Merge branch 'v0.1.1'

README.md

@@ -56,8 +56,8 @@
 To run this application, please make sure the following packages are installed on your system:
 
 - Node.js v20.11.1 & npm
-- PHP 8.1.27 & php-fpm & Composer
-- Python 3.9.5 & pip
+- PHP 8.1 & php-fpm & Composer
+- Python 3.10 & pip
 - Nginx or Apache
 - Redis 6.0.20
 - CUDA
@@ -84,6 +84,7 @@ Alternatively, you can refer to the following steps to install the entire system
      ./production_update.sh
      cd ../kernel
      pip install -r requirement.txt
+     sudo chown -R $(whoami):www-data /var/www/html
      ```
 
    - For Windows:

README_TW.md

@@ -56,8 +56,8 @@
 為了執行此應用，請確保您的系統上安裝了以下套件：
 
 - Node.js v20.11.1 & npm
-- PHP 8.1.27 & php-fpm & Composer
-- Python 3.9.5 & pip
+- PHP 8.1 & php-fpm & Composer
+- Python 3.10 & pip
 - Nginx 或 Apache
 - Redis 6.0.20
 - CUDA
@@ -84,6 +84,7 @@
      ./production_update.sh
      cd ../kernel
      pip install -r requirement.txt
+     sudo chown -R $(whoami):www-data /var/www/html
      ```
 
    - 對於 Windows:

docker/dbqa.yaml

@@ -0,0 +1,21 @@
+services:
+  dbqa-executor:
+    build:
+      context: ../
+      dockerfile: docker/executor/Dockerfile
+    image: kuwa-executor
+    environment:
+      CUSTOM_EXECUTOR_PATH: ./docqa/docqa.py
+      EXECUTOR_ACCESS_CODE: db-qa
+      EXECUTOR_NAME: DB QA
+    volumes: [ "</path/to/vector-database>:/var/database" ]
+    depends_on:
+      - kernel
+      - multi-chat
+    command: [
+      "--api_base_url", "http://web/",
+      "--access_code", "db-qa"
+      "--database", "/var/database"
+      ]
+    restart: unless-stopped
+    networks: ["backend", "frontend"]

docker/docqa_webqa.yaml

@@ -0,0 +1,20 @@
+services:
+  docqa-executor:
+    build:
+      context: ../
+      dockerfile: docker/executor/Dockerfile
+    image: kuwa-executor
+    environment:
+      CUSTOM_EXECUTOR_PATH: ./docqa/docqa.py
+      EXECUTOR_ACCESS_CODE: doc-qa;web-qa
+      EXECUTOR_NAME: DocQA;WebQA
+    depends_on:
+      - kernel
+      - multi-chat
+    command: [
+      "--api_base_url", "http://web/",
+      "--access_code", "doc-qa",
+      "--alt_access_code", "web-qa"
+      ]
+    restart: unless-stopped
+    networks: ["backend", "frontend"]

docker/executor/Dockerfile

@@ -10,6 +10,8 @@ COPY src/executor/requirements.txt ./
 RUN sed -i '/\.[/]*/d' ./requirements.txt &&\
     sed -i '/torch.*/d' ./requirements.txt
 RUN pip install --no-cache-dir -r requirements.txt
+COPY src/executor/docqa/requirements.txt ./docqa/requirements.txt
+RUN pip install --no-cache-dir -r ./docqa/requirements.txt
 
 # Install the executor framework
 COPY src/executor/. .

docker/executor/docker-entrypoint

@@ -1,19 +1,37 @@
-#!/bin/sh
+#!/bin/bash
 set -xeu
 
 add_executor(){
     if [ -z "${EXECUTOR_NAME:-}" ] || [ -z "${EXECUTOR_ACCESS_CODE:-}" ]
     then
         return
     fi
-    multi-chat-client add-executor "${EXECUTOR_ACCESS_CODE}" "${EXECUTOR_NAME}" || return 0
+    IFS=';' read -r -a access_code_arr <<< "${EXECUTOR_ACCESS_CODE}"
+    IFS=';' read -r -a name_arr <<< "${EXECUTOR_NAME}"
+    for idx in "${!name_arr[@]}"
+    do
+        access_code="${access_code_arr[idx]}"
+        name="${name_arr[idx]}"
+        multi-chat-client add-executor "${access_code}" "${name}" || true
+    done
 }
 
 if [ "${ADD_EXECUTOR_TO_MULTI_CHAT}" = true ]
 then
     add_executor
 fi
+
+if [ -z "${CUSTOM_EXECUTOR_PATH:-}" ]
+then
 exec kuwa-executor ${EXECUTOR_TYPE} \
     --kernel_url ${KERNEL_URL} \
     --access_code ${EXECUTOR_ACCESS_CODE} \
-    $@ 2>&1
+    $@ 2>&1
+else
+dir_name="$(dirname "${CUSTOM_EXECUTOR_PATH}")"
+file_name="$(basename "${CUSTOM_EXECUTOR_PATH}")"
+cd ${dir_name}
+exec python "${file_name}" \
+    --kernel_url ${KERNEL_URL} \
+    $@ 2>&1
+fi

docker/gemini.yaml

@@ -11,6 +11,6 @@ services:
     depends_on:
       - kernel
       - multi-chat
-    command: []
+    command: ["--api_key", "<YOUR_GLOBAL_API_KEY_HERE>"]
     restart: unless-stopped
     networks: ["backend"]

scripts/translate.py

@@ -20,11 +20,12 @@ def invoke_model(prompt, model="gemini-pro", base_url="https://chatdev.gai.tw"):
         if not resp.ok:
             raise RuntimeError(f'Request failed with status {response.status_code}')
         for line in resp.iter_lines(decode_unicode=True):
-            if not line or line == "event: end": break
+            if line == "event: close": break
             elif line.startswith("data: "):
                 chunk = json.loads(line[len("data: "):])["choices"][0]["delta"]["content"]
                 yield chunk
 
+
 def translate(input_filename, target_filename=None, lang="en", quite=False):
     prompt = {
         "en": "Translate to English.",
@@ -35,7 +36,7 @@ def translate(input_filename, target_filename=None, lang="en", quite=False):
     with open(input_filename,encoding="utf-8") as f:
         input_content = f.read()
     prompt += "\n" + input_content
-
+    
     output_content = ''
     for chunk in invoke_model(prompt):
         output_content += chunk

src/executor/docqa/.gitignore

@@ -0,0 +1 @@
+web_data/

src/executor/docqa/README.md

@@ -0,0 +1,27 @@
+Doc QA
+---
+A Retrieval-Augmented Generation Executor.
+
+## Production Deployment
+
+This executor has two configurations.
+1. DocQA/WebQA: Retrieve information from the documents or web pages
+2. DatabaseQA(DBQA): Retrieve information from a pre-built vector database
+
+### Prerequisite
+1. Install the dependency
+```sh
+cd src/executor/webqa
+pip install -r requirements.txt
+```
+2. You can obtain the KUWA_TOKEN from the website's profile > API Token Management > Kuwa Chat API Token
+
+### DocQA/WebQA
+```
+python ./docqa.py --access_code doc-qa --api_base_url http://localhost/ --api_key <KUWA_TOKEN> [--model <MODEL_NAME>]
+```
+
+### DBQA
+```
+python ./docqa.py --access_code dbqa --api_base_url http://localhost/ --api_key <KUWA_TOKEN> [--model <MODEL_NAME>] --database /path/to/pre-built/vector-db
+```

src/executor/working/docqa/README.md → src/executor/docqa/archive/README.md

@@ -1,6 +1,6 @@
-Document QA (DocQA)
+Web QA
 ---
-Retrieve-Augmented Generation Worker.
+A Retrieval-Augmented Generation Executor.
 
 ## Production Deployment
 

src/executor/docqa/docqa.py

@@ -0,0 +1,154 @@
+#!/bin/python3
+# -#- coding: UTF-8 -*-
+
+import os
+import re
+import sys
+import logging
+import asyncio
+import functools
+import itertools
+import requests
+import json
+import i18n
+sys.path.append(os.path.dirname(os.path.abspath(__file__)))
+
+from typing import Generator
+from urllib.parse import urljoin
+from kuwa.executor import LLMExecutor
+
+from src.docqa import DocQa
+from src.kuwa_llm_client import KuwaLlmClient
+from src.document_store import DocumentStore
+
+logger = logging.getLogger(__name__)
+
+class NoUrlException(Exception):
+    def __init__(self, msg):
+        self.msg = msg
+    def __str__(self):
+        return self.msg
+
+class DocQaExecutor(LLMExecutor):
+    def __init__(self):
+        super().__init__()
+
+    def _try_register(self):
+        super()._try_register()
+        if self.alt_access_code:
+            resp = requests.post(
+                url=urljoin(self.kernel_url, f"{self.executor_iface_version}/worker/register"),
+                data={"name": self.alt_access_code, "endpoint": self.get_reg_endpoint()}
+            )
+            if not resp.ok or resp.text == "Failed":
+                raise RuntimeWarning("The server failed to register to kernel.")
+
+    def _shut_down(self):
+        super()._shut_down()
+        if self.alt_access_code:
+            try:
+                response = requests.post(
+                    urljoin(self.kernel_url, f"{self.executor_iface_version}/worker/unregister"),
+                    data={"name": self.alt_access_code,"endpoint": self.get_reg_endpoint()}
+                )
+                if not response.ok or response.text == "Failed":
+                    raise RuntimeWarning()
+                else:
+                    logger.info("Unregistered from kernel.")
+            except requests.exceptions.ConnectionError as e:
+                logger.warning("Failed to unregister from kernel")
+
+    def extend_arguments(self, parser):
+        parser.add_argument('--lang', default="en", help='The language code to internationalize the aplication. See \'lang/\'')
+        parser.add_argument('--database', default=None, type=str, help='The path the the pre-built database.')
+        parser.add_argument('--api_base_url', default="http://127.0.0.1/", help='The API base URL of Kuwa multi-chat WebUI')
+        parser.add_argument('--api_key', default=None, help='The API authentication token of Kuwa multi-chat WebUI')
+        parser.add_argument('--limit', default=3072, type=int, help='The limit of the LLM\'s context window')
+        parser.add_argument('--model', default=None, help='The model name (access code) on Kuwa multi-chat WebUI')
+        parser.add_argument('--mmr_k', default=6, type=int, help='Number of chunk to retrieve after Maximum Marginal Relevance (MMR).')
+        parser.add_argument('--mmr_fetch_k', default=12, type=int, help='Number of chunk to retrieve before Maximum Marginal Relevance (MMR).')
+        parser.add_argument('--chunk_size', default=512, type=int, help='The charters in the chunk.')
+        parser.add_argument('--chunk_overlap', default=128, type=int, help='The overlaps between chunks.')
+        parser.add_argument('--alt_access_code', default=None, type=str, help='The alternate access code.')
+
+    def setup(self):
+        i18n.load_path.append(f'lang/{self.args.lang}/')
+        i18n.config.set("error_on_missing_translation", True)
+        i18n.config.set("locale", self.args.lang)
+
+        self.pre_built_db = self.args.database
+        self.llm = KuwaLlmClient(
+            base_url = self.args.api_base_url,
+            kernel_base_url = self.kernel_url,
+            model=self.args.model,
+            auth_token=self.args.api_key
+        )
+        self.document_store = DocumentStore(
+            mmr_k = self.args.mmr_k,
+            mmr_fetch_k = self.args.mmr_fetch_k,
+            chunk_size = self.args.chunk_size,
+            chunk_overlap = self.args.chunk_overlap
+        )
+        self.docqa = DocQa(
+            document_store = self.document_store,
+            vector_db = self.pre_built_db,
+            llm = self.llm,
+            lang = self.args.lang
+        )
+        self.alt_access_code = self.args.alt_access_code
+        self.proc = False
+
+    def extract_last_url(self, chat_history: list):
+        """
+        Find the latest URL provided by the user and trim the chat history to there.
+        """
+
+        url = None
+        begin_index = 0
+        user_records = list(filter(lambda x: not x["isbot"], chat_history))
+        for i, record in enumerate(reversed(user_records)):
+
+            urls_in_msg = re.findall(r'^(https?://[^\s]+)$', record["msg"])
+            if len(urls_in_msg) != 0: 
+                url = urls_in_msg[-1]
+                begin_index = len(chat_history) - i - 1
+                break
+
+        return url, chat_history[begin_index:]
+
+    async def llm_compute(self, data):
+        chat_history = json.loads(data.get("input"))
+        auth_token = data.get("user_token") or self.args.api_key
+        url = None
+
+        try:
+            if self.pre_built_db == None:
+                url, chat_history = self.extract_last_url(chat_history)
+                if url == None : raise NoUrlException(i18n.t('docqa.no_url_exception'))
+
+                chat_history = [{"isbot": False, "msg": None}] + chat_history[1:]
+            self.proc = True
+            response_generator = self.docqa.process(urls=[url], chat_history=chat_history, auth_token=auth_token)
+            async for reply in response_generator:
+                if not self.proc:
+                    await response_generator.aclose()
+                yield reply
+
+        except NoUrlException as e:
+            yield str(e)
+
+        except Exception as e:
+            await asyncio.sleep(2) # To prevent SSE error of web page.
+            logger.exception('Unexpected error')
+            yield i18n.t("docqa.default_exception_msg")
+
+    async def abort(self):
+        if self.proc:
+            self.proc = False
+            logger.debug("aborted")
+            return "Aborted"
+        return "No process to abort"
+
+if __name__ == "__main__":
+    executor = DocQaExecutor()
+    executor.run()

src/executor/docqa/lang/en/docqa.en.yml

@@ -0,0 +1,7 @@
+en:
+  no_url_exception: No URL found.
+  default_exception_msg: An error occurred, please try again or contact the administrator.
+  summary_prompt: "Please provide a summary of this article:"
+  error_fetching_document: An error occurred while trying to fetch the document {}. Please make sure the submitted document exists and is publicly available.
+  summary_question: When, where, what, outcome, summary
+  summary_prefix: "Here is a summary of the article:"

src/executor/docqa/lang/en/prompt_template/llm_input_qa.mustache

@@ -0,0 +1,7 @@
+Please answer the questions based on the content below.{{#ref}}If there are references, please include them at the end of the answer.{{/ref}}
+
+{{#docs}}{{page_content}}
+
+{{/docs}}
+
+Question: {{question}}

src/executor/docqa/lang/en/prompt_template/llm_input_summary.mustache

@@ -0,0 +1,10 @@
+Article Content:
+
+{{#docs}}{{page_content}}
+
+{{/docs}}
+
+
+Please provide a highlight summary of the above article.
+
+Summary:

src/executor/docqa/lang/en/prompt_template/llm_input_translate.mustache

@@ -0,0 +1 @@
+Translate to English: {{question}}