Spaces:
Runtime error
Runtime error
File size: 13,827 Bytes
440deef |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 |
import os
from langchain import FAISS
from langchain.chains import ConversationalRetrievalChain
from langchain.chat_models import ChatOpenAI
from langchain.document_loaders import CSVLoader
from langchain.memory import ConversationBufferMemory
from langchain.prompts import ChatPromptTemplate, HumanMessagePromptTemplate, SystemMessagePromptTemplate
from langchain.text_splitter import RecursiveCharacterTextSplitter
def search_index_from_docs(source_chunks, embeddings):
# print("source chunks: " + str(len(source_chunks)))
# print("embeddings: " + str(embeddings))
search_index = FAISS.from_documents(source_chunks, embeddings)
return search_index
def get_chat_history(inputs) -> str:
res = []
for human, ai in inputs:
res.append(f"Human:{human}\nAI:{ai}")
return "\n".join(res)
class GraderQA:
def __init__(self, grader, embeddings):
self.grader = grader
self.llm = self.grader.llm
self.index_file = "vector_stores/canvas-discussions.faiss"
self.pickle_file = "vector_stores/canvas-discussions.pkl"
self.rubric_text = grader.rubric_text
self.search_index = self.get_search_index(embeddings)
self.chain = self.create_chain(embeddings)
self.tokens = None
self.question = None
def get_search_index(self, embeddings):
if os.path.isfile(self.pickle_file) and os.path.isfile(self.index_file) and os.path.getsize(
self.pickle_file) > 0:
# Load index from pickle file
search_index = self.load_index(embeddings)
else:
search_index = self.create_index(embeddings)
print("Created index")
return search_index
def load_index(self, embeddings):
# Load index
db = FAISS.load_local(
folder_path="vector_stores/",
index_name="canvas-discussions", embeddings=embeddings,
)
print("Loaded index")
return db
def create_index(self, embeddings):
source_chunks = self.create_chunk_documents()
search_index = search_index_from_docs(source_chunks, embeddings)
FAISS.save_local(search_index, folder_path="vector_stores/", index_name="canvas-discussions")
return search_index
def create_chunk_documents(self):
sources = self.fetch_data_for_embeddings()
splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=0)
source_chunks = splitter.split_documents(sources)
print("chunks: " + str(len(source_chunks)))
print("sources: " + str(len(sources)))
return source_chunks
def fetch_data_for_embeddings(self):
document_list = self.get_csv_files()
print("document list: " + str(len(document_list)))
return document_list
def get_csv_files(self):
loader = CSVLoader(file_path=self.grader.csv, source_column="student_name")
document_list = loader.load()
return document_list
def create_chain(self, embeddings):
if not self.search_index:
self.search_index = self.load_index(embeddings)
question_prompt, combine_prompt = self.create_map_reduce_prompt()
# create agent, 1 chain for summary based question, 2nd chain for semantic retrieval based question
chain = ConversationalRetrievalChain.from_llm(llm=self.llm, chain_type='map_reduce',
retriever=self.search_index.as_retriever(search_type='mmr',
search_kwargs={
'lambda_mult': 1,
'fetch_k': 50,
'k': 30}),
return_source_documents=True,
verbose=True,
memory=ConversationBufferMemory(memory_key='chat_history',
return_messages=True,
output_key='answer'),
condense_question_llm=ChatOpenAI(temperature=0,
model='gpt-3.5-turbo'),
combine_docs_chain_kwargs={"question_prompt": question_prompt,
"combine_prompt": combine_prompt})
return chain
def create_map_reduce_prompt(self):
system_template = f"""Use the following portion of a long grading results document to answer the question BUT ONLY FOR THE STUDENT MENTIONED. Use the following examples to take guidance on how to answer the question.
Examples:
Question: How many students participated in the discussion?
Answer: This student participated in the discussion./This student did not participate in the discussion.
Question: What was the average score for the discussion?
Answer: This student received a score of 10/10 for the discussion.
Question: How many students received a full score?/How many students did not receive a full score?
Answer: This student received a full score./This student did not receive a full score.
Question: How many students lost marks in X category of the rubric?
Answer: This student lost marks in X category of the rubric./This student did not lose marks in X category of the rubric.
Question: Give me 3 best responses received for the discussion.
Answer: This student gave the following responses for the discussion and received a score of 10/10.
______________________
Grading Result For:
{{context}}
______________________
Following are the instructions and rubric of the discussion post for reference, used to grade the discussion.
----------------
Instructions and Rubric:
{self.rubric_text}
"""
messages = [
SystemMessagePromptTemplate.from_template(system_template),
HumanMessagePromptTemplate.from_template("{question}"),
]
CHAT_QUESTION_PROMPT = ChatPromptTemplate.from_messages(messages)
system_template = """You are Canvas Discussions Grading + Feedback QA Bot. Have a conversation with a human, answering the questions about the grading results, feedback, answers as accurately as possible.
Use the following answers for each student to answer the users question as accurately as possible.
You are an expert at basic calculations and answering questions on grading results and can answer the following questions with ease.
If you don't know the answer, just say that you don't know. Don't try to make up an answer.
______________________
{summaries}"""
messages = [
SystemMessagePromptTemplate.from_template(system_template),
HumanMessagePromptTemplate.from_template("{question}"),
]
CHAT_COMBINE_PROMPT = ChatPromptTemplate.from_messages(messages)
return CHAT_QUESTION_PROMPT, CHAT_COMBINE_PROMPT
def create_prompt(self):
system_template = f"""You are Canvas Discussions Grading + Feedback QA Bot. Have a conversation with a human, answering the questions about the grading results, feedback, answers as accurately as possible.
You are a grading assistant who graded the canvas discussions to create the following grading results and feedback.
Use the following instruction, rubric of the discussion which were used to grade the discussions and refine the answer if needed.
----------------
{self.rubric_text}
----------------
Use the following pieces of the grading results, score, feedback and summary of student responses to answer the users question as accurately as possible.
{{context}}"""
messages = [
SystemMessagePromptTemplate.from_template(system_template),
HumanMessagePromptTemplate.from_template("{question}"),
]
return ChatPromptTemplate.from_messages(messages)
def get_tokens(self):
total_tokens = 0
for doc in self.docs:
chat_prompt = self.prompt.format(context=doc, question=self.question)
num_tokens = self.llm.get_num_tokens(chat_prompt)
total_tokens += num_tokens
# summary = self.llm(summary_prompt)
# print (f"Summary: {summary.strip()}")
# print ("\n")
return total_tokens
def run_qa_chain(self, question):
self.question = question
self.get_tokens()
answer = self.chain(question)
return answer
# system_template = """You are Canvas Discussions Grading + Feedback QA Bot. Have a conversation with a human, answering the following questions as best you can.
# You are a grading assistant who graded the canvas discussions to create the following grading results and feedback. Use the following pieces of the grading results and feedback to answer the users question.
# Use the following pieces of context to answer the users question.
# ----------------
# {context}"""
#
# messages = [
# SystemMessagePromptTemplate.from_template(system_template),
# HumanMessagePromptTemplate.from_template("{question}"),
# ]
# CHAT_PROMPT = ChatPromptTemplate.from_messages(messages)
#
#
# def get_search_index(embeddings):
# global vectorstore_index
# if os.path.isfile(pickle_file) and os.path.isfile(index_file) and os.path.getsize(pickle_file) > 0:
# # Load index from pickle file
# search_index = load_index(embeddings)
# else:
# search_index = create_index(model)
# print("Created index")
#
# vectorstore_index = search_index
# return search_index
#
#
# def create_index(embeddings):
# source_chunks = create_chunk_documents()
# search_index = search_index_from_docs(source_chunks, embeddings)
# # search_index.persist()
# FAISS.save_local(search_index, folder_path="vector_stores/", index_name="canvas-discussions")
# # Save index to pickle file
# # with open(pickle_file, "wb") as f:
# # pickle.dump(search_index, f)
# return search_index
#
#
# def search_index_from_docs(source_chunks, embeddings):
# # print("source chunks: " + str(len(source_chunks)))
# # print("embeddings: " + str(embeddings))
# search_index = FAISS.from_documents(source_chunks, embeddings)
# return search_index
#
#
# def get_html_files():
# loader = DirectoryLoader('docs', glob="**/*.html", loader_cls=UnstructuredHTMLLoader, recursive=True)
# document_list = loader.load()
# for document in document_list:
# document.metadata["name"] = document.metadata["source"].split("/")[-1].split(".")[0]
# return document_list
#
#
# def get_text_files():
# loader = DirectoryLoader('docs', glob="**/*.txt", loader_cls=TextLoader, recursive=True)
# document_list = loader.load()
# return document_list
#
#
# def create_chunk_documents():
# sources = fetch_data_for_embeddings()
#
# splitter = RecursiveCharacterTextSplitter.from_language(
# language=Language.HTML, chunk_size=500, chunk_overlap=0
# )
#
# source_chunks = splitter.split_documents(sources)
#
# print("chunks: " + str(len(source_chunks)))
# print("sources: " + str(len(sources)))
#
# return source_chunks
#
#
# def create_chain(question, llm, embeddings):
# db = load_index(embeddings)
#
# # Create chain
# chain = ConversationalRetrievalChain.from_llm(llm, db.as_retriever(search_type='mmr',
# search_kwargs={'lambda_mult': 1, 'fetch_k': 50,
# 'k': 30}),
# return_source_documents=True,
# verbose=True,
# memory=ConversationSummaryBufferMemory(memory_key='chat_history',
# llm=llm, max_token_limit=40,
# return_messages=True,
# output_key='answer'),
# get_chat_history=get_chat_history,
# combine_docs_chain_kwargs={"prompt": CHAT_PROMPT})
#
# result = chain({"question": question})
#
# sources = []
# print(result)
#
# for document in result['source_documents']:
# sources.append("\n" + str(document.metadata))
# print(sources)
#
# source = ',\n'.join(set(sources))
# return result['answer'] + '\nSOURCES: ' + source
#
#
# def load_index(embeddings):
# # Load index
# db = FAISS.load_local(
# folder_path="vector_stores/",
# index_name="canvas-discussions", embeddings=embeddings,
# )
# return db
#
#
# def get_chat_history(inputs) -> str:
# res = []
# for human, ai in inputs:
# res.append(f"Human:{human}\nAI:{ai}")
# return "\n".join(res)
|