Merge pull request #40 from aws-samples/development

Development

kendra_retriever_samples/README.md

@@ -30,6 +30,15 @@ If you are using Conda
 conda env create -f environment.yml 
 ```
 
+### For Bedrock
+If you are using Bedrock, make sure that you have a boto3 client with bedrock library and you use an AWS_PROFILE that has access to bedrock.
+
+```
+wget https://xxxxx/Documentation/SDK/bedrock-python-sdk.zip
+unzip bedrock-python-sdk.zip 
+pip install *.whl
+```
+
 ## Running samples
 Before you run the sample, you need to deploy a Large Language Model (or get an API key if you using Anthropic or OPENAI). The samples in this repository have been tested on models deployed using SageMaker Jumpstart. The model id for the LLMS are specified in the table below.
 
@@ -40,6 +49,9 @@ Before you run the sample, you need to deploy a Large Language Model (or get an
 | Flan XXL | FLAN_XXL_ENDPOINT | huggingface-text2text-flan-t5-xxl | flanxxl |
 | Falcon 40B instruct | FALCON_40B_ENDPOINT | huggingface-llm-falcon-40b-instruct-bf16 | falcon40b |
 | Llama2 70B instruct | LLAMA_2_ENDPOINT | meta-textgeneration-llama-2-70b-f | llama2 |
+| Bedrock Titan | None | | bedrock_titan|
+| Bedrock Claude | None | | bedrock_claude|
+| Bedrock Claude V2 | None | | bedrock_claudev2|
 
 
 after deploying the LLM, set up environment variables for kendra id, aws_region and the endpoint name (or the API key for an external provider)
@@ -49,15 +61,17 @@ For example, for running the `kendra_chat_flan_xl.py` sample, these environment
 You can use commands as below to set the environment variables. Only set the environment variable for the provider that you are using. For example, if you are using Flan-xl only set the FLAN_XXL_ENDPOINT. There is no need to set the other Endpoints and keys. 
 
 ```bash
-export AWS_REGION="<YOUR-AWS-REGION>"
-export KENDRA_INDEX_ID="<YOUR-KENDRA-INDEX-ID>"
-export FLAN_XL_ENDPOINT="<YOUR-SAGEMAKER-ENDPOINT-FOR-FLAN-T-XL>" # only if you are using FLAN_XL
-export FLAN_XXL_ENDPOINT="<YOUR-SAGEMAKER-ENDPOINT-FOR-FLAN-T-XXL>" # only if you are using FLAN_XXL
-export FALCON_40B_ENDPOINT="<YOUR-SAGEMAKER-ENDPOINT-FOR-FALCON>" # only if you are using falcon as the endpoint
-export LLAMA_2_ENDPOINT="<YOUR-SAGEMAKER-ENDPOINT-FOR-LLAMA2>" #only if you are using llama2 as the endpoint
-
-export OPENAI_API_KEY="<YOUR-OPEN-AI-API-KEY>" # only if you are using OPENAI as the endpoint
-export ANTHROPIC_API_KEY="<YOUR-ANTHROPIC-API-KEY>" # only if you are using Anthropic as the endpoint
+export AWS_REGION=<YOUR-AWS-REGION>
+export AWS_PROFILE=<AWS Profile>
+export KENDRA_INDEX_ID=<YOUR-KENDRA-INDEX-ID>
+
+export FLAN_XL_ENDPOINT=<YOUR-SAGEMAKER-ENDPOINT-FOR-FLAN-T-XL> # only if you are using FLAN_XL
+export FLAN_XXL_ENDPOINT=<YOUR-SAGEMAKER-ENDPOINT-FOR-FLAN-T-XXL> # only if you are using FLAN_XXL
+export FALCON_40B_ENDPOINT=<YOUR-SAGEMAKER-ENDPOINT-FOR-FALCON> # only if you are using falcon as the endpoint
+export LLAMA_2_ENDPOINT=<YOUR-SAGEMAKER-ENDPOINT-FOR-LLAMA2> #only if you are using llama2 as the endpoint
+
+export OPENAI_API_KEY=<YOUR-OPEN-AI-API-KEY> # only if you are using OPENAI as the endpoint
+export ANTHROPIC_API_KEY=<YOUR-ANTHROPIC-API-KEY> # only if you are using Anthropic as the endpoint
 ```
 
 

kendra_retriever_samples/app.py

@@ -8,6 +8,11 @@
 import kendra_chat_open_ai as openai
 import kendra_chat_falcon_40b as falcon40b
 import kendra_chat_llama_2 as llama2
+import kendra_chat_bedrock_titan as bedrock_titan
+import kendra_chat_bedrock_claude as bedrock_claude
+import kendra_chat_bedrock_claudev2 as bedrock_claudev2
+
+
 
 USER_ICON = "images/user-icon.png"
 AI_ICON = "images/ai-icon.png"
@@ -21,6 +26,17 @@
  'llama2' : 'Llama 2'
 }
 
+#function to read a properties file and create environment variables
+def read_properties_file(filename):
+ import os
+ import re
+ with open(filename, 'r') as f:
+ for line in f:
+ m = re.match(r'^\s*(\w+)\s*=\s*(.*)\s*$', line)
+ if m:
+ os.environ[m.group(1)] = m.group(2)
+
+
 # Check if the user ID is already stored in the session state
 if 'user_id' in st.session_state:
  user_id = st.session_state['user_id']
@@ -51,10 +67,19 @@
  elif (sys.argv[1] == 'llama2'):
  st.session_state['llm_app'] = llama2
  st.session_state['llm_chain'] = llama2.build_chain()
+ elif (sys.argv[1] == 'bedrock_titan'):
+ st.session_state['llm_app'] = bedrock_titan
+ st.session_state['llm_chain'] = bedrock_titan.build_chain()
+ elif (sys.argv[1] == 'bedrock_claude'):
+ st.session_state['llm_app'] = bedrock_claude
+ st.session_state['llm_chain'] = bedrock_claude.build_chain()
+ elif (sys.argv[1] == 'bedrock_claudev2'):
+ st.session_state['llm_app'] = bedrock_claudev2
+ st.session_state['llm_chain'] = bedrock_claudev2.build_chain()
  else:
  raise Exception("Unsupported LLM: ", sys.argv[1])
  else:
- raise Exception("Usage: streamlit run app.py <anthropic|flanxl|flanxxl|openai>")
+ raise Exception("Usage: streamlit run app.py <anthropic|flanxl|flanxxl|openai|bedrock_titan|bedrock_claude|bedrock|claudev2>")
 
 if 'chat_history' not in st.session_state:
  st.session_state['chat_history'] = []

kendra_retriever_samples/kendra_chat_bedrock_claude.py

@@ -0,0 +1,112 @@
+# from aws_langchain.kendra import AmazonKendraRetriever #custom library
+from langchain.retrievers import AmazonKendraRetriever
+from langchain.chains import ConversationalRetrievalChain
+from langchain.prompts import PromptTemplate
+from langchain.llms.bedrock import Bedrock
+from langchain.chains.llm import LLMChain
+import sys
+import os
+
+class bcolors:
+ HEADER = '\033[95m'
+ OKBLUE = '\033[94m'
+ OKCYAN = '\033[96m'
+ OKGREEN = '\033[92m'
+ WARNING = '\033[93m'
+ FAIL = '\033[91m'
+ ENDC = '\033[0m'
+ BOLD = '\033[1m'
+ UNDERLINE = '\033[4m'
+
+MAX_HISTORY_LENGTH = 5
+
+def build_chain():
+ region = os.environ["AWS_REGION"]
+ kendra_index_id = os.environ["KENDRA_INDEX_ID"]
+ credentials_profile_name = os.environ['AWS_PROFILE']
+
+ print(credentials_profile_name)
+
+
+ llm = Bedrock(
+ credentials_profile_name=credentials_profile_name,
+ region_name = region,
+ model_kwargs={"max_tokens_to_sample":300,"temperature":1,"top_k":250,"top_p":0.999,"anthropic_version":"bedrock-2023-05-31"},
+ model_id="anthropic.claude-v1"
+ )
+
+ retriever = AmazonKendraRetriever(index_id=kendra_index_id,top_k=5,region_name=region)
+
+
+ prompt_template = """Human: This is a friendly conversation between a human and an AI. 
+ The AI is talkative and provides specific details from its context but limits it to 240 tokens.
+ If the AI does not know the answer to a question, it truthfully says it 
+ does not know.
+
+ Assistant: OK, got it, I'll be a talkative truthful AI assistant.
+
+ Human: Here are a few documents in <documents> tags:
+ <documents>
+ {context}
+ </documents>
+ Based on the above documents, provide a detailed answer for, {question} 
+ Answer "don't know" if not present in the document. 
+
+ Assistant:
+ """
+ PROMPT = PromptTemplate(
+ template=prompt_template, input_variables=["context", "question"]
+ )
+
+ condense_qa_template = """Human: 
+ Given the following conversation and a follow up question, rephrase the follow up question 
+ to be a standalone question.
+ Chat History:
+ {chat_history}
+ Follow Up Input: {question}
+ Standalone question: 
+
+ Assistant:"""
+ standalone_question_prompt = PromptTemplate.from_template(condense_qa_template)
+
+
+
+ qa = ConversationalRetrievalChain.from_llm(
+ llm=llm, 
+ retriever=retriever, 
+ condense_question_prompt=standalone_question_prompt, 
+ return_source_documents=True, 
+ combine_docs_chain_kwargs={"prompt":PROMPT},
+ verbose=True)
+
+ # qa = ConversationalRetrievalChain.from_llm(llm=llm, retriever=retriever, qa_prompt=PROMPT, return_source_documents=True)
+ return qa
+
+
+def run_chain(chain, prompt: str, history=[]):
+ return chain({"question": prompt, "chat_history": history})
+
+
+if __name__ == "__main__":
+ chat_history = []
+ qa = build_chain()
+ print(bcolors.OKBLUE + "Hello! How can I help you?" + bcolors.ENDC)
+ print(bcolors.OKCYAN + "Ask a question, start a New search: or CTRL-D to exit." + bcolors.ENDC)
+ print(">", end=" ", flush=True)
+ for query in sys.stdin:
+ if (query.strip().lower().startswith("new search:")):
+ query = query.strip().lower().replace("new search:","")
+ chat_history = []
+ elif (len(chat_history) == MAX_HISTORY_LENGTH):
+ chat_history.pop(0)
+ result = run_chain(qa, query, chat_history)
+ chat_history.append((query, result["answer"]))
+ print(bcolors.OKGREEN + result['answer'] + bcolors.ENDC)
+ if 'source_documents' in result:
+ print(bcolors.OKGREEN + 'Sources:')
+ for d in result['source_documents']:
+ print(d.metadata['source'])
+ print(bcolors.ENDC)
+ print(bcolors.OKCYAN + "Ask a question, start a New search: or CTRL-D to exit." + bcolors.ENDC)
+ print(">", end=" ", flush=True)
+ print(bcolors.OKBLUE + "Bye" + bcolors.ENDC)

kendra_retriever_samples/kendra_chat_bedrock_claudev2.py

@@ -0,0 +1,112 @@
+# from aws_langchain.kendra import AmazonKendraRetriever #custom library
+from langchain.retrievers import AmazonKendraRetriever
+from langchain.chains import ConversationalRetrievalChain
+from langchain.prompts import PromptTemplate
+from langchain.llms.bedrock import Bedrock
+from langchain.chains.llm import LLMChain
+import sys
+import os
+
+class bcolors:
+ HEADER = '\033[95m'
+ OKBLUE = '\033[94m'
+ OKCYAN = '\033[96m'
+ OKGREEN = '\033[92m'
+ WARNING = '\033[93m'
+ FAIL = '\033[91m'
+ ENDC = '\033[0m'
+ BOLD = '\033[1m'
+ UNDERLINE = '\033[4m'
+
+MAX_HISTORY_LENGTH = 5
+
+def build_chain():
+ region = os.environ["AWS_REGION"]
+ kendra_index_id = os.environ["KENDRA_INDEX_ID"]
+ credentials_profile_name = os.environ['AWS_PROFILE']
+
+ print(credentials_profile_name)
+
+
+ llm = Bedrock(
+ credentials_profile_name=credentials_profile_name,
+ region_name = region,
+ model_kwargs={"max_tokens_to_sample":300,"temperature":1,"top_k":250,"top_p":0.999,"anthropic_version":"bedrock-2023-05-31"},
+ model_id="anthropic.claude-v2"
+ )
+
+ retriever = AmazonKendraRetriever(index_id=kendra_index_id,top_k=5,region_name=region)
+
+
+ prompt_template = """Human: This is a friendly conversation between a human and an AI. 
+ The AI is talkative and provides specific details from its context but limits it to 240 tokens.
+ If the AI does not know the answer to a question, it truthfully says it 
+ does not know.
+
+ Assistant: OK, got it, I'll be a talkative truthful AI assistant.
+
+ Human: Here are a few documents in <documents> tags:
+ <documents>
+ {context}
+ </documents>
+ Based on the above documents, provide a detailed answer for, {question} 
+ Answer "don't know" if not present in the document. 
+
+ Assistant:
+ """
+ PROMPT = PromptTemplate(
+ template=prompt_template, input_variables=["context", "question"]
+ )
+
+ condense_qa_template = """Human: 
+ Given the following conversation and a follow up question, rephrase the follow up question 
+ to be a standalone question.
+ Chat History:
+ {chat_history}
+ Follow Up Input: {question}
+ Standalone question: 
+
+ Assistant:"""
+ standalone_question_prompt = PromptTemplate.from_template(condense_qa_template)
+
+
+
+ qa = ConversationalRetrievalChain.from_llm(
+ llm=llm, 
+ retriever=retriever, 
+ condense_question_prompt=standalone_question_prompt, 
+ return_source_documents=True, 
+ combine_docs_chain_kwargs={"prompt":PROMPT},
+ verbose=True)
+
+ # qa = ConversationalRetrievalChain.from_llm(llm=llm, retriever=retriever, qa_prompt=PROMPT, return_source_documents=True)
+ return qa
+
+
+def run_chain(chain, prompt: str, history=[]):
+ return chain({"question": prompt, "chat_history": history})
+
+
+if __name__ == "__main__":
+ chat_history = []
+ qa = build_chain()
+ print(bcolors.OKBLUE + "Hello! How can I help you?" + bcolors.ENDC)
+ print(bcolors.OKCYAN + "Ask a question, start a New search: or CTRL-D to exit." + bcolors.ENDC)
+ print(">", end=" ", flush=True)
+ for query in sys.stdin:
+ if (query.strip().lower().startswith("new search:")):
+ query = query.strip().lower().replace("new search:","")
+ chat_history = []
+ elif (len(chat_history) == MAX_HISTORY_LENGTH):
+ chat_history.pop(0)
+ result = run_chain(qa, query, chat_history)
+ chat_history.append((query, result["answer"]))
+ print(bcolors.OKGREEN + result['answer'] + bcolors.ENDC)
+ if 'source_documents' in result:
+ print(bcolors.OKGREEN + 'Sources:')
+ for d in result['source_documents']:
+ print(d.metadata['source'])
+ print(bcolors.ENDC)
+ print(bcolors.OKCYAN + "Ask a question, start a New search: or CTRL-D to exit." + bcolors.ENDC)
+ print(">", end=" ", flush=True)
+ print(bcolors.OKBLUE + "Bye" + bcolors.ENDC)