Merge pull request #82 from mit-submit/main

Initial Release for 801 Class Project

.github/workflows/prod-801-ci-cd.yaml

@@ -0,0 +1,70 @@
+name: Deploy A2rchi Prod for 8.01
+run-name: ${{ github.actor }} deploys A2rchi for 8.01 to prod
+on:
+  push:
+    branches:
+      - release-8.01
+jobs:
+  deploy-prod-system:
+    runs-on: ubuntu-latest
+    env:
+      SSH_AUTH_SOCK: /tmp/ssh_agent.sock
+    steps:
+      # boilerplate message and pull repository to CI runner
+      - run: echo "🎉 The job was automatically triggered by a ${{ github.event_name }} event."
+      - uses: actions/checkout@v3
+      - run: echo "The ${{ github.repository }} repository has been cloned to the runner."
+
+      # setup SSH 
+      - name: Setup SSH
+        run: |
+          mkdir -p /home/runner/.ssh/
+          echo "${{ secrets.SSH_PRIVATE_KEY_MDRUSSO }}" > /home/runner/.ssh/id_rsa_submit
+          chmod 600 /home/runner/.ssh/id_rsa_submit
+          echo "${{ secrets.SSH_SUBMIT_KNOWN_HOSTS }}" > ~/.ssh/known_hosts
+          cp ${{ github.workspace }}/deploy/ssh_config /home/runner/.ssh/config
+          ssh-agent -a $SSH_AUTH_SOCK > /dev/null
+          ssh-add /home/runner/.ssh/id_rsa_submit
+
+      # create secrets files for docker-compose
+      - name: Create Secrets Files
+        run: |
+          mkdir -p ${{ github.workspace }}/deploy/prod-801/secrets/
+          touch ${{ github.workspace }}/deploy/prod-801/secrets/flask_uploader_app_secret_key.txt
+          echo "${{ secrets.PROD_FLASK_UPLOADER_APP_SECRET_KEY }}" >> ${{ github.workspace }}/deploy/prod-801/secrets/flask_uploader_app_secret_key.txt
+          chmod 400 ${{ github.workspace }}/deploy/prod-801/secrets/flask_uploader_app_secret_key.txt
+          touch ${{ github.workspace }}/deploy/prod-801/secrets/uploader_salt.txt
+          echo "${{ secrets.PROD_UPLOADER_SALT }}" >> ${{ github.workspace }}/deploy/prod-801/secrets/uploader_salt.txt
+          chmod 400 ${{ github.workspace }}/deploy/prod-801/secrets/uploader_salt.txt
+          touch ${{ github.workspace }}/deploy/prod-801/secrets/openai_api_key.txt
+          echo "${{ secrets.OPENAI_API_KEY }}" >> ${{ github.workspace }}/deploy/prod-801/secrets/openai_api_key.txt
+          chmod 400 ${{ github.workspace }}/deploy/prod-801/secrets/openai_api_key.txt
+          touch ${{ github.workspace }}/deploy/prod-801/secrets/hf_token.txt
+          echo "${{ secrets.HF_TOKEN }}" >> ${{ github.workspace }}/deploy/prod-801/secrets/hf_token.txt
+          chmod 400 ${{ github.workspace }}/deploy/prod-801/secrets/hf_token.txt
+
+      # stop any existing docker compose that's running
+      - name: Stop Docker Compose
+        run: |
+          ssh submit-t3desk 'bash -s' < ${{ github.workspace }}/deploy/prod-801/prod-801-stop.sh
+
+      # copy repository to machine
+      - name: Copy Repository
+        run: |
+          rsync -e ssh -r ${{ github.workspace}}/* --exclude .git/ --delete submit-t3desk:~/A2rchi-prod-801/
+
+      # run deploy script
+      - name: Run Deploy Script
+        run: |
+          ssh submit-t3desk 'bash -s' < ${{ github.workspace }}/deploy/prod-801/prod-801-install.sh
+
+      # clean up secret files
+      - name: Remove Secrets from Runner
+        run: |
+          rm ${{ github.workspace }}/deploy/prod-801/secrets/flask_uploader_app_secret_key.txt
+          rm ${{ github.workspace }}/deploy/prod-801/secrets/uploader_salt.txt
+          rm ${{ github.workspace }}/deploy/prod-801/secrets/openai_api_key.txt
+          rm ${{ github.workspace }}/deploy/prod-801/secrets/hf_token.txt
+
+      # print job status
+      - run: echo "🍏 This job's status is ${{ job.status }}."

.gitignore

@@ -10,3 +10,4 @@ venv
 *.egg-info
 *sqlite_db
 .vscode
+801-content/

A2rchi/chains/prompts.py

@@ -1,27 +1,31 @@
 # flake8: noqa
 from langchain.prompts.prompt import PromptTemplate
+from A2rchi.utils.config_loader import Config_Loader
 
-condense_history_template = """Given the following conversation between you (the AI named A2rchi), a human user who needs help, and an expert, and a follow up question, rephrase the follow up question to be a standalone question, in its original language.
+config = Config_Loader().config["chains"]["prompts"]
 
-Chat History:
-{chat_history}
-Follow Up Input: {question}
-Standalone question:"""
+def read_prompt(prompt_filepath, is_condense_prompt=False, is_main_prompt=False):
+    with open(prompt_filepath, "r") as f:
+        raw_prompt = f.read()
 
-prompt_template = """You are a conversational chatbot named A2rchi who helps people navigate a computing resource named subMIT. You will be provided context to help you answer their questions. 
-Using your linux and computing knowledge, answer the question at the end. Unless otherwise indicated, assume the users are not well versed computing.
- Please do not assume that subMIT machines have anything installed on top of native linux except if the context mentions it.
-If you don't know, say "I don't know", if you need to ask a follow up question, please do.
+    prompt = ""
+    for line in raw_prompt.split("\n"):
+        if len(line.lstrip())>0 and line.lstrip()[0:1] != "#":
+            prompt += line + "\n"
 
-Context: {context} Additionally, it is always preferred to use conda, if possible.
+    if is_condense_prompt and ("{chat_history}" not in prompt or "{question}" not in prompt):
+        raise ValueError("""Condensing prompt must contain \"{chat_history}\" and \"{question}\" tags. Instead, found prompt to be:
+                         """ + prompt)
+    if is_main_prompt and ("{context}" not in prompt or "{question}" not in prompt):
+        raise ValueError("""Condensing prompt must contain \"{context}\" and \"{question}\" tags. Instead, found prompt to be:
+                         """ + prompt)
 
-Question: {question}
-Helpful Answer:"""
+    return prompt
 
 QA_PROMPT = PromptTemplate(
-    template=prompt_template, input_variables=["context", "question"]
+    template=read_prompt(config["MAIN_PROMPT"], is_main_prompt=True), input_variables=["context", "question"]
 )
 
 CONDENSE_QUESTION_PROMPT = PromptTemplate(
-    template=condense_history_template, input_variables=["chat_history", "question"]
+    template=read_prompt(config["CONDENSING_PROMPT"], is_condense_prompt=True), input_variables=["chat_history", "question"]
 )

A2rchi/interfaces/chat_app/app.py

@@ -153,7 +153,7 @@ def __init__(self, app, **configs):
         CORS(self.app)
 
         # add endpoints for flask app
-        self.add_endpoint('/get_chat_response', 'get_chat_response', self.get_chat_response, methods=["POST"])
+        self.add_endpoint('/api/get_chat_response', 'get_chat_response', self.get_chat_response, methods=["POST"])
         self.add_endpoint('/', '', self.index)
         self.add_endpoint('/terms', 'terms', self.terms)
 

A2rchi/interfaces/chat_app/static/script.js

@@ -42,7 +42,7 @@ const refreshChat = async () => {
 }
 
 const getChatResponse = async (incomingChatDiv) => {
-    const API_URL = "http://0.0.0.0:7861/get_chat_response";
+    const API_URL = "http://t3desk019.mit.edu:7861/api/get_chat_response";
     const pElement = document.createElement("div");
 
      // Define the properties and data for the API request

A2rchi/interfaces/chat_app/static/script.js-template

@@ -42,7 +42,7 @@ const refreshChat = async () => {
 }
 
 const getChatResponse = async (incomingChatDiv) => {
-    const API_URL = "http://0.0.0.0:XX-HTTP_PORT-XX/get_chat_response";
+    const API_URL = "http://t3desk019.mit.edu:XX-HTTP_PORT-XX/api/get_chat_response";
     const pElement = document.createElement("div");
 
      // Define the properties and data for the API request

A2rchi/utils/config_loader.py

@@ -16,9 +16,9 @@ def load_config(self):
         """
         Small function for loading the config.yaml file
         """
-        prod_or_dev = os.getenv("PROD_OR_DEV")
+        env = os.getenv("RUNTIME_ENV")
         try:
-            with open(f"./config/{prod_or_dev}-config.yaml", "r") as f:
+            with open(f"./config/{env}-config.yaml", "r") as f:
                 config = yaml.load(f, Loader=yaml.FullLoader)
 
             # change the model class parameter from a string to an actual class

A2rchi/utils/scraper.py

@@ -59,7 +59,7 @@ def collect_urls_from_lists(self):
                 data = f.read()
 
             for line in data.split("\n"):
-                if len(line) > 0 and line[0] != '#':
+                if len(line.lstrip())>0 and line.lstrip()[0:1] != "#":
                     urls.append(line)
 
         return urls

config/dev-config.yaml

@@ -32,6 +32,11 @@ chains:
       - User
       - A2rchi
       - Expert 
+  prompts:
+    # prompt that serves to condense a history and a question into a single question
+    CONDENSING_PROMPT: config/prompts/condense.prompt
+    # main prompt which takes in a single question and a context.
+    MAIN_PROMPT: config/prompts/submit.prompt
   chain: 
     # pick one of the models listed in the model class map below
     MODEL_NAME: OpenAILLM # LlamaLLM

config/prod-801-config.yaml

@@ -0,0 +1,110 @@
+global:
+  TRAINED_ON: "8.01"    #used to create name of the specific version of a2rchi we're using
+  DATA_PATH: "/root/data/"
+  ACCOUNTS_PATH: "/root/.accounts/"
+  LOCAL_VSTORE_PATH: "/root/data/vstore/"
+  ACCEPTED_FILES: 
+    -".txt"
+    -".html"
+    -".pdf"
+
+interfaces:
+  chat_app:
+    PORT: 7861
+    EXTERNAL_PORT: 7683
+    HOST: "0.0.0.0" # either "0.0.0.0" (for public) or "127.0.0.1" (for internal)
+    HOSTNAME: "ppc.mit.edu"  # careful, this is used for the chat service
+    template_folder: "/root/A2rchi/A2rchi/interfaces/chat_app/templates"
+    static_folder: "/root/A2rchi/A2rchi/interfaces/chat_app/static"
+  uploader_app:
+    PORT: 5001
+    HOST: "0.0.0.0" # either "0.0.0.0" (for public) or "127.0.0.1" (for internal)
+    template_folder: "/root/A2rchi/A2rchi/interfaces/uploader_app/templates"
+
+chains:
+  input_lists:
+    - empty.list
+    - miscellanea.list
+  base:
+    # roles that A2rchi knows about
+    ROLES: 
+      - User
+      - A2rchi
+      - Expert 
+  prompts:
+  # prompt that serves to condense a history and a question into a single question
+    CONDENSING_PROMPT: config/prompts/condense.prompt
+    # main prompt which takes in a single question and a context.
+    MAIN_PROMPT: config/prompts/801.prompt
+  chain: 
+    # pick one of the models listed in the model class map below
+    MODEL_NAME: OpenAILLM 
+    # map of all the class models and their keyword arguments
+    MODEL_CLASS_MAP:
+      OpenAILLM:
+        class: OpenAILLM
+        kwargs:
+          model_name: gpt-4
+          temperature: 1
+      DumbLLM:
+        class: DumbLLM
+        kwargs: 
+          filler: null
+      LlamaLLM:
+        class: LlamaLLM
+        kwargs:
+          base_model: "meta-llama/Llama-2-7b-chat-hf" #the location of the model (ex. meta-llama/Llama-2-70b)
+          peft_model: null #the location of the finetuning of the model. Can be none
+          enable_salesforce_content_safety: True # Enable safety check with Salesforce safety flan t5
+          quantization: True #enables 8-bit quantization
+          max_new_tokens: 4096 #The maximum numbers of tokens to generate
+          seed: null #seed value for reproducibility
+          do_sample: True #Whether or not to use sampling ; use greedy decoding otherwise.
+          min_length: null #The minimum length of the sequence to be generated, input prompt + min_new_tokens
+          use_cache: True  #[optional] Whether or not the model should use the past last key/values attentions Whether or not the model should use the past last key/values attentions (if applicable to the model) to speed up decoding.
+          top_p: .9 # [optional] If set to float < 1, only the smallest set of most probable tokens with probabilities that add up to top_p or higher are kept for generation.
+          temperature: .6 # [optional] The value used to modulate the next token probabilities.
+          top_k: 50 # [optional] The number of highest probability vocabulary tokens to keep for top-k-filtering.
+          repetition_penalty: 1.0 #The parameter for repetition penalty. 1.0 means no penalty.
+          length_penalty: 1 #[optional] Exponential penalty to the length that is used with beam-based generation.
+          max_padding_length: null # the max padding length to be used with tokenizer padding the prompts.
+    chain_update_time: 10 # the amount of time (in seconds) which passes between when the chain updates to the newest version of the vectorstore
+utils: 
+  cleo: 
+    cleo_update_time: 10
+  mailbox: 
+    IMAP4_PORT: 143
+    mailbox_update_time: 10
+  data_manager:
+    CHUNK_SIZE: 1000
+    CHUNK_OVERLAP: 0
+    use_HTTP_chromadb_client: True   # recommended: True (use http client for the chromadb vectorstore?)
+    # use_HTTP_chromadb_client: False
+    vectordb_update_time: 10
+    chromadb_host: chromadb-prod-801
+    chromadb_port: 8000
+    collection_name: "prod_801_collection" #unique in case vector stores are ever combined.
+    reset_collection: True           # reset the entire collection each time it is accessed by a new data manager instance
+  embeddings:
+    # choose one embedding from list below
+    EMBEDDING_NAME: OpenAIEmbeddings
+    # list of possible embeddings to use in vectorstore
+    EMBEDDING_CLASS_MAP:
+      OpenAIEmbeddings:
+        class: OpenAIEmbeddings
+        kwargs: 
+          model: text-embedding-ada-002
+        similarity_score_reference: 0.4
+      HuggingFaceEmbeddings:
+        class: HuggingFaceEmbeddings
+        kwargs:
+          model_name: "sentence-transformers/all-mpnet-base-v2"
+          model_kwargs:
+            device: 'cpu'
+          encode_kwargs: 
+            normalize_embeddings: True
+        similarity_score_reference: 0.9
+  scraper:
+    reset_data: True         # delete websites and sources.yml in data folder
+    verify_urls: False       # should be true when possible
+    enable_warnings: False   # keeps output clean if verify == False

config/prod-config.yaml

@@ -32,6 +32,11 @@ chains:
       - User
       - A2rchi
       - Expert 
+  prompts:
+    # prompt that serves to condense a history and a question into a single question
+    CONDENSING_PROMPT: config/prompts/condense.prompt
+    # main prompt which takes in a single question and a context.
+    MAIN_PROMPT: config/prompts/submit.prompt
   chain: 
     # pick one of the models listed in the model class map below
     MODEL_NAME: OpenAILLM 

config/prompts/801.prompt

@@ -0,0 +1,16 @@
+# Prompt used to qurery LLM with appropriate context and question.
+# This prompt is specific to 8.01 taught at MIT and likely will not perform well for other applications, where it is recommeneded to write your own prompt and change it in the config
+# 
+# All final promptsd must have the following tags in them, which will be filled with the appropriate information:
+#      {question}
+#      {context}
+#
+You are a conversational chatbot and teaching assisitant named A2rchi who helps students taking Classical Mechanics 1 at MIT (also called 8.01). You will be provided context to help you answer their questions. 
+Using your physics, math, and problem solving knowledge, answer the question at the end. Unless otherwise indicated, assume the users know high school level physics.
+Since you are a teaching assisitant, please try to give throughou answers to questions with explanations, instead of just giving the answer.
+If you don't know, say "I don't know". It is extremely important you only give correct answers. If you need to ask a follow up question, please do.
+
+Context: {context} 
+
+Question: {question}
+Helpful Answer:

config/prompts/condense.prompt

@@ -0,0 +1,13 @@
+# Prompt used to condense a chat history and a follow up question into a stand alone question. 
+# This is a very general prompt for condensing histories, so for base installs it will not need to be modified
+# 
+# All condensing prompts must have the following tags in them, which will be filled with the appropriate information:
+#      {chat_history}
+#      {question}
+#
+Given the following conversation between you (the AI named A2rchi), a human user who needs help, and an expert, and a follow up question, rephrase the follow up question to be a standalone question, in its original language.
+
+Chat History:
+{chat_history}
+Follow Up Input: {question}
+Standalone question:

config/prompts/submit.prompt

@@ -0,0 +1,16 @@
+# Prompt used to qurery LLM with appropriate context and question.
+# This prompt is specific to subMIT and likely will not perform well for other applications, where it is recommeneded to write your own prompt and change it in the config
+# 
+# All final promptsd must have the following tags in them, which will be filled with the appropriate information:
+#      {question}
+#      {context}
+#
+You are a conversational chatbot named A2rchi who helps people navigate a computing resource named subMIT. You will be provided context to help you answer their questions. 
+Using your linux and computing knowledge, answer the question at the end. Unless otherwise indicated, assume the users are not well versed computing.
+ Please do not assume that subMIT machines have anything installed on top of native linux except if the context mentions it.
+If you don't know, say "I don't know", if you need to ask a follow up question, please do.
+
+Context: {context} Additionally, it is always preferred to use conda, if possible.
+
+Question: {question}
+Helpful Answer: