From a180f7bef10c1907c7ef10d833c5b63d9153def0 Mon Sep 17 00:00:00 2001
From: Bill Cui <bill.cui.work@gmail.com>
Date: Mon, 19 Jun 2023 09:11:05 -0400
Subject: [PATCH 1/7] chatgpt can talk to acl

---
 pyserini/demo/acl-chatgpt.py |  27 ++++++++
 pyserini/demo/acl-server.py  | 116 +++++++++++++++++++++++++++++++++++
 pyserini/demo/skill.py       |  48 +++++++++++++++
 3 files changed, 191 insertions(+)
 create mode 100644 pyserini/demo/acl-chatgpt.py
 create mode 100644 pyserini/demo/acl-server.py
 create mode 100644 pyserini/demo/skill.py

diff --git a/pyserini/demo/acl-chatgpt.py b/pyserini/demo/acl-chatgpt.py
new file mode 100644
index 000000000..e5320c77f
--- /dev/null
+++ b/pyserini/demo/acl-chatgpt.py
@@ -0,0 +1,27 @@
+import semantic_kernel as sk
+from semantic_kernel.connectors.ai.open_ai import OpenAITextCompletion, AzureTextCompletion
+
+from pyserini.demo.skill import PyseriniSkill
+
+kernel = sk.Kernel()
+
+api_key, org_id = sk.openai_settings_from_dot_env()
+kernel.add_text_completion_service("dv", OpenAITextCompletion("text-davinci-003", api_key, org_id))
+
+kernel.import_skill(PyseriniSkill(),"pyserini")
+
+
+sk_prompt = """
+{{pyserini.search $input}}
+
+Based on the information provided above, please answer the following question:
+
+{{$input}}
+"""
+
+context = kernel.create_new_context()
+context["url"] = "http://127.0.0.1:8080/search"
+
+acl_function = kernel.create_semantic_function(sk_prompt, max_tokens=200, temperature=0, top_p=0.5)
+
+print(acl_function("Tell me about the Computational Power of Transformers and Its Implications in Sequence Modeling", context=context))
\ No newline at end of file
diff --git a/pyserini/demo/acl-server.py b/pyserini/demo/acl-server.py
new file mode 100644
index 000000000..bca8eed6b
--- /dev/null
+++ b/pyserini/demo/acl-server.py
@@ -0,0 +1,116 @@
+#
+# Pyserini: Reproducible IR research with sparse and dense representations
+#
+# 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.
+#
+
+"""
+This script provides an interactive web interface demo for retrieval on the ACL dataset.
+It requires `flask` (`pip install flask~=2.2.0`).
+An example command looks like `python -m pyserini.demo.acl` that starts up a server on port 8080.
+The demo can be accessed via "http://localhost:8080" in a web browser.
+Additional arguments include:
+    --port [PORT] --hits [Number of hits]
+    --k1 [BM25 k1] --b [BM25 b] --device [cpu, cuda]
+"""
+import json
+import logging
+from argparse import ArgumentParser
+from functools import partial
+from typing import Callable, Optional, Tuple, Union
+
+from flask import Flask, render_template, request, flash, jsonify
+from pyserini.search import LuceneSearcher, FaissSearcher, AutoQueryEncoder
+
+logging.basicConfig(
+    format='%(asctime)s | %(levelname)s | %(name)s | %(message)s',
+    datefmt='%Y-%m-%d %H:%M:%S',
+    level=logging.INFO,
+)
+logger = logging.getLogger('acl-demo')
+
+VERSION = '1.0'
+Searcher = Union[FaissSearcher, LuceneSearcher]
+
+
+def create_app(k: int, load_searcher_fn: Callable[[str], Tuple[Searcher, str]]):
+    app = Flask(__name__)
+
+    lang = 'en'
+    searcher, retriever = load_searcher_fn(lang)
+
+    @app.route('/search', methods=['POST'])
+    def search():
+        nonlocal lang, searcher, retriever
+        query = request.json['query']
+        if not query:
+            search_results = []
+        else:
+            hits = searcher.search(query, k=k)
+            docs = [searcher.doc(hit.docid) for hit in hits]
+            search_results = [
+                {
+                    'rank': r + 1,
+                    'docid': hit.docid,
+                    'doc': docs[r].contents(),
+                    'score': hit.score,
+                }
+                for r, hit in enumerate(hits)
+            ]
+        return search_results
+
+
+    return app
+
+
+def _load_sparse_searcher(language: str, k1: Optional[float]=None, b: Optional[float]=None) -> (Searcher, str):
+    searcher = LuceneSearcher('indexes/lucene-index-acl-paragraph')
+    searcher.set_language(language)
+    if k1 is not None and b is not None:
+        searcher.set_bm25(k1, b)
+        retriever_name = f'BM25 (k1={k1}, b={b})'
+    else:
+        retriever_name = 'BM25'
+
+    return searcher, retriever_name
+
+
+def main():
+    parser = ArgumentParser()
+
+    parser.add_argument('--k1', type=float, help='BM25 k1 parameter.')
+    parser.add_argument('--b', type=float, help='BM25 b parameter.')
+    parser.add_argument('--hits', type=int, default=10, help='Number of hits returned by the retriever')
+    parser.add_argument(
+        '--device',
+        type=str,
+        default='cpu',
+        help='Device to run query encoder, cpu or [cuda:0, cuda:1, ...] (used only when index is based on FAISS)',
+    )
+    parser.add_argument(
+        '--port',
+        default=8080,
+        type=int,
+        help='Web server port',
+    )
+
+    args = parser.parse_args()
+
+    load_fn = partial(_load_sparse_searcher, k1=args.k1, b=args.b)
+
+    app = create_app(args.hits, load_fn)
+    app.run(host='0.0.0.0', port=args.port)
+
+
+if __name__ == '__main__':
+    main()
diff --git a/pyserini/demo/skill.py b/pyserini/demo/skill.py
new file mode 100644
index 000000000..df512e527
--- /dev/null
+++ b/pyserini/demo/skill.py
@@ -0,0 +1,48 @@
+# Copyright (c) Microsoft. All rights reserved.
+
+import json
+
+import aiohttp
+
+from semantic_kernel.orchestration.sk_context import SKContext
+from semantic_kernel.skill_definition import sk_function, sk_function_context_parameter
+
+
+class PyseriniSkill:
+    """
+    A skill that uses Pyserini to search a corpus of documents.
+
+    Usage:
+        kernel.import_skill(PyseriniSkill(), "http")
+
+    Examples:
+
+        {{pyserini.search $query}}
+    """
+
+    @sk_function(description="Searches a corpus of documents using Pyserini using the specified query.", name="search")
+    @sk_function_context_parameter(name="url", description="The url of the request")
+    async def search(self, query: str, context: SKContext) -> str:
+        """
+        Searches a corpus of documents using Pyserini using the specified query.
+        Return the response body as a string.
+
+        params:
+            query: The query to search for.
+            context: The SKContext containing the url of the request.
+        returns:
+            The response body as a string.
+        """
+        _, url = context.variables.get("url")
+        if not url:
+            raise ValueError("url cannot be `None` or empty")
+
+        headers = {"Content-Type": "application/json"}
+        body = {"query": query}
+        data = json.dumps(body)
+        async with aiohttp.ClientSession() as session:
+            async with session.post(
+                    url, headers=headers, data=data, raise_for_status=True
+            ) as response:
+                return await response.text()
+

From 2ce0747a9da02ed8763e4d352f0e67b55efe8afb Mon Sep 17 00:00:00 2001
From: Bill Cui <bill.cui.work@gmail.com>
Date: Mon, 19 Jun 2023 09:51:44 -0400
Subject: [PATCH 2/7] chatbot done?

---
 pyserini/demo/acl-chatgpt.py                  | 27 -----------
 pyserini/demo/aclchatgpt/chatbot.py           | 46 +++++++++++++++++++
 .../{acl-server.py => aclchatgpt/server.py}   |  0
 pyserini/demo/{ => aclchatgpt}/skill.py       | 10 ++--
 4 files changed, 49 insertions(+), 34 deletions(-)
 delete mode 100644 pyserini/demo/acl-chatgpt.py
 create mode 100644 pyserini/demo/aclchatgpt/chatbot.py
 rename pyserini/demo/{acl-server.py => aclchatgpt/server.py} (100%)
 rename pyserini/demo/{ => aclchatgpt}/skill.py (78%)

diff --git a/pyserini/demo/acl-chatgpt.py b/pyserini/demo/acl-chatgpt.py
deleted file mode 100644
index e5320c77f..000000000
--- a/pyserini/demo/acl-chatgpt.py
+++ /dev/null
@@ -1,27 +0,0 @@
-import semantic_kernel as sk
-from semantic_kernel.connectors.ai.open_ai import OpenAITextCompletion, AzureTextCompletion
-
-from pyserini.demo.skill import PyseriniSkill
-
-kernel = sk.Kernel()
-
-api_key, org_id = sk.openai_settings_from_dot_env()
-kernel.add_text_completion_service("dv", OpenAITextCompletion("text-davinci-003", api_key, org_id))
-
-kernel.import_skill(PyseriniSkill(),"pyserini")
-
-
-sk_prompt = """
-{{pyserini.search $input}}
-
-Based on the information provided above, please answer the following question:
-
-{{$input}}
-"""
-
-context = kernel.create_new_context()
-context["url"] = "http://127.0.0.1:8080/search"
-
-acl_function = kernel.create_semantic_function(sk_prompt, max_tokens=200, temperature=0, top_p=0.5)
-
-print(acl_function("Tell me about the Computational Power of Transformers and Its Implications in Sequence Modeling", context=context))
\ No newline at end of file
diff --git a/pyserini/demo/aclchatgpt/chatbot.py b/pyserini/demo/aclchatgpt/chatbot.py
new file mode 100644
index 000000000..e5d147558
--- /dev/null
+++ b/pyserini/demo/aclchatgpt/chatbot.py
@@ -0,0 +1,46 @@
+import semantic_kernel as sk
+from semantic_kernel.connectors.ai.open_ai import OpenAITextCompletion
+
+from pyserini.demo.aclchatgpt.skill import PyseriniSkill
+
+kernel = sk.Kernel()
+
+api_key, org_id = sk.openai_settings_from_dot_env()
+kernel.add_text_completion_service("dv", OpenAITextCompletion("text-davinci-003", api_key, org_id))
+
+kernel.import_skill(PyseriniSkill(),"pyserini")
+
+
+sk_prompt = """
+ChatBot can answer questions based on query_results.
+It can only answer questions based on query_results.
+If query_results is empty, say "I don't know"
+
+============
+query_results: {{pyserini.search $input}}
+============
+
+Question: {{$input}}
+"""
+
+
+
+context = kernel.create_new_context()
+context["url"] = "http://127.0.0.1:8080/search"
+
+acl_chat_function = kernel.create_semantic_function(sk_prompt, max_tokens=200, temperature=0, top_p=0.5)
+
+
+def chat(input_text: str) -> None:
+    # Save new message in the context variables
+
+
+    # Process the user message and get an answer
+    answer = acl_chat_function(input_text,context=context)
+
+    # Show the response
+    print(f"ChatBot: {answer}")
+
+
+while True:
+    chat(input("User: "))
\ No newline at end of file
diff --git a/pyserini/demo/acl-server.py b/pyserini/demo/aclchatgpt/server.py
similarity index 100%
rename from pyserini/demo/acl-server.py
rename to pyserini/demo/aclchatgpt/server.py
diff --git a/pyserini/demo/skill.py b/pyserini/demo/aclchatgpt/skill.py
similarity index 78%
rename from pyserini/demo/skill.py
rename to pyserini/demo/aclchatgpt/skill.py
index df512e527..fa27ed9de 100644
--- a/pyserini/demo/skill.py
+++ b/pyserini/demo/aclchatgpt/skill.py
@@ -3,6 +3,7 @@
 import json
 
 import aiohttp
+import requests
 
 from semantic_kernel.orchestration.sk_context import SKContext
 from semantic_kernel.skill_definition import sk_function, sk_function_context_parameter
@@ -37,12 +38,7 @@ async def search(self, query: str, context: SKContext) -> str:
         if not url:
             raise ValueError("url cannot be `None` or empty")
 
-        headers = {"Content-Type": "application/json"}
         body = {"query": query}
-        data = json.dumps(body)
-        async with aiohttp.ClientSession() as session:
-            async with session.post(
-                    url, headers=headers, data=data, raise_for_status=True
-            ) as response:
-                return await response.text()
+        result  = requests.post(url, json = body).json()[0]["doc"]
+        return result
 

From 4c296598ad84597b04bd8d868acef4611be370d8 Mon Sep 17 00:00:00 2001
From: Bill Cui <bill.cui.work@gmail.com>
Date: Mon, 19 Jun 2023 11:45:39 -0400
Subject: [PATCH 3/7] remove answer

---
 pyserini/demo/aclchatgpt/chatbot.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/pyserini/demo/aclchatgpt/chatbot.py b/pyserini/demo/aclchatgpt/chatbot.py
index e5d147558..222956b2f 100644
--- a/pyserini/demo/aclchatgpt/chatbot.py
+++ b/pyserini/demo/aclchatgpt/chatbot.py
@@ -20,7 +20,7 @@
 query_results: {{pyserini.search $input}}
 ============
 
-Question: {{$input}}
+{{$input}}
 """
 
 

From 00f397568a96c7123fbebc226178389804dc9366 Mon Sep 17 00:00:00 2001
From: Bill Cui <bill.cui.work@gmail.com>
Date: Mon, 19 Jun 2023 11:54:06 -0400
Subject: [PATCH 4/7] documents

---
 pyserini/demo/aclchatgpt/readme.md | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)
 create mode 100644 pyserini/demo/aclchatgpt/readme.md

diff --git a/pyserini/demo/aclchatgpt/readme.md b/pyserini/demo/aclchatgpt/readme.md
new file mode 100644
index 000000000..64b49b409
--- /dev/null
+++ b/pyserini/demo/aclchatgpt/readme.md
@@ -0,0 +1,19 @@
+# Talking to ACL Anthology with GPT
+
+By default, GPT-3 text-davinci-003 is used. You can change this in `chatbot.py`.
+
+`chatbot.py` is the front-end script. As a CLI, the user can input a query and the chatbot will return a response.
+
+`server.py` is the back-end script. It is a Flask server that will return a response to a query sent to the server.
+
+You must have both running for `chatbot.py` to work.
+
+## Setup
+
+1. Follow the instructions in [Indexing the ACL Anthology with Anserini](https://github.com/castorini/pyserini/blob/master/docs/working-with-acl-anthology.md) to setup the project and generate a `lucene-index-acl-paragraph` index.
+2. Copy the generated `lucene-index-acl-paragraph` index from the `acl-anthology` folder to `pyserini/indexes`
+3. You will need Semantic Kernel as well. 
+   - `pip3 install --upgrade semantic-kernel`
+4. Start the server with `python -m pyserini.demo.aclchatgpt.server`
+5. Start the chatbot with `python -m pyserini.demo.aclchatgpt.chatbot`
+6. Start chatting!
\ No newline at end of file

From 2fd2a61cfd8dcc2814b6bd29f081ae7f42f8c17c Mon Sep 17 00:00:00 2001
From: Bill Cui <bill.cui.work@gmail.com>
Date: Tue, 20 Jun 2023 21:26:39 -0400
Subject: [PATCH 5/7] added conversation capability

---
 pyserini/demo/aclchatgpt/chatbot.py | 47 +++++++++++++++++++++++------
 pyserini/demo/aclchatgpt/readme.md  |  8 +++++
 2 files changed, 45 insertions(+), 10 deletions(-)

diff --git a/pyserini/demo/aclchatgpt/chatbot.py b/pyserini/demo/aclchatgpt/chatbot.py
index 222956b2f..ec3d4cb53 100644
--- a/pyserini/demo/aclchatgpt/chatbot.py
+++ b/pyserini/demo/aclchatgpt/chatbot.py
@@ -11,36 +11,63 @@
 kernel.import_skill(PyseriniSkill(),"pyserini")
 
 
-sk_prompt = """
-ChatBot can answer questions based on query_results.
-It can only answer questions based on query_results.
-If query_results is empty, say "I don't know"
 
-============
+
+acl_chat_prompt = """
+Given the `query_results` below, your task is to formulate an answer using only the information 
+provided in these results. You should not draw from other sources or attempt to provide information that is not 
+contained within the `query_results`. If the `query_results` are empty, simply state "I'm sorry, but I do not have 
+enough information to provide an answer."
+
+===================
 query_results: {{pyserini.search $input}}
-============
+===================
 
-{{$input}}
+Based on the above `query_results`, what is your response to {{$input}}?
 """
 
+absolute_question_prompt = """
+Task: You are an AI language model tasked with transforming given questions into 
+absolute questions. An absolute question is a question that can stand on its own and carries all the context needed 
+to be answered. Here's an example:
+
+User: Who is Alan Turing?
+ChatBot: Who is Alan Turing?
+User: How old is he?
+ChatBot: How old is Alan Turing?
+
+===================
+History: {{$history}}
+===================
+
+Using the history as context, transform the following question into an absolute question: {{$input}}
+"""
 
 
 context = kernel.create_new_context()
 context["url"] = "http://127.0.0.1:8080/search"
+context["history"] = ""
 
-acl_chat_function = kernel.create_semantic_function(sk_prompt, max_tokens=200, temperature=0, top_p=0.5)
+acl_chat_function = kernel.create_semantic_function(acl_chat_prompt, max_tokens=200, temperature=0, top_p=0.5)
+absolute_question_function = kernel.create_semantic_function(absolute_question_prompt, max_tokens=200, temperature=0, top_p=0.5)
 
 
 def chat(input_text: str) -> None:
-    # Save new message in the context variables
 
+    print("---------------------------------------------")
+    absolute_question = absolute_question_function(input_text,context=context)
+
+    print (f"Absolute Question: {absolute_question}")
 
     # Process the user message and get an answer
-    answer = acl_chat_function(input_text,context=context)
+    answer = acl_chat_function(str(absolute_question),context=context)
 
     # Show the response
     print(f"ChatBot: {answer}")
 
+    context["history"] += f"\nUser: {input_text}\nChatBot: {answer}\n"
+
 
 while True:
+    print("=============================================")
     chat(input("User: "))
\ No newline at end of file
diff --git a/pyserini/demo/aclchatgpt/readme.md b/pyserini/demo/aclchatgpt/readme.md
index 64b49b409..df9c7cbb4 100644
--- a/pyserini/demo/aclchatgpt/readme.md
+++ b/pyserini/demo/aclchatgpt/readme.md
@@ -8,6 +8,14 @@ By default, GPT-3 text-davinci-003 is used. You can change this in `chatbot.py`.
 
 You must have both running for `chatbot.py` to work.
 
+## Environment Varaibles
+You will need to create an `.env` file in directory that you are running in.
+The `.env` file should contain the following:
+```
+OPENAI_API_KEY="<your key>"
+OPENAI_ORG_ID="<your org id>"
+```
+
 ## Setup
 
 1. Follow the instructions in [Indexing the ACL Anthology with Anserini](https://github.com/castorini/pyserini/blob/master/docs/working-with-acl-anthology.md) to setup the project and generate a `lucene-index-acl-paragraph` index.

From c18e506129f0e01bb034d7c4e8c24977ff8f4a16 Mon Sep 17 00:00:00 2001
From: Bill Cui <bill.cui.work@gmail.com>
Date: Wed, 21 Jun 2023 18:50:01 -0400
Subject: [PATCH 6/7] pyserini skill

---
 pyserini/demo/aclchatgpt/chatbot.py | 120 +++++++++++++++++-----------
 pyserini/demo/aclchatgpt/server.py  | 116 ---------------------------
 pyserini/demo/aclchatgpt/skill.py   |  38 ++++++++-
 3 files changed, 110 insertions(+), 164 deletions(-)
 delete mode 100644 pyserini/demo/aclchatgpt/server.py

diff --git a/pyserini/demo/aclchatgpt/chatbot.py b/pyserini/demo/aclchatgpt/chatbot.py
index ec3d4cb53..6e80cae66 100644
--- a/pyserini/demo/aclchatgpt/chatbot.py
+++ b/pyserini/demo/aclchatgpt/chatbot.py
@@ -1,73 +1,103 @@
+from dataclasses import dataclass
+from argparse import ArgumentParser
+from typing import Optional
+
 import semantic_kernel as sk
 from semantic_kernel.connectors.ai.open_ai import OpenAITextCompletion
 
-from pyserini.demo.aclchatgpt.skill import PyseriniSkill
+from pyserini.demo.aclchatgpt.skill import PyseriniSkill, PyseriniConfig
+
+
+@dataclass
+class OpenAIConfig:
+    api_key: str
+    org_id: str
 
-kernel = sk.Kernel()
+class ChatBot:
 
-api_key, org_id = sk.openai_settings_from_dot_env()
-kernel.add_text_completion_service("dv", OpenAITextCompletion("text-davinci-003", api_key, org_id))
+    acl_chat_prompt = """
+    Given the `query_results` below, your task is to formulate an answer using only the information 
+    provided in these results. You should not draw from other sources or attempt to provide information that is not 
+    contained within the `query_results`. If the `query_results` are empty, simply state "I'm sorry, but I do not have 
+    enough information to provide an answer."
+    
+    ===================
+    query_results: {{pyserini.search $input}}
+    ===================
+    
+    Based on the above `query_results`, what is your response to {{$input}}?
+    """
 
-kernel.import_skill(PyseriniSkill(),"pyserini")
+    absolute_question_prompt = """
+    Task: You are an AI language model tasked with transforming given questions into 
+    absolute questions. An absolute question is a question that can stand on its own and carries all the context needed 
+    to be answered. Here's an example:
+    
+    User: Who is Alan Turing?
+    ChatBot: Who is Alan Turing?
+    User: How old is he?
+    ChatBot: How old is Alan Turing?
+    
+    ===================
+    History: {{$history}}
+    ===================
+    
+    Using the history as context, transform the following question into an absolute question: {{$input}}
+    """
 
+    def __init__(self, pyserini_config: PyseriniConfig, openai_config: OpenAIConfig):
 
+        self.kernel = sk.Kernel()
+        self.kernel.add_text_completion_service("dv", OpenAITextCompletion("text-davinci-003", openai_config.api_key, openai_config.org_id))
+        self.kernel.import_skill(PyseriniSkill(pyserini_config),"pyserini")
+        self.context = self.kernel.create_new_context()
+        self.context["url"] = "http://127.0.0.1:8080/search"
+        self.context["history"] = ""
 
+        self.acl_chat_function = self.kernel.create_semantic_function(self.acl_chat_prompt, max_tokens=200, temperature=0, top_p=0.5)
+        self.absolute_question_function = self.kernel.create_semantic_function(self.absolute_question_prompt, max_tokens=200, temperature=0, top_p=0.5)
 
-acl_chat_prompt = """
-Given the `query_results` below, your task is to formulate an answer using only the information 
-provided in these results. You should not draw from other sources or attempt to provide information that is not 
-contained within the `query_results`. If the `query_results` are empty, simply state "I'm sorry, but I do not have 
-enough information to provide an answer."
 
-===================
-query_results: {{pyserini.search $input}}
-===================
+    def _chat(self,input_text: str) -> None:
 
-Based on the above `query_results`, what is your response to {{$input}}?
-"""
+        print("---------------------------------------------")
+        absolute_question = self.absolute_question_function(input_text,context=self.context)
 
-absolute_question_prompt = """
-Task: You are an AI language model tasked with transforming given questions into 
-absolute questions. An absolute question is a question that can stand on its own and carries all the context needed 
-to be answered. Here's an example:
+        print (f"Absolute Question: {absolute_question}")
 
-User: Who is Alan Turing?
-ChatBot: Who is Alan Turing?
-User: How old is he?
-ChatBot: How old is Alan Turing?
+        # Process the user message and get an answer
+        answer = self.acl_chat_function(str(absolute_question),context=self.context)
 
-===================
-History: {{$history}}
-===================
+        # Show the response
+        print(f"ChatBot: {answer}")
 
-Using the history as context, transform the following question into an absolute question: {{$input}}
-"""
+        self.context["history"] += f"\nUser: {input_text}\nChatBot: {answer}\n"
 
+    def chat(self) -> None:
 
-context = kernel.create_new_context()
-context["url"] = "http://127.0.0.1:8080/search"
-context["history"] = ""
+        while True:
+            print("=============================================")
+            self._chat(input("User: "))
 
-acl_chat_function = kernel.create_semantic_function(acl_chat_prompt, max_tokens=200, temperature=0, top_p=0.5)
-absolute_question_function = kernel.create_semantic_function(absolute_question_prompt, max_tokens=200, temperature=0, top_p=0.5)
+def main():
 
 
-def chat(input_text: str) -> None:
+    parser = ArgumentParser()
 
-    print("---------------------------------------------")
-    absolute_question = absolute_question_function(input_text,context=context)
+    parser.add_argument('--k1', type=float, help='BM25 k1 parameter.')
+    parser.add_argument('--b', type=float, help='BM25 b parameter.')
+    parser.add_argument('--hits', type=int, default=10, help='Number of hits returned by the retriever')
 
-    print (f"Absolute Question: {absolute_question}")
+    args = parser.parse_args()
+    api_key, org_id = sk.openai_settings_from_dot_env()
+    open_ai_config = OpenAIConfig(api_key,org_id)
+    pyserini_config = PyseriniConfig(args.k1, args.b, args.hits)
 
-    # Process the user message and get an answer
-    answer = acl_chat_function(str(absolute_question),context=context)
+    chatbot = ChatBot(pyserini_config=pyserini_config,openai_config=open_ai_config)
+    chatbot.chat()
 
-    # Show the response
-    print(f"ChatBot: {answer}")
 
-    context["history"] += f"\nUser: {input_text}\nChatBot: {answer}\n"
 
 
-while True:
-    print("=============================================")
-    chat(input("User: "))
\ No newline at end of file
+if __name__ == '__main__':
+    main()
\ No newline at end of file
diff --git a/pyserini/demo/aclchatgpt/server.py b/pyserini/demo/aclchatgpt/server.py
deleted file mode 100644
index bca8eed6b..000000000
--- a/pyserini/demo/aclchatgpt/server.py
+++ /dev/null
@@ -1,116 +0,0 @@
-#
-# Pyserini: Reproducible IR research with sparse and dense representations
-#
-# 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.
-#
-
-"""
-This script provides an interactive web interface demo for retrieval on the ACL dataset.
-It requires `flask` (`pip install flask~=2.2.0`).
-An example command looks like `python -m pyserini.demo.acl` that starts up a server on port 8080.
-The demo can be accessed via "http://localhost:8080" in a web browser.
-Additional arguments include:
-    --port [PORT] --hits [Number of hits]
-    --k1 [BM25 k1] --b [BM25 b] --device [cpu, cuda]
-"""
-import json
-import logging
-from argparse import ArgumentParser
-from functools import partial
-from typing import Callable, Optional, Tuple, Union
-
-from flask import Flask, render_template, request, flash, jsonify
-from pyserini.search import LuceneSearcher, FaissSearcher, AutoQueryEncoder
-
-logging.basicConfig(
-    format='%(asctime)s | %(levelname)s | %(name)s | %(message)s',
-    datefmt='%Y-%m-%d %H:%M:%S',
-    level=logging.INFO,
-)
-logger = logging.getLogger('acl-demo')
-
-VERSION = '1.0'
-Searcher = Union[FaissSearcher, LuceneSearcher]
-
-
-def create_app(k: int, load_searcher_fn: Callable[[str], Tuple[Searcher, str]]):
-    app = Flask(__name__)
-
-    lang = 'en'
-    searcher, retriever = load_searcher_fn(lang)
-
-    @app.route('/search', methods=['POST'])
-    def search():
-        nonlocal lang, searcher, retriever
-        query = request.json['query']
-        if not query:
-            search_results = []
-        else:
-            hits = searcher.search(query, k=k)
-            docs = [searcher.doc(hit.docid) for hit in hits]
-            search_results = [
-                {
-                    'rank': r + 1,
-                    'docid': hit.docid,
-                    'doc': docs[r].contents(),
-                    'score': hit.score,
-                }
-                for r, hit in enumerate(hits)
-            ]
-        return search_results
-
-
-    return app
-
-
-def _load_sparse_searcher(language: str, k1: Optional[float]=None, b: Optional[float]=None) -> (Searcher, str):
-    searcher = LuceneSearcher('indexes/lucene-index-acl-paragraph')
-    searcher.set_language(language)
-    if k1 is not None and b is not None:
-        searcher.set_bm25(k1, b)
-        retriever_name = f'BM25 (k1={k1}, b={b})'
-    else:
-        retriever_name = 'BM25'
-
-    return searcher, retriever_name
-
-
-def main():
-    parser = ArgumentParser()
-
-    parser.add_argument('--k1', type=float, help='BM25 k1 parameter.')
-    parser.add_argument('--b', type=float, help='BM25 b parameter.')
-    parser.add_argument('--hits', type=int, default=10, help='Number of hits returned by the retriever')
-    parser.add_argument(
-        '--device',
-        type=str,
-        default='cpu',
-        help='Device to run query encoder, cpu or [cuda:0, cuda:1, ...] (used only when index is based on FAISS)',
-    )
-    parser.add_argument(
-        '--port',
-        default=8080,
-        type=int,
-        help='Web server port',
-    )
-
-    args = parser.parse_args()
-
-    load_fn = partial(_load_sparse_searcher, k1=args.k1, b=args.b)
-
-    app = create_app(args.hits, load_fn)
-    app.run(host='0.0.0.0', port=args.port)
-
-
-if __name__ == '__main__':
-    main()
diff --git a/pyserini/demo/aclchatgpt/skill.py b/pyserini/demo/aclchatgpt/skill.py
index fa27ed9de..52300a86a 100644
--- a/pyserini/demo/aclchatgpt/skill.py
+++ b/pyserini/demo/aclchatgpt/skill.py
@@ -1,13 +1,22 @@
 # Copyright (c) Microsoft. All rights reserved.
 
 import json
+import logging
+from dataclasses import dataclass
 
 import aiohttp
 import requests
 
 from semantic_kernel.orchestration.sk_context import SKContext
+from typing import Callable, Optional, Tuple, Union
 from semantic_kernel.skill_definition import sk_function, sk_function_context_parameter
+from pyserini.search import LuceneSearcher, FaissSearcher, AutoQueryEncoder
 
+@dataclass
+class PyseriniConfig:
+    k1: Optional[float]=None
+    b: Optional[float]=None
+    hits: Optional[int]=1
 
 class PyseriniSkill:
     """
@@ -21,6 +30,18 @@ class PyseriniSkill:
         {{pyserini.search $query}}
     """
 
+
+    def __init__(self,pyserini_config:PyseriniConfig):
+        self.lang = 'en'
+        self.searcher = LuceneSearcher('indexes/lucene-index-acl-paragraph')
+        self.searcher.set_language(self.lang)
+        if pyserini_config.k1 is not None and pyserini_config.b is not None:
+            self.searcher.set_bm25(pyserini_config.k1, pyserini_config.b)
+            self.retriever_name = f'BM25 (k1={pyserini_config.k1}, b={pyserini_config.b})'
+        else:
+            self.retriever_name = 'BM25'
+        self.hits = pyserini_config.hits
+
     @sk_function(description="Searches a corpus of documents using Pyserini using the specified query.", name="search")
     @sk_function_context_parameter(name="url", description="The url of the request")
     async def search(self, query: str, context: SKContext) -> str:
@@ -38,7 +59,18 @@ async def search(self, query: str, context: SKContext) -> str:
         if not url:
             raise ValueError("url cannot be `None` or empty")
 
-        body = {"query": query}
-        result  = requests.post(url, json = body).json()[0]["doc"]
-        return result
+        if not query:
+            search_results = []
+        else:
+            hits = self.searcher.search(query, k=self.hits)
+            docs = [self.searcher.doc(hit.docid) for hit in hits]
+            search_results = [
+                {
+                    'rank': r + 1,
+                    'docid': hit.docid,
+                    'doc': docs[r].contents(),
+                }
+                for r, hit in enumerate(hits)
+            ]
+        return search_results[0]["doc"]
 

From 39070d8095f65c22674b531dc125b2dd88ce20fb Mon Sep 17 00:00:00 2001
From: Bill Cui <bill.cui.work@gmail.com>
Date: Wed, 21 Jun 2023 22:36:19 -0400
Subject: [PATCH 7/7] made chatbot smarter

---
 pyserini/demo/aclchatgpt/chatbot.py | 15 ++++++++-------
 pyserini/demo/aclchatgpt/readme.md  | 11 ++---------
 pyserini/demo/aclchatgpt/skill.py   |  3 +--
 3 files changed, 11 insertions(+), 18 deletions(-)

diff --git a/pyserini/demo/aclchatgpt/chatbot.py b/pyserini/demo/aclchatgpt/chatbot.py
index 6e80cae66..770e59172 100644
--- a/pyserini/demo/aclchatgpt/chatbot.py
+++ b/pyserini/demo/aclchatgpt/chatbot.py
@@ -15,17 +15,16 @@ class OpenAIConfig:
 
 class ChatBot:
 
-    acl_chat_prompt = """
-    Given the `query_results` below, your task is to formulate an answer using only the information 
-    provided in these results. You should not draw from other sources or attempt to provide information that is not 
-    contained within the `query_results`. If the `query_results` are empty, simply state "I'm sorry, but I do not have 
-    enough information to provide an answer."
+    acl_chat_prompt = """Given the query_results below, your task is to formulate an answer. You may choose to use or not to use the 
+    information in the query_results.
+    If you use the query_results, you must reference the docid of the document used by appending to the answer with "(docid: doc-id-here)".
+    If you do not use the query_results, do not reference it in your answer.
     
     ===================
     query_results: {{pyserini.search $input}}
     ===================
     
-    Based on the above `query_results`, what is your response to {{$input}}?
+    What is your response to {{$input}}?
     """
 
     absolute_question_prompt = """
@@ -74,6 +73,7 @@ def _chat(self,input_text: str) -> None:
         self.context["history"] += f"\nUser: {input_text}\nChatBot: {answer}\n"
 
     def chat(self) -> None:
+        print("Hi, I'm the ACL ChatBot. Ask me a question about ACL Anthology papers and I'll do my best to answer it.")
 
         while True:
             print("=============================================")
@@ -81,7 +81,6 @@ def chat(self) -> None:
 
 def main():
 
-
     parser = ArgumentParser()
 
     parser.add_argument('--k1', type=float, help='BM25 k1 parameter.')
@@ -93,6 +92,8 @@ def main():
     open_ai_config = OpenAIConfig(api_key,org_id)
     pyserini_config = PyseriniConfig(args.k1, args.b, args.hits)
 
+    print("Starting ChatBot...")
+
     chatbot = ChatBot(pyserini_config=pyserini_config,openai_config=open_ai_config)
     chatbot.chat()
 
diff --git a/pyserini/demo/aclchatgpt/readme.md b/pyserini/demo/aclchatgpt/readme.md
index df9c7cbb4..0f3a5809a 100644
--- a/pyserini/demo/aclchatgpt/readme.md
+++ b/pyserini/demo/aclchatgpt/readme.md
@@ -2,12 +2,6 @@
 
 By default, GPT-3 text-davinci-003 is used. You can change this in `chatbot.py`.
 
-`chatbot.py` is the front-end script. As a CLI, the user can input a query and the chatbot will return a response.
-
-`server.py` is the back-end script. It is a Flask server that will return a response to a query sent to the server.
-
-You must have both running for `chatbot.py` to work.
-
 ## Environment Varaibles
 You will need to create an `.env` file in directory that you are running in.
 The `.env` file should contain the following:
@@ -22,6 +16,5 @@ OPENAI_ORG_ID="<your org id>"
 2. Copy the generated `lucene-index-acl-paragraph` index from the `acl-anthology` folder to `pyserini/indexes`
 3. You will need Semantic Kernel as well. 
    - `pip3 install --upgrade semantic-kernel`
-4. Start the server with `python -m pyserini.demo.aclchatgpt.server`
-5. Start the chatbot with `python -m pyserini.demo.aclchatgpt.chatbot`
-6. Start chatting!
\ No newline at end of file
+4. Start the chatbot with `python -m pyserini.demo.aclchatgpt.chatbot`
+5. Start chatting!
\ No newline at end of file
diff --git a/pyserini/demo/aclchatgpt/skill.py b/pyserini/demo/aclchatgpt/skill.py
index 52300a86a..3209d0373 100644
--- a/pyserini/demo/aclchatgpt/skill.py
+++ b/pyserini/demo/aclchatgpt/skill.py
@@ -30,7 +30,6 @@ class PyseriniSkill:
         {{pyserini.search $query}}
     """
 
-
     def __init__(self,pyserini_config:PyseriniConfig):
         self.lang = 'en'
         self.searcher = LuceneSearcher('indexes/lucene-index-acl-paragraph')
@@ -72,5 +71,5 @@ async def search(self, query: str, context: SKContext) -> str:
                 }
                 for r, hit in enumerate(hits)
             ]
-        return search_results[0]["doc"]
+        return "docid:" + search_results[0]["docid"] + ",doc:" + search_results[0]["doc"]