style="color:#100170;font-size:24px;">context</h1>', unsafe_allow_html=True)
|
396 |
+
st.text_area(label="", value=context)
|
397 |
+
st.markdown('<h1 style="color:#100170;font-size:24px;">Augmented knowledge</h1>', unsafe_allow_html=True)
|
398 |
+
st.text_area(label="", value=response["source_documents"])
|
399 |
+
|
400 |
+
|
401 |
+
|
402 |
+
# if st.session_state.clicked:
|
403 |
+
|
404 |
+
# # The message and nested widget will remain on the page
|
405 |
+
# hallu_eval(response["query"], response["result"], "blah blah")
|
406 |
+
|
407 |
+
|
408 |
+
# print("in if for hallu")
|
409 |
+
|
410 |
+
|
411 |
+
|
412 |
+
with tab2:
|
413 |
+
|
414 |
+
|
415 |
+
|
416 |
+
with st.form(" LLM-aasisted evaluation of Hallucination"):
|
417 |
+
|
418 |
+
|
419 |
+
#print("lenght in tab2, ", len(vectorstore.serialize_to_bytes()))
|
420 |
+
question = st.text_input(label="**Question**", value="", label_visibility="visible", disabled=False)
|
421 |
+
answer = st.text_input(label="**answer**", value="", label_visibility="visible", disabled=False)
|
422 |
+
context = st.text_input(label="**context**", value="", label_visibility="visible", disabled=False)
|
423 |
+
|
424 |
+
|
425 |
+
if st.form_submit_button("Evaluate"):
|
426 |
+
hallu_eval(question, answer, context)
|
427 |
+
|
428 |
+
|
429 |
+
with tab3:
|
430 |
+
|
431 |
+
|
432 |
+
with st.form("RAG scoring"):
|
433 |
+
|
434 |
+
|
435 |
+
#print("lenght in tab2, ", len(vectorstore.serialize_to_bytes()))
|
436 |
+
question = st.text_input(label="**Question**", value="", label_visibility="visible", disabled=False)
|
437 |
+
answer = st.text_input(label="**answer**", value="", label_visibility="visible", disabled=False)
|
438 |
+
context = st.text_input(label="**context**", value="", label_visibility="visible", disabled=False)
|
439 |
+
|
440 |
+
|
441 |
+
if st.form_submit_button("Evaluate"):
|
442 |
+
scoring_eval(question, answer, context)
|
443 |
+
|
444 |
+
|
445 |
+
|
446 |
+
print("activ session: ", px.active_session().get_spans_dataframe())
|
447 |
+
trace_df = px.active_session().get_spans_dataframe()
|
448 |
+
|
449 |
+
st.session_state['trace_df'] = trace_df
|
450 |
+
|
451 |
+
# with tab3:
|
452 |
+
|
453 |
+
|
454 |
+
|
455 |
+
# with st.form(" trace"):
|
456 |
+
|
457 |
+
# if px.active_session():
|
458 |
+
# df0 = px.active_session().get_spans_dataframe()
|
459 |
+
# if not df0.empty:
|
460 |
+
# df= df0.fillna('')
|
461 |
+
# st.dataframe(df)
|
462 |
+
|
463 |
+
|
464 |
+
|
465 |
+
|
466 |
+
|
467 |
+
def rag():
|
468 |
+
print("in rag")
|
469 |
+
options = ["true", "false"]
|
470 |
+
question = st.text_input(label="user question", value="", label_visibility="visible", disabled=False)
|
471 |
+
evaluate = st.selectbox(label="select evaluation",options=options, index=0, placeholder="Choose an option", disabled=False, label_visibility="visible")
|
472 |
+
|
473 |
+
|
474 |
+
|
475 |
+
if st.button("do RAG"):
|
476 |
+
chain = RetrievalQA.from_chain_type(
|
477 |
+
llm=llm,
|
478 |
+
retriever=retriever,
|
479 |
+
callbacks=[handler],
|
480 |
+
return_source_documents=True
|
481 |
+
)
|
482 |
+
|
483 |
+
#response = chain("how tredence brought good insight?")
|
484 |
+
response = chain(question)
|
485 |
+
print(response["result"])
|
486 |
+
|
487 |
+
# time.sleep(4)
|
488 |
+
|
489 |
+
# df = px.active_session().get_spans_dataframe()
|
490 |
+
# print(px.active_session())
|
491 |
+
# print(px.active_session().get_spans_dataframe())
|
492 |
+
# print(df.count())
|
493 |
+
# df_sorted = df.sort_values(by='end_time',ascending=False)
|
494 |
+
|
495 |
+
# model_input = json.loads(df_sorted[df_sorted["name"] == "LLMChain"]["attributes.input.value"][0])
|
496 |
+
# context = model_input["context"]
|
497 |
+
|
498 |
+
# print(context)
|
499 |
+
|
500 |
+
# if evaluate:
|
501 |
+
# score = _create_evaluation_scenario({
|
502 |
+
# "question": question,
|
503 |
+
# "answer": response['result'],
|
504 |
+
# "context": context
|
505 |
+
# })
|
506 |
+
# else:
|
507 |
+
# score = "Evaluation is Turned OFF"
|
508 |
+
|
509 |
+
# return {"question": question, "answer": response['result'], "context": context, "score": score}
|
510 |
+
rag_response(response)
|
511 |
+
|
512 |
+
# if st.button("click me"):
|
513 |
+
# click_button(response)
|
514 |
+
|
515 |
+
click = st.button("Do you want to see more?")
|
516 |
+
if click:
|
517 |
+
st.session_state.more_stuff = True
|
518 |
+
|
519 |
+
if st.session_state.more_stuff:
|
520 |
+
click_button(response)
|
521 |
+
#st.write("Doing more optional stuff")
|
522 |
+
|
523 |
+
|
524 |
+
return(response)
|
525 |
+
|
526 |
+
|
527 |
+
a = st.markdown("""
|
528 |
+
<style>
|
529 |
+
div.stTextArea > textarea {
|
530 |
+
background-color: #0099ff;
|
531 |
+
height: 1400px;
|
532 |
+
width: 800px;
|
533 |
+
}
|
534 |
+
</style>""", unsafe_allow_html=True)
|
pages/DocIndex.py
ADDED
@@ -0,0 +1,61 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
#from langchain.retrievers import KNNRetriever
|
3 |
+
from langchain.storage import LocalFileStore
|
4 |
+
from langchain.embeddings import CacheBackedEmbeddings
|
5 |
+
from langchain.vectorstores import FAISS
|
6 |
+
#from streamapp import *
|
7 |
+
from PIL import Image
|
8 |
+
|
9 |
+
from langchain.document_loaders import WebBaseLoader
|
10 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
11 |
+
|
12 |
+
|
13 |
+
st.sidebar.image(Image.open("./test-logo.png"), use_column_width=True)
|
14 |
+
|
15 |
+
|
16 |
+
print("Loading Index Page!!")
|
17 |
+
|
18 |
+
#if 'vectorstore' in st.session_state.keys():
|
19 |
+
vectorstore = st.session_state['vectorstore']
|
20 |
+
# else:
|
21 |
+
# retriever = initialize_vectorstore()
|
22 |
+
# vectorstore = st.session_state['vectorstore']
|
23 |
+
|
24 |
+
def _text_splitter(doc):
|
25 |
+
text_splitter = RecursiveCharacterTextSplitter(
|
26 |
+
chunk_size=600,
|
27 |
+
chunk_overlap=50,
|
28 |
+
length_function=len,
|
29 |
+
)
|
30 |
+
return text_splitter.transform_documents(doc)
|
31 |
+
|
32 |
+
def _load_docs(path: str):
|
33 |
+
load_doc = WebBaseLoader(path).load()
|
34 |
+
doc = _text_splitter(load_doc)
|
35 |
+
return doc
|
36 |
+
|
37 |
+
|
38 |
+
with st.form("Index documents to Vector Store"):
|
39 |
+
|
40 |
+
file_path = st.text_input(label="Enter the web link", value="", placeholder="", label_visibility="visible", disabled=False)
|
41 |
+
print("file_path " ,file_path)
|
42 |
+
|
43 |
+
submitted = st.form_submit_button("Submit")
|
44 |
+
|
45 |
+
if submitted:
|
46 |
+
st.write("Submitted web link: " + file_path)
|
47 |
+
webpage_loader = _load_docs(file_path)
|
48 |
+
|
49 |
+
webpage_chunks = _text_splitter(webpage_loader)
|
50 |
+
|
51 |
+
# store embeddings in vector store
|
52 |
+
print("vectorstore length before addition, ", len(vectorstore.serialize_to_bytes()))
|
53 |
+
vectorstore.add_documents(webpage_chunks)
|
54 |
+
print("vectorstore length after addition, ", len(vectorstore.serialize_to_bytes()))
|
55 |
+
|
56 |
+
st.session_state['vectorstore'] = vectorstore
|
57 |
+
retriever = vectorstore.as_retriever()
|
58 |
+
st.session_state['retriever'] = retriever
|
59 |
+
st.session_state['docadd'] = 1
|
60 |
+
|
61 |
+
st.markdown('<h2 style="color:#100170;font-size:24px;">Document loaded to vector store successfully!!</h2>', unsafe_allow_html=True)
|
pages/InsightTrace.py
ADDED
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
import streamlit as st
|
3 |
+
from PIL import Image
|
4 |
+
|
5 |
+
#from .streamapp import trace_df
|
6 |
+
st.sidebar.image(Image.open("./test-logo.png"), use_column_width=True)
|
7 |
+
|
8 |
+
print("trace_df ", st.session_state['trace_df'])
|
9 |
+
|
10 |
+
trace_df = st.session_state['trace_df']
|
11 |
+
print(list(trace_df))
|
12 |
+
|
13 |
+
trace_df = trace_df.loc[:,['name', 'span_kind', 'start_time', 'end_time', 'attributes.__computed__.latency_ms', 'status_code', 'status_message', 'attributes.llm.invocation_parameters', 'attributes.llm.prompts', 'attributes.input.value', 'attributes.output.value', 'attributes.llm.prompt_template.template', 'attributes.llm.prompt_template.variables', 'attributes.llm.prompt_template.version', 'attributes.retrieval.documents']]
|
14 |
+
trace_df = trace_df.sort_values(by='start_time', ascending = False)
|
15 |
+
|
16 |
+
blankIndex=[''] * len(trace_df)
|
17 |
+
trace_df.index=blankIndex
|
18 |
+
|
19 |
+
st.dataframe(trace_df)
|
20 |
+
|
21 |
+
# if px.active_session():
|
22 |
+
# df0 = px.active_session().get_spans_dataframe()
|
23 |
+
# if not df0.empty:
|
24 |
+
# df= df0.fillna('')
|
25 |
+
# st.dataframe(df)
|
26 |
+
|
27 |
+
|
28 |
+
#'name', 'span_kind', 'start_time', 'end_time', 'status_code', 'status_message', 'attributes.llm.invocation_parameters', 'attributes.llm.prompts', 'attributes.input.value', 'attributes.output.value', 'attributes.__computed__.latency_ms', 'attributes.llm.prompt_template.template', 'attributes.llm.prompt_template.variables', 'attributes.llm.prompt_template.version', 'attributes.retrieval.documents'
|
requirements.txt
ADDED
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
fastapi==0.74.*
|
2 |
+
requests==2.27.*
|
3 |
+
uvicorn[standard]==0.17.*
|
4 |
+
sentencepiece==0.1.*
|
5 |
+
torch==1.12.*
|
6 |
+
transformers==4.*
|
7 |
+
sentence-transformers
|
8 |
+
langchain==0.0.301
|
9 |
+
arize-phoenix
|
10 |
+
huggingface_hub
|
11 |
+
faiss-cpu
|
12 |
+
bs4==0.0.1
|
13 |
+
streamlit
|
test-logo.png
ADDED