Llama index agentic rag

bootstraprag/cli.py

@@ -34,6 +34,7 @@ def create(project_name, framework, template, observability):
             'rag-with-hyde',
             'rag-with-flare',
             'rag-with-self-correction',
+            'rag-with-controllable-agents',
             'llama-deploy-with-simplemq',
             'llama-deploy-with-rabbitmq',
             'llama-deploy-with-kafka'

bootstraprag/templates/llamaindex/rag_with_controllable_agents/.env

@@ -0,0 +1,29 @@
+DB_URL='http://localhost:6333'
+DB_API_KEY='th3s3cr3tk3y'
+COLLECTION_NAME='CONTROLLABLE_AGENTS_COLLECTION'
+
+OPENAI_API_KEY='sk-'
+OPENAI_EMBED_MODEL='text-embedding-3-small'
+# both of these models can be same as well.
+OPENAI_AGENT_MODEL='gpt-4-turbo-2024-04-09'
+OPENAI_QUERY_MODEL='gpt-4o'
+
+# use this incase you are prefering to experiment with local models.
+OLLAMA_BASE_URL='http://localhost:11434'
+OLLAMA_LLM_MODEL='llama3.1'
+OLLAMA_EMBED_MODEL='nomic-embed-text:latest'
+
+# logger can be controlled usiing env
+CRITICAL = 50
+FATAL = 50
+ERROR = 40
+WARNING = 30
+WARN = 30
+INFO = 20
+DEBUG = 10
+NOTSET = 0
+
+LIT_SERVER_PORT=8000
+LIT_SERVER_WORKERS_PER_DEVICE=4
+
+IS_EVALUATION_NEEDED=true

bootstraprag/templates/llamaindex/rag_with_controllable_agents/Dockerfile

@@ -0,0 +1,24 @@
+# Use the official Python image from the Docker Hub
+FROM python:3.9-slim
+
+# Set the working directory in the container
+WORKDIR /app
+
+# Copy the requirements file to the container
+COPY requirements.txt .
+
+# Install the required dependencies
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copy the current directory contents into the container at /app
+COPY . .
+
+# Set environment variables (you can replace these with values from your .env file or other configs)
+ENV DB_URL='http://host.docker.internal:6333' \
+    OLLAMA_BASE_URL='http://host.docker.internal:11434'
+
+# Expose port 8000 for external access
+EXPOSE 8000
+
+# Command to run your application
+CMD ["python", "api_server.py"]

bootstraprag/templates/llamaindex/rag_with_controllable_agents/client.py

@@ -0,0 +1,17 @@
+# Copyright The Lightning AI team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import requests
+
+response = requests.post("http://127.0.0.1:8000/predict", json={"input": 4.0})
+print(f"Status: {response.status_code}\nResponse:\n {response.text}")

bootstraprag/templates/llamaindex/rag_with_controllable_agents/controllable_agents_with_human_in_the_loop.py

@@ -0,0 +1,114 @@
+import os
+from llama_index.core import (
+    SimpleDirectoryReader,
+    VectorStoreIndex,
+    StorageContext,
+    Settings,
+)
+from llama_index.core.node_parser import SentenceSplitter
+from llama_index.llms.openai import OpenAI
+from llama_index.core.tools import QueryEngineTool, ToolMetadata
+from llama_index.core.agent import ReActAgent
+from llama_index.vector_stores.qdrant import QdrantVectorStore
+from rag_evaluator import RAGEvaluator
+from dotenv import load_dotenv, find_dotenv
+import qdrant_client
+import logging
+
+
+class ControllableAgentsWithHumanInLoop:
+    def __init__(self, input_dir: str, show_progress: bool = True, required_exts: list[str] = ['.pdf', '.txt'],
+                 similarity_top_k: int = 3, chunk_size: int = 512, chunk_overlap: int = 200, max_iterations: int = 20):
+        load_dotenv(find_dotenv())
+
+        logging.basicConfig(level=logging.INFO)
+        self.logger = logging.getLogger(__name__)
+
+        llm = OpenAI(model=os.environ.get('OPENAI_QUERY_MODEL'))
+
+        Settings.chunk_size = chunk_size
+        Settings.chunk_overlap = chunk_overlap
+        Settings.llm = llm
+
+        self.rag_evaluator = RAGEvaluator()
+
+        self.similarity_top_k = similarity_top_k
+
+        self.text_parser = SentenceSplitter(chunk_size=Settings.chunk_size, chunk_overlap=Settings.chunk_overlap)
+
+        self.client = qdrant_client.QdrantClient(url=os.environ['DB_URL'], api_key=os.environ['DB_API_KEY'])
+        self.vector_store = QdrantVectorStore(client=self.client, collection_name=os.environ['COLLECTION_NAME'])
+        self.vector_index = None
+
+        self.mlops_data = SimpleDirectoryReader(input_dir=input_dir, required_exts=required_exts).load_data(
+            show_progress=show_progress)
+
+        self.mlops_tool = self.get_tool("mlops_tool", "MLOps Tool", documents=self.mlops_data)
+
+        self.query_engine_tools = [self.mlops_tool]
+
+        agent_llm = OpenAI(model=os.environ.get('OPENAI_AGENT_MODEL'))
+        self.agent = ReActAgent.from_tools(
+            self.query_engine_tools, llm=agent_llm, verbose=True, max_iterations=max_iterations
+        )
+
+    def get_tool(self, name, full_name, documents=None):
+        self.logger.info("initializing the storage context")
+        storage_context = StorageContext.from_defaults(vector_store=self.vector_store)
+
+        if not self.client.collection_exists(collection_name=os.environ.get('COLLECTION_NAME')):
+            self.logger.info("indexing the nodes in VectorStoreIndex")
+            self.vector_index = VectorStoreIndex.from_documents(
+                documents=documents,
+                storage_context=storage_context,
+                transformations=Settings.transformations,
+            )
+        else:
+            self.vector_index = VectorStoreIndex.from_vector_store(vector_store=self.vector_store)
+
+        query_engine = self.vector_index.as_query_engine(similarity_top_k=self.similarity_top_k, llm=Settings.llm)
+        query_engine_tool = QueryEngineTool(
+            query_engine=query_engine,
+            metadata=ToolMetadata(
+                name=name,
+                description=(
+                    "Provides information about mlops and its details"
+                    f" {full_name}"
+                ),
+            ),
+        )
+        return query_engine_tool
+
+    def chat_repl(self, user_query: str, exit_when_done: bool = True):
+        task_message = user_query
+
+        task = self.agent.create_task(task_message)
+
+        response = None
+        step_output = None
+        message = None
+        while message != "exit":
+            if message is None or message == "":
+                step_output = self.agent.run_step(task.task_id)
+            else:
+                step_output = self.agent.run_step(task.task_id, input=message)
+            if exit_when_done and step_output.is_last:
+                print(">> Task marked as finished by the agent, executing task execution.")
+                break
+
+            message = input(">> Add feedback during step? (press enter/leave blank to continue, "
+                            "and type 'exit' to stop): ")
+            if message == "exit":
+                break
+
+        if step_output is None:
+            print(">> You haven't run the agent. Task is discarded.")
+        elif not step_output.is_last:
+            print(">> The agent hasn't finished yet. Task is discarded.")
+        else:
+            response = self.agent.finalize_response(task.task_id)
+        print(f"Agent: {str(response)}")
+
+        if os.environ.get('IS_EVALUATION_NEEDED') == 'true':
+            self.rag_evaluator.evaluate(user_query=user_query, response_obj=response)
+        return response

bootstraprag/templates/llamaindex/rag_with_controllable_agents/main.py

@@ -0,0 +1,17 @@
+# driver code
+from controllable_agents_with_human_in_the_loop import ControllableAgentsWithHumanInLoop
+
+
+controllable_agent = ControllableAgentsWithHumanInLoop(input_dir='data', show_progress=True)
+
+# Start a loop to continually get input from the user
+while True:
+    # Get a query from the user
+    user_query = input("Enter your query [type 'bye' to 'exit']: ")
+
+    # Check if the user wants to terminate the loop
+    if user_query.lower() == "bye" or user_query.lower() == "exit":
+        break
+
+    response = controllable_agent.chat_repl(user_query=user_query)
+    print(response)

bootstraprag/templates/llamaindex/rag_with_controllable_agents/rag_evaluator.py

@@ -0,0 +1,34 @@
+from deepeval.integrations.llama_index import (
+    DeepEvalFaithfulnessEvaluator,
+    DeepEvalAnswerRelevancyEvaluator,
+    DeepEvalContextualRelevancyEvaluator
+)
+from dotenv import load_dotenv, find_dotenv
+from typing import Any
+import os
+import logging
+
+_ = load_dotenv(find_dotenv())
+logging.basicConfig(level=int(os.environ['INFO']))
+logger = logging.getLogger(__name__)
+
+
+class RAGEvaluator:
+    def __init__(self):
+        self.faithfulness_evaluator = DeepEvalFaithfulnessEvaluator()
+        self.answer_relevancy_evaluator = DeepEvalAnswerRelevancyEvaluator()
+        self.context_relevancy_evaluator = DeepEvalContextualRelevancyEvaluator()
+
+    def evaluate(self, user_query: str, response_obj: Any):
+        logger.info(f"calling evaluation, user_query: {user_query}, response_obj: {response_obj}")
+        retrieval_context = [node.get_content() for node in response_obj.source_nodes]
+        actual_output = response_obj.response
+        faithfulness_evaluation_response = self.faithfulness_evaluator.evaluate(query=user_query, response=actual_output,
+                                                                                contexts=retrieval_context)
+        answer_relevancy_response = self.answer_relevancy_evaluator.evaluate(query=user_query, response=actual_output,
+                                                                             contexts=retrieval_context)
+        context_relevancy_response = self.context_relevancy_evaluator.evaluate(query=user_query, response=actual_output,
+                                                                               contexts=retrieval_context)
+        logger.info(f"faithfulness_evaluation_response: {faithfulness_evaluation_response.score}")
+        logger.info(f"answer_relevancy_response: {answer_relevancy_response.score}")
+        logger.info(f"context_relevancy_response: {context_relevancy_response.score}")

bootstraprag/templates/llamaindex/rag_with_controllable_agents/readme.md

@@ -0,0 +1,10 @@
+## Instructions to run the code
+
+- Navigate to the root of the project and run the below command
+- `pip install -r requirements.txt`
+- open `.env` file update your qdrant password in the property `DB_API_KEY`
+- In the data folder place your data preferably any ".pdf"
+#### Note: ensure your qdrant and ollama (if LLM models are pointing to local) are running
+- run `python main.py`
+
+Note: This is Human in the loop agent, so keep a watch on the console to pass the human feedback to the agent.

bootstraprag/templates/llamaindex/rag_with_controllable_agents/requirements.txt

@@ -0,0 +1,10 @@
+python-dotenv==1.0.1
+llama-index==0.11.16
+llama-index-llms-openai==0.2.11
+llama-index-llms-ollama==0.3.3
+llama-index-embeddings-openai==0.2.5
+llama-index-embeddings-ollama==0.3.1
+llama-index-vector-stores-qdrant==0.3.0
+qdrant-client==1.11.3
+pydantic==2.9.2
+deepeval==1.3.4

bootstraprag/templates/llamaindex/rag_with_controllable_agents_with_observability/.env

@@ -0,0 +1,29 @@
+DB_URL='http://localhost:6333'
+DB_API_KEY='th3s3cr3tk3y'
+COLLECTION_NAME='CONTROLLABLE_AGENTS_COLLECTION'
+
+OPENAI_API_KEY='sk-'
+OPENAI_EMBED_MODEL='text-embedding-3-small'
+# both of these models can be same as well.
+OPENAI_AGENT_MODEL='gpt-4-turbo-2024-04-09'
+OPENAI_QUERY_MODEL='gpt-4o'
+
+# use this incase you are prefering to experiment with local models.
+OLLAMA_BASE_URL='http://localhost:11434'
+OLLAMA_LLM_MODEL='llama3.1'
+OLLAMA_EMBED_MODEL='nomic-embed-text:latest'
+
+# logger can be controlled usiing env
+CRITICAL = 50
+FATAL = 50
+ERROR = 40
+WARNING = 30
+WARN = 30
+INFO = 20
+DEBUG = 10
+NOTSET = 0
+
+LIT_SERVER_PORT=8000
+LIT_SERVER_WORKERS_PER_DEVICE=4
+
+IS_EVALUATION_NEEDED=true

bootstraprag/templates/llamaindex/rag_with_controllable_agents_with_observability/Dockerfile

@@ -0,0 +1,24 @@
+# Use the official Python image from the Docker Hub
+FROM python:3.9-slim
+
+# Set the working directory in the container
+WORKDIR /app
+
+# Copy the requirements file to the container
+COPY requirements.txt .
+
+# Install the required dependencies
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copy the current directory contents into the container at /app
+COPY . .
+
+# Set environment variables (you can replace these with values from your .env file or other configs)
+ENV DB_URL='http://host.docker.internal:6333' \
+    OLLAMA_BASE_URL='http://host.docker.internal:11434'
+
+# Expose port 8000 for external access
+EXPOSE 8000
+
+# Command to run your application
+CMD ["python", "api_server.py"]

bootstraprag/templates/llamaindex/rag_with_controllable_agents_with_observability/client.py

@@ -0,0 +1,17 @@
+# Copyright The Lightning AI team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import requests
+
+response = requests.post("http://127.0.0.1:8000/predict", json={"input": 4.0})
+print(f"Status: {response.status_code}\nResponse:\n {response.text}")