Added logging and generated response

.gitignore

@@ -0,0 +1,2 @@
+venv/
+__pycache__/

data/load_dataset.py

@@ -1,4 +1,8 @@
+import logging
 from datasets import load_dataset
 
 def load_data():
-    return load_dataset("rungalileo/ragbench", 'covidqa', split="train")
+    logging.info("Loading dataset")
+    dataset = load_dataset("rungalileo/ragbench", 'covidqa', split="train")
+    logging.info("Dataset loaded successfully")
+    return dataset

generator/create_prompt.py

@@ -0,0 +1,134 @@
+from langchain.docstore.document import Document
+
+def create_prompt(documents, question, response):
+    prompt = f""" I asked someone to answer a question based on one or more documents. Your task is to review their response and assess whether or not each sentence in that response is supported by text in the documents. If supported, identify which sentences in the documents provide that support. Additionally, identify which documents contain useful information for answering the question, and which documents the answer was sourced from.
+
+    Here are the documents, each split into sentences with associated keys:
+    {documents}
+
+    The question was:
+    {question}
+
+    Here is their response, split into sentences with associated keys:
+    {response}
+
+    Provide a JSON response with an answer , relevance_explanation, all_relevant_sentence_keys, overall_supported_explanation, overall_supported, sentence_support_information(this can be a json inside a json and with these fields response_sentence_key, explanation, supporting_sentence_keys, fully_supported) and all_utilized_sentence_keys below are the definitions
+
+    You must respond with a JSON object matching this schema:
+
+    {{
+      "answer": "string",
+      "relevance_explanation": "string",
+      "all_relevant_sentence_keys": ["string"],
+      "overall_supported_explanation": "string",
+      "overall_supported": "boolean",
+      "sentence_support_information": [
+        {{
+          "response_sentence_key": "string",
+          "explanation": "string",
+          "supporting_sentence_keys": ["string"],
+          "fully_supported": "boolean"
+        }},
+      ],
+      "all_utilized_sentence_keys": ["string"]
+    }}
+
+          The relevance_explanation field is a string explaining which documents
+        contain useful information for answering the question. Provide a step-by-step
+        breakdown of information provided in the documents and how it is useful for
+        answering the question.
+        The all_relevant_sentence_keys field is a list of all document sentences keys
+        (e.g. ’0a’) that are revant to the question. Include every sentence that is
+        useful and relevant to the question, even if it was not used in the response,
+        or if only parts of the sentence are useful. Ignore the provided response when
+        making this judgement and base your judgement solely on the provided documents
+        and question. Omit sentences that, if removed from the document, would not
+        impact someone’s ability to answer the question.
+        The overall_supported_explanation field is a string explaining why the response
+        *as a whole* is or is not supported by the documents. In this field, provide a
+        step-by-step breakdown of the claims made in the response and the support (or
+        lack thereof) for those claims in the documents. Begin by assessing each claim
+        separately, one by one; don’t make any remarks about the response as a whole
+        until you have assessed all the claims in isolation.
+        The overall_supported field is a boolean indicating whether the response as a
+        whole is supported by the documents. This value should reflect the conclusion
+        you drew at the end of your step-by-step breakdown in overall_supported_explanation.
+        In the sentence_support_information field, provide information about the support
+        *for each sentence* in the response.
+        The sentence_support_information field is a list of objects, one for each sentence
+        in the response. Each object MUST have the following fields:
+        - response_sentence_key: a string identifying the sentence in the response.
+        This key is the same as the one used in the response above.
+        16
+        - explanation: a string explaining why the sentence is or is not supported by the
+        documents.
+        - supporting_sentence_keys: keys (e.g. ’0a’) of sentences from the documents that
+        support the response sentence. If the sentence is not supported, this list MUST
+        be empty. If the sentence is supported, this list MUST contain one or more keys.
+        In special cases where the sentence is supported, but not by any specific sentence,
+        you can use the string "supported_without_sentence" to indicate that the sentence
+        is generally supported by the documents. Consider cases where the sentence is
+        expressing inability to answer the question due to lack of relevant information in
+        the provided contex as "supported_without_sentence". In cases where the sentence
+        is making a general statement (e.g. outlining the steps to produce an answer, or
+        summarizing previously stated sentences, or a transition sentence), use the
+        sting "general".In cases where the sentence is correctly stating a well-known fact,
+        like a mathematical formula, use the string "well_known_fact". In cases where the
+        sentence is performing numerical reasoning (e.g. addition, multiplication), use
+        the string "numerical_reasoning".
+        - fully_supported: a boolean indicating whether the sentence is fully supported by
+        the documents.
+        - This value should reflect the conclusion you drew at the end of your step-by-step
+        breakdown in explanation.
+        - If supporting_sentence_keys is an empty list, then fully_supported must be false.
+        - Otherwise, use fully_supported to clarify whether everything in the response
+        sentence is fully supported by the document text indicated in supporting_sentence_keys
+        (fully_supported = true), or whether the sentence is only partially or incompletely
+        supported by that document text (fully_supported = false).
+        The all_utilized_sentence_keys field is a list of all sentences keys (e.g. ’0a’) that
+        were used to construct the answer. Include every sentence that either directly supported
+        the answer, or was implicitly used to construct the answer, even if it was not used
+        in its entirety. Omit sentences that were not used, and could have been removed from
+        the documents without affecting the answer.
+        You must respond with a valid JSON string. Use escapes for quotes, e.g. ‘\\"‘, and
+        newlines, e.g. ‘\\n‘. Do not write anything before or after the JSON string. Do not
+        wrap the JSON string in backticks like ‘‘‘ or ‘‘‘json.
+        As a reminder: your task is to review the response and assess which documents contain
+        useful information pertaining to the question, and how each sentence in the response
+        is supported by the text in the documents.
+
+    """
+    return prompt
+    prompt = f""" I asked someone to answer a question based on one or more documents. Your task is to review their response and assess whether or not each sentence in that response is supported by text in the documents. If supported, identify which sentences in the documents provide that support. Additionally, identify which documents contain useful information for answering the question, and which documents the answer was sourced from.
+
+    Here are the documents, each split into sentences with associated keys:
+    {documents}
+
+    The question was:
+    {question}
+
+    Here is their response, split into sentences with associated keys:
+    {response}
+
+    Provide a JSON response with an answer , relevance_explanation, all_relevant_sentence_keys, overall_supported_explanation, overall_supported, sentence_support_information(this can be a json inside a json and with these fields response_sentence_key, explanation, supporting_sentence_keys, fully_supported) and all_utilized_sentence_keys below are the definitions
+
+    You must respond with a JSON object matching this schema:
+
+    {{
+      "answer": "string",
+      "relevance_explanation": "string",
+      "all_relevant_sentence_keys": ["string"],
+      "overall_supported_explanation": "string",
+      "overall_supported": "boolean",
+      "sentence_support_information": [
+        {{
+          "response_sentence_key": "string",
+          "explanation": "string",
+          "supporting_sentence_keys": ["string"],
+          "fully_supported": "boolean"
+        }},
+      ],
+      "all_utilized_sentence_keys": ["string"]
+    }}
+"""
+    return prompt

generator/extract_attributes.py

@@ -0,0 +1,18 @@
+import json
+from generator.create_prompt import create_prompt
+from generator.initialize_llm import initialize_llm
+
+# Initialize the LLM
+llm = initialize_llm()
+
+# Function to extract attributes
+def extract_attributes(question, relevant_docs, response):
+    # Format documents into a string by accessing the `page_content` attribute of each Document
+    formatted_documents = "\n".join([f"Doc {i+1}: {doc.page_content}" for i, doc in enumerate(relevant_docs)])
+
+    attribute_prompt = create_prompt(formatted_documents, question, response)
+
+    # Instead of using BaseMessage, pass the formatted prompt directly to invoke
+    result = llm.invoke(attribute_prompt)
+
+    return result

generator/generate_response.py

@@ -1,6 +1,7 @@
 from langchain.chains import RetrievalQA
 
 def generate_response(llm, vector_store, question):
+    # Create a retrieval-based question-answering chain using the relevant documents
     qa_chain = RetrievalQA.from_chain_type(
         llm=llm,
         retriever=vector_store.as_retriever(),

main.py

@@ -1,35 +1,70 @@
+import logging, json
 from data.load_dataset import load_data
 from retriever.chunk_documents import chunk_documents
 from retriever.embed_documents import embed_documents
 from retriever.retrieve_documents import retrieve_top_k_documents
 from generator.initialize_llm import initialize_llm
 from generator.generate_response import generate_response
+from generator.extract_attributes import extract_attributes
+
+# Configure logging
+logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
 
 def main():
+    logging.info("Starting the RAG pipeline")
+
     # Load the dataset
     dataset = load_data()
-    
+    logging.info("Dataset loaded")
+
     # Chunk the dataset
     documents = chunk_documents(dataset)
-    
+    logging.info("Documents chunked")
+
     # Embed the documents
     vector_store = embed_documents(documents)
-    
+    logging.info("Documents embedded")
+
     # Initialize the LLM
     llm = initialize_llm()
-    
+    logging.info("LLM initialized")
+
     # Sample question
     sample_question = dataset[0]['question']
-    
+    logging.info(f"Sample question: {sample_question}")
+
     # Retrieve relevant documents
     relevant_docs = retrieve_top_k_documents(vector_store, sample_question, top_k=5)
-    
-    # Generate a response
+    logging.info("Relevant documents retrieved :", print(len(relevant_docs)))
+    # Log each retrieved document individually
+    #for i, doc in enumerate(relevant_docs):
+        #logging.info(f"Relevant document {i+1}: {doc}")
+
+    # Generate a response using the relevant documents
     response, source_docs = generate_response(llm, vector_store, sample_question)
-    
+    logging.info("Response generated")
+
     # Print the response
     print(f"Response: {response}")
     print(f"Source Documents: {source_docs}")
 
+     # Extract attributes from the response and source documents
+    attributes = extract_attributes(sample_question, relevant_docs, response)
+    
+    # Only proceed if the content is not empty
+    if attributes.content:
+        result_content = attributes.content  # Access the content attribute
+        
+        # Extract the JSON part from the result_content
+        json_start = result_content.find("{")
+        json_end = result_content.rfind("}") + 1
+        json_str = result_content[json_start:json_end]
+        
+        try:
+            result_json = json.loads(json_str)
+            print(json.dumps(result_json, indent=2))
+        except json.JSONDecodeError as e:
+            logging.error(f"JSONDecodeError: {e}")
+
 if __name__ == "__main__":
     main()

requirements.txt

@@ -6,5 +6,5 @@ sentence-transformers
 langchain
 llama-index
 langchain-community
-groq-sdk
-langchain_groq
+langchain_groq
+langchain-huggingface

retriever/embed_documents.py

@@ -0,0 +1,7 @@
+from langchain_huggingface import HuggingFaceEmbeddings
+from langchain_community.vectorstores import FAISS
+
+def embed_documents(documents):
+    embedding_model = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
+    vector_store = FAISS.from_texts([doc['text'] for doc in documents], embedding_model)
+    return vector_store