Merging in release ROOT to staging branch

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

@@ -37,6 +37,7 @@ jobs:
           /bin/bash ${workspace}/deploy/create_secret.sh flask_uploader_app_secret_key.txt ${{ secrets.PROD_FLASK_UPLOADER_APP_SECRET_KEY }}
           /bin/bash ${workspace}/deploy/create_secret.sh uploader_salt.txt ${{ secrets.PROD_UPLOADER_SALT }}
           /bin/bash ${workspace}/deploy/create_secret.sh openai_api_key.txt ${{ secrets.OPENAI_API_KEY }}
+          /bin/bash ${workspace}/deploy/create_secret.sh anthropic_api_key.txt ${{ secrets.ANTHROPIC_API_KEY }}
           /bin/bash ${workspace}/deploy/create_secret.sh hf_token.txt ${{ secrets.HF_TOKEN }}
           /bin/bash ${workspace}/deploy/create_secret.sh pg_password.txt ${{ secrets.PROD_ROOT_PG_PASSWORD }}
           /bin/bash ${workspace}/deploy/create_secret.sh grafana_password.txt ${{ secrets.PROD_ROOT_GRAFANA_PG_PASSWORD }}
@@ -52,30 +53,30 @@ jobs:
       # create deployment directory if it doesn't already exist
       - name: Create Directory
         run: |
-          ssh submit06 "mkdir -p ~/A2rchi-prod-root/"
+          ssh submit-t3desk018 "mkdir -p ~/A2rchi-prod-root/"
 
       # stop any existing docker compose that's running
       - name: Stop Docker Compose
         run: |
-          ssh submit06 'bash -s' < ${{ github.workspace }}/deploy/prod-root/prod-root-stop.sh
+          ssh submit-t3desk018 'bash -s' < ${{ github.workspace }}/deploy/prod-root/prod-root-stop.sh
 
       # copy repository to machine
       - name: Copy Repository
         run: |
-          rsync -e ssh -r ${{ github.workspace}}/* --exclude .git/ --delete submit06:~/A2rchi-prod-root/
+          rsync -e ssh -r ${{ github.workspace}}/* --exclude .git/ --delete submit-t3desk018:~/A2rchi-prod-root/
 
       # run deploy script
       - name: Run Deploy Script
         run: |
           export tag="${GITHUB_REF#refs/heads/}"
           export tag="${tag//\//-}.${GITHUB_SHA}"
           sed -i "s/BASE_TAG/${tag}/" ${{ github.workspace }}/deploy/prod-root/prod-root-install.sh
-          ssh submit06 'bash -s' < ${{ github.workspace }}/deploy/prod-root/prod-root-install.sh
+          ssh submit-t3desk018 'bash -s' < ${{ github.workspace }}/deploy/prod-root/prod-root-install.sh
 
       # clean up secret files
       - name: Remove Secrets from Runner
         run: |
           rm ${{ github.workspace }}/deploy/prod-root/secrets/*.txt
 
       # print job status
-      - run: echo "🍏 This job's status is ${{ job.status }}."
+      - run: echo "🍏 This job's status is ${{ job.status }}."

README.md

@@ -1,4 +1,4 @@
-# A2rchi
+# A2rchi 
 
 An AI Augmented Research Chat Intelligence for MIT's subMIT project in the physics department
 

a2rchi/bin/service_chat.py

@@ -8,6 +8,7 @@
 import os
 
 # set openai
+os.environ['ANTHROPIC_API_KEY'] = read_secret("ANTHROPIC_API_KEY")
 os.environ['OPENAI_API_KEY'] = read_secret("OPENAI_API_KEY")
 os.environ['HUGGING_FACE_HUB_TOKEN'] = read_secret("HUGGING_FACE_HUB_TOKEN")
 config = Config_Loader().config["interfaces"]["chat_app"]

a2rchi/bin/service_uploader.py

@@ -13,6 +13,8 @@
 
 
 # set openai
+
+os.environ['ANTHROPIC_API_KEY'] = read_secret("ANTHROPIC_API_KEY")
 os.environ['OPENAI_API_KEY'] = read_secret("OPENAI_API_KEY")
 os.environ['HUGGING_FACE_HUB_TOKEN'] = read_secret("HUGGING_FACE_HUB_TOKEN")
 

a2rchi/chains/base.py

@@ -1,18 +1,23 @@
 """Chain for chatting with a vector database."""
 from __future__ import annotations
+from pydantic import BaseModel
 from loguru import logger
-from langchain.callbacks import FileCallbackHandler
+from langchain_core.callbacks.file import FileCallbackHandler
 
 from a2rchi.chains.prompts import CONDENSE_QUESTION_PROMPT, QA_PROMPT
 from a2rchi.utils.config_loader import Config_Loader
 
-from langchain.base_language import BaseLanguageModel
-from langchain.chains.combine_documents.stuff import StuffDocumentsChain
+from langchain_core.language_models.base import BaseLanguageModel
+from langchain.chains.combine_documents.stuff import StuffDocumentsChain #################### ****************
 from langchain.chains.conversational_retrieval.base import BaseConversationalRetrievalChain
-from langchain.chains.llm import LLMChain
-from langchain.schema import BaseRetriever, Document
-from langchain.schema.prompt_template import BasePromptTemplate
+from langchain.chains.llm import LLMChain #################### ****************
+from langchain_core.retrievers import BaseRetriever
+from langchain_core.documents import Document
+from langchain_core.prompts.base import BasePromptTemplate
+from langchain_core.runnables import RunnableSequence, RunnablePassthrough
 from typing import Any, Dict, List, Optional, Tuple
+from typing import Callable
+# from pydantic import model_rebuild
 import os 
 
 
@@ -46,7 +51,20 @@ class BaseSubMITChain(BaseConversationalRetrievalChain):
     """
     retriever: BaseRetriever # Index to connect to
     max_tokens_limit: Optional[int] = None # restrict doc length to return from store, enforced only for StuffDocumentChain
-    get_chat_history: Optional[function] = _get_chat_history
+    get_chat_history: Optional[Callable[[List[Tuple[str, str]]], str]] = _get_chat_history
+
+    # kuangfei: the rebuild logic is default or rewrite
+    # @classmethod
+    # def model_rebuild(
+    #     cls,
+    #     *,
+    #     force: bool = False,
+    #     raise_errors: bool = True,
+    #     _parent_namespace_depth: int = 2,
+    #     _types_namespace: dict[str, Any] | None = None,
+    # ) -> bool | None:
+    #     return False
+
 
     def _reduce_tokens_below_limit(self, docs: List[Document]) -> List[Document]:
         num_docs = len(docs)
@@ -104,6 +122,17 @@ def from_llm(
             callbacks = [handler],
             verbose=verbose,
         )
+
+        # llm_chain = RunnableSequence(
+        #     {
+        #         "sentence": RunnablePassthrough(),
+        #         "language": RunnablePassthrough()
+        #     }
+        #     | _prompt
+        #     | llm
+        #     # | output_parser
+        # )
+
         doc_chain = StuffDocumentsChain(
             llm_chain=llm_chain,
             document_variable_name=document_variable_name,
@@ -114,7 +143,7 @@ def from_llm(
         condense_question_chain = LLMChain(
             llm=_llm, prompt=condense_question_prompt, callbacks = [handler], verbose=verbose
         )
-
+        
         return cls(
             retriever=retriever,
             combine_docs_chain=doc_chain,

a2rchi/chains/chain.py

@@ -1,7 +1,7 @@
 from a2rchi.chains.base import BaseSubMITChain as BaseChain
 
 from chromadb.config import Settings
-from langchain.vectorstores import Chroma
+from langchain_chroma.vectorstores import Chroma
 
 import chromadb
 import time

a2rchi/chains/models.py

@@ -4,10 +4,16 @@
 import torch
 import time
 
-from langchain.callbacks.manager import CallbackManagerForLLMRun
-from langchain.llms.base import LLM
-from langchain.chat_models import ChatOpenAI
-from langchain.llms import LlamaCpp
+from langchain_core.callbacks.manager import CallbackManagerForLLMRun
+from langchain_core.language_models.llms import LLM
+from langchain_openai import ChatOpenAI
+from langchain_anthropic import ChatAnthropic
+
+
+import requests
+from typing import Optional, List
+
+
 
 class BaseCustomLLM(LLM):
     """
@@ -46,6 +52,21 @@ def _call(
         time.sleep(sleep_time)
         return "I am just a dumb LLM, I will give you a number: " + str(np.random.randint(10000, 99999))
 
+
+class AnthropicLLM(ChatAnthropic):
+    """
+    Loading Anthropic model from langchain package and specifying version. Options include:
+        model: str = "claude-3-opus-20240229"
+        model: str = "claude-3-sonnet-20240229"
+
+    Model comparison: https://docs.anthropic.com/en/docs/about-claude/models#model-comparison 
+    """
+
+    model: str = "claude-3-opus-20240229"
+    temp: int = 1
+
+
+
 class LlamaLLM(BaseCustomLLM):
     """
     Loading the Llama LLM from facebook. Make sure that the model
@@ -217,3 +238,77 @@ def __call__(self, output_text):
             report += "|" + "|".join(f"{n:^10}" for n in scores.keys()) + "|\n"
             report += "|" + "|".join(f"{n:^10}" for n in scores.values()) + "|\n"
         return "Salesforce Content Safety Flan T5 Base", is_safe, report
+
+class BaseCustomLLM(LLM):
+    """
+    Abstract class used to load a custom LLM
+    """
+    n_tokens: int = 100 # this has to be here for parent LLM class
+
+    @property
+    def _llm_type(self) -> str:
+        return "custom"
+
+    @abstractmethod
+    def _call(
+        self,
+        prompt: str,
+        stop: Optional[List[str]] = None,
+        run_manager: Optional[CallbackManagerForLLMRun] = None,
+    ) -> str:
+        pass
+
+
+class ClaudeLLM(BaseCustomLLM):
+    """
+    An LLM class that uses Anthropic's Claude model.
+    """
+    #TODO: obscure api key in final production version
+    api_key: str  = "INSERT KEY HERE!!!" # Claude API key
+    base_url: str = "https://api.anthropic.com/v1/messages"  # Anthropic API endpoint
+    model_name: str = "claude-3-5-sonnet-20240620"  # Specify the model version to use
+
+    verbose: bool = False
+
+    def _call(
+        self,
+        prompt: str = None,
+        stop: Optional[List[str]] = None,
+        run_manager: Optional[CallbackManagerForLLMRun] = None,
+        max_tokens: int = 1024,
+    ) -> str:
+
+        if stop is not None:
+            print("WARNING : currently this model does not support stop tokens")
+
+        if self.verbose:
+            print(f"INFO : Starting call to Claude with prompt: {prompt}")
+
+        headers = {
+            "x-api-key": self.api_key,  # Use the API key for the x-api-key header
+            "anthropic-version": "2023-06-01",  # Add the required version header
+            "Content-Type": "application/json"
+        }
+
+        # Modify the payload to match the required structure
+        payload = {
+            "model": self.model_name,  # You can keep this dynamic based on your code
+            "max_tokens": max_tokens,  # Update to match the required max_tokens
+            "messages": [  # Use a list of messages where each message has a role and content
+                {"role": "user", "content": prompt}  # Prompt becomes part of the message content
+            ]
+        }
+
+        if self.verbose:
+            print("INFO: Sending request to Claude API")
+
+        # Send request to Claude API
+        response = requests.post(self.base_url, headers=headers, json=payload)
+
+        if response.status_code == 200:
+            completion = response.json()["content"][0]["text"]
+            if self.verbose:
+                print(f"INFO : received response from Claude API: {completion}")
+            return completion
+        else:
+            raise Exception(f"API request to Claude failed with status {response.status_code}, {response.text}")

a2rchi/chains/prompts.py

@@ -1,5 +1,5 @@
 # flake8: noqa
-from langchain.prompts.prompt import PromptTemplate
+from langchain_core.prompts import PromptTemplate
 from a2rchi.utils.config_loader import Config_Loader
 
 config = Config_Loader().config["chains"]["prompts"]

a2rchi/utils/config_loader.py

@@ -1,7 +1,7 @@
-from a2rchi.chains.models import OpenAILLM, DumbLLM, LlamaLLM
+from a2rchi.chains.models import OpenAILLM, DumbLLM, LlamaLLM, AnthropicLLM
 
-from langchain.embeddings.openai import OpenAIEmbeddings
-from langchain.embeddings import HuggingFaceEmbeddings
+from langchain_openai import OpenAIEmbeddings
+from langchain_huggingface import HuggingFaceEmbeddings
 
 import os
 import yaml
@@ -23,6 +23,7 @@ def load_config(self):
 
             # change the model class parameter from a string to an actual class
             MODEL_MAPPING = {
+                "AnthropicLLM": AnthropicLLM,
                 "OpenAILLM": OpenAILLM,
                 "DumbLLM": DumbLLM,
                 "LlamaLLM": LlamaLLM

a2rchi/utils/data_manager.py

@@ -1,13 +1,12 @@
 from a2rchi.utils.scraper import Scraper
 
 from chromadb.config import Settings
-from langchain.document_loaders import TextLoader
-from langchain.document_loaders import PyPDFLoader
-from langchain.memory import ConversationBufferMemory
-from langchain.document_loaders import BSHTMLLoader
-from langchain.embeddings.openai import OpenAIEmbeddings
-from langchain.vectorstores import Chroma
-from langchain.text_splitter import CharacterTextSplitter
+from langchain_community.document_loaders.text import TextLoader
+from langchain_community.document_loaders import PyPDFLoader
+from langchain_community.document_loaders import BSHTMLLoader
+from langchain_openai import OpenAIEmbeddings
+from langchain_chroma.vectorstores import Chroma
+from langchain_text_splitters.character import CharacterTextSplitter
 
 import chromadb
 import hashlib

a2rchi/utils/scraper.py

@@ -1,4 +1,3 @@
-from piazza_api import Piazza
 
 import hashlib
 import os

config/prod-root-config.yaml

@@ -13,7 +13,7 @@ interfaces:
     PORT: 7861
     EXTERNAL_PORT: 7684
     HOST: "0.0.0.0" # either "0.0.0.0" (for public) or "127.0.0.1" (for internal)
-    HOSTNAME: "submit06.mit.edu"  # careful, this is used for the chat service
+    HOSTNAME: "t3desk018.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"
     num_responses_until_feedback: 3 #the number of responses given by A2rchi until she asks for feedback.
@@ -28,7 +28,7 @@ chains:
     - empty.list
     - miscellanea.list
     - root-docs.list
-    - root-tutorial.list
+    # - root-tutorial.list
     # - root-forum.list
   base:
     # roles that A2rchi knows about
@@ -47,13 +47,18 @@ chains:
     MAIN_PROMPT: config/prompts/root.prompt
   chain: 
     # pick one of the models listed in the model class map below
-    MODEL_NAME: OpenAILLM 
+    MODEL_NAME: AnthropicLLM 
     # map of all the class models and their keyword arguments
     MODEL_CLASS_MAP:
+      AnthropicLLM:
+        class: AnthropicLLM
+        kwargs:
+          model: claude-3-opus-20240229
+          temperature: 1 # not sure if 1 is best value? for now keeping consistent with prior settings.
       OpenAILLM:
         class: OpenAILLM
         kwargs:
-          model_name: gpt-4
+          model_name: gpt-4o-mini
           temperature: 1
       DumbLLM:
         class: DumbLLM
@@ -120,4 +125,4 @@ utils:
   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
+    enable_warnings: False   # keeps output clean if verify == False