Update README.md

.env_template

@@ -1,5 +1,7 @@
-RECIPE_PATH=AMUseBotFront/ai_talks/AMUseBotBackend/recipe/ 
-DIALOG_PATH=AMUseBotFront/ai_talks/AMUseBotBackend/dialog/
-INTENT_DICT_PATH=ai_talks/AMUseBotBackend/utils/intent_dict.json
-MODEL_IDENTIFIER_PATH=ai_talks/AMUseBotBackend/models/NLU/roberta-base-cookdial.txt
-INGREDIENTS_RECIPES_MERGED=
+RECIPE_PATH=recipe/
+DIALOG_PATH=dialog/
+INTENT_DICT_PATH=intent_dict.json
+MODEL_IDENTIFIER_PATH=roberta-base-cookdial-v1_1.txt
+INGREDIENTS_RECIPES_MERGED=ingredients_recipes_merged.csv
+CHARACTERS_DICT=characters_dict.json
+API_KEY=

README.md

@@ -0,0 +1,36 @@
+# Cooking taskbot project
+
+## Run system
+
+#### With Conda
+    
+    conda create -n "my_env" python=3.9.12 ipython
+    conda activate my_env
+    pip install -r requirements.txt
+    streamlit run ai_talks\chat.py
+
+After running system, model saves in dir:
+
+Linux
+
+    ~/.cache/huggingface/transformers
+
+Windows
+
+    C:\Users\username\.cache\huggingface\transformers
+
+To use the purely experimental generative features, for now, an OpenAI API key is needed. Insert it into the following file:
+
+    AMUseBot/.env_template
+    
+## Requirements
+
+    Python 3.9.12
+
+## Dataset
+
+[YiweiJiang2015/CookDial](https://github.com/YiweiJiang2015/CookDial)
+
+## NLU model HF repo
+
+[kedudzic/roberta-base-cookdial](https://huggingface.co/AMUseBot/roberta-base-cookdial-v1_1)

ai_talks/AMUseBotBackend/src/DP/dp.py

@@ -4,14 +4,16 @@ from AMUseBotBackend.src.NLG.nlg import NLG
 from AMUseBotBackend.src.tools.search import search_recipe
 
 import AMUseBotBackend.consts as c
+import json
 
+import streamlit as st
 
 class DP:
 
-    def __init__(self, dst: DST):
+    def __init__(self, dst: DST, llm_rephrasing=True, character='default'): #TODO: a way to set llm_rephrasing status and a character
         self.dst_module = dst
-
-
+        self.llm_rephrasing = llm_rephrasing
+        self.character = character
 
     def generate_response(self, intents: List[str]) -> str:
 
@@ -31,8 +33,13 @@ class DP:
                 if found_recipe:
                     recipe_name = self.dst_module.set_recipe(found_recipe)
                     self.dst_module.set_next_step()
-                    return NLG.MESSAGE_CHOOSEN_RECIPE(recipe_name=recipe_name) + "\n" \
-                        + self.dst_module.generate_state(c.STEPS_KEY)[self.dst_module.generate_state(c.CURR_STEP_KEY)]
+                    if self.llm_rephrasing:
+                        return NLG.MESSAGE_CHOOSEN_RECIPE(recipe_name=recipe_name) + "\n" \
+                            + NLG.llm_rephrase_recipe(self.character, self.dst_module.generate_state(c.STEPS_KEY)[self.dst_module.generate_state(c.CURR_STEP_KEY)])
+                    else:
+                        return NLG.MESSAGE_CHOOSEN_RECIPE(recipe_name=recipe_name) + "\n" \
+                            + self.dst_module.generate_state(c.STEPS_KEY)[self.dst_module.generate_state(c.CURR_STEP_KEY)]
+
                 if not found_recipe:
                     return NLG.MESSAGE_NOT_UNDERSTAND_SUGGEST_RECIPE(self.dst_module.get_random_recipes(3))
             # not understand ask recipe
@@ -41,6 +48,8 @@ class DP:
         # Recipe choosen
         if (None != self.dst_module.generate_state(c.RECIPE_ID_KEY) and "" != self.dst_module.generate_state(
                 c.RECIPE_ID_KEY)):
+            if ("req_substitute" in intents):
+                return NLG.llm_substitute_product(self.character, self.dst_module.generate_state(c.DIALOG_HISTORY_KEY)[-1][c.USER_MESSAGE_KEY])
             if ("req_ingredient_list" in intents
                     or "req_ingredient" in intents):
                 return NLG.MESSAGE_INGREDIENTS(self.dst_module.generate_state(c.INGREDIENTS_KEY))
@@ -51,7 +60,10 @@ class DP:
                     or "req_instruction" in intents):
                 next_step = self.dst_module.set_next_step()
                 if (next_step):
-                    return self.dst_module.generate_state(c.STEPS_KEY)[self.dst_module.generate_state(c.CURR_STEP_KEY)]
+                    if self.llm_rephrasing:
+                        return NLG.llm_rephrase_recipe(self.character, self.dst_module.generate_state(c.STEPS_KEY)[self.dst_module.generate_state(c.CURR_STEP_KEY)])
+                    else:
+                        return self.dst_module.generate_state(c.STEPS_KEY)[self.dst_module.generate_state(c.CURR_STEP_KEY)]
                 if (not next_step):
                     self.dst_module.restart()
                     return NLG.RECIPE_OVER_ANSWER

ai_talks/AMUseBotBackend/src/DST/dst.py

@@ -77,19 +77,21 @@ class DST:
         return [self.recipes[id] for id in recipes_id]
 
     def __set_steps(self):
-        dialog_files = [] 
+        dialog_files = []
         steps = {}
-        for (_, _, filenames) in walk(self.__dialog_path):
+        for (_, _, filenames) in walk(self.__recipe_path):
             dialog_files.extend(filenames)
             break
         for dialog_title in dialog_files:
             if dialog_title.startswith(f"{self.__recipe_id:03d}"):
-                with open(self.__dialog_path + "/" + dialog_title) as f:
+                with open(self.__recipe_path + dialog_title) as f:
                     data = json.load(f)
-                    for message in data["messages"]:
-                        if "inform_instruction" in message["annotations"]:
-                            steps[len(steps)] = message["utterance"]
+                    for row in data['content']:
+                        if row['type']=='instruction':
+                            steps[len(steps)] = row['text'].split(maxsplit=1)[1]
+
         self.__steps = steps
+
     
     def __set_ingredients(self):
         dialog_files = []

ai_talks/AMUseBotBackend/src/NLG/nlg.py

@@ -1,3 +1,4 @@
+import streamlit as st
 class NLG:
     MESSAGE_PROMPT = "Hello! I'm AMUseBot, a virtual cooking assistant. Please tell me the name of the dish that you'd like to prepare today."
     MESSAGE_HI   = "Hi! What do you want to make today?"
@@ -5,6 +6,7 @@ class NLG:
     BYE_ANSWER    = "Bye, hope to see you soon!"
     RECIPE_OVER_ANSWER    = "Congratulations! You finished preparing the dish, bon appetit!"
     NOT_UNDERSTAND_ANSWER = "I'm sorry, I don't understand. Could you rephrase?"
+    CANNOT_HELP_ANSWER = "I'm sorry I can't help you with that."
 
     @staticmethod
     def MESSAGE_INGREDIENTS(ingr_list):
@@ -24,3 +26,38 @@ class NLG:
             suggestions = ", ".join(recipes_list[0:-1]) + f" or {recipes_list[-1]}"
 
         return f"I'm sorry, I don't know a recipe like that. Instead, I can suggest you {suggestions}."
+
+    
+    def llm_create_response(character, input):
+        model = st.session_state.characters_dict['model']
+        prompt = st.session_state.characters_dict['characters'][character]['prompt']
+
+        message = [{'role': 'system', 'content': prompt}, {'role': 'user', 'content': input}]
+
+        response = st.session_state.openai.ChatCompletion.create(
+            model=model, messages=message, temperature=1, max_tokens=128
+        )
+        rephrased_response = response.choices[0].message.content
+
+        return rephrased_response
+    
+    def llm_rephrase_recipe(character, response):
+
+        input = st.session_state.characters_dict['task_paraphrase'] + f'"{response}".' + st.session_state.characters_dict['characters'][character]['task_specification']
+        try:
+            return NLG.llm_create_response(character, input)
+        except:
+            print('OpenAI API call failed during response paraphrasing! Returning input response')
+            return response
+
+    def llm_substitute_product(character, user_message):
+
+        input = st.session_state.characters_dict['task_substitute'] + f'"{user_message}".' + st.session_state.characters_dict['characters'][character]['task_specification']
+
+        try:
+            return NLG.llm_create_response(character, input)
+        except:
+            print('OpenAI API call failed during response paraphrasing! Returning input response')
+            return NLG.CANNOT_HELP_ANSWER
+
+

ai_talks/AMUseBotBackend/src/tools/search.py

@@ -7,7 +7,7 @@ from rank_bm25 import BM25Okapi
 import os
 from dotenv import load_dotenv
 
-load_dotenv()
+load_dotenv('.env_template')
 
 INGREDIENTS_RECIPES_MERGED = os.getenv('INGREDIENTS_RECIPES_MERGED')  
 

ai_talks/chat.py

@@ -9,7 +9,12 @@ import streamlit as st
 from PIL import Image
 from src.utils.conversation import get_user_input, show_chat_buttons, show_conversation
 from src.utils.lang import en
-
+import openai
+import copy
+import json
+import string
+import streamlit.components.v1 as components
+import re
 import os
 from dotenv import load_dotenv
 
@@ -25,11 +30,6 @@ if __name__ == '__main__':
     favicon: Path = icons_dir / "favicons/0.png"
     # --- GENERAL SETTINGS ---
     LANG_PL: str = "Pl"
-    AI_MODEL_OPTIONS: list[str] = [
-        "gpt-3.5-turbo",
-        "gpt-4",
-        "gpt-4-32k",
-    ]
 
     CONFIG = {"page_title": "AMUsebot", "page_icon": Image.open(favicon)}
 
@@ -39,10 +39,12 @@ if __name__ == '__main__':
     with open(css_file) as f:
         st.markdown(f"<style>{f.read()}</style>", unsafe_allow_html=True)
 
-    load_dotenv()
+    load_dotenv('.env_template')
 
     DIALOG_PATH = os.getenv('DIALOG_PATH')
     RECIPE_PATH = os.getenv('RECIPE_PATH')
+    CHARACTERS_DICT = os.getenv('CHARACTERS_DICT')
+    API_KEY = os.getenv('API_KEY')
 
     # Storing The Context
     if "locale" not in st.session_state:
@@ -55,8 +57,6 @@ if __name__ == '__main__':
         st.session_state.messages = []
     if "user_text" not in st.session_state:
         st.session_state.user_text = ""
-    if "input_kind" not in st.session_state:
-        st.session_state.input_kind = st.session_state.locale.input_kind_1
     if "seed" not in st.session_state:
         st.session_state.seed = randrange(10 ** 3)  # noqa: S311
     if "costs" not in st.session_state:
@@ -67,51 +67,69 @@ if __name__ == '__main__':
         st.session_state.dst = DST(recipe_path=RECIPE_PATH, dialog_path=DIALOG_PATH)
     if "dp" not in st.session_state:
         st.session_state.dp = DP(dst=st.session_state.dst)
+    if "openai" not in st.session_state:
+        st.session_state.openai = openai
+        st.session_state.openai.api_key = API_KEY
+    if "characters_dict" not in st.session_state:
+        with open(CHARACTERS_DICT) as f:
+            st.session_state.characters_dict = json.load(f)
+        
+def mermaid(code: str) -> None:
+    components.html(
+        f"""
+        <pre class="mermaid">
+            %%{{init: {{'themeVariables': {{ 'edgeLabelBackground': 'transparent'}}}}}}%%
+            flowchart TD;
+            {code}
+            linkStyle default fill:white,color:white,stroke-width:2px,background-color:lime;
+        </pre>
+        <script type="module">
+            import mermaid from 'https://cdn.jsdelivr.net/npm/mermaid@10/dist/mermaid.esm.min.mjs';
+            mermaid.initialize({{ startOnLoad: true }});
+        </script>
+        """, height=1000
+    )
 
-def show_graph():
+def graph():
     # Create a graphlib graph object
     if st.session_state.generated:
-        user, chatbot = [], []
-        graph = graphviz.Digraph()
-        for i in range(len(st.session_state.past)):
-            chatbot.append(st.session_state.generated[i])
-            user.append(st.session_state.past[i])
-        for x in range(len(user)):
-            chatbot_text = [word + '\n' if i % 5 == 0 and i > 0 else word for i, word in enumerate(st.session_state.generated[x].split(' '))]
-            user_text    = [word + '\n' if i % 5 == 0 and i > 0 else word for i, word in enumerate(st.session_state.past[x].split(' '))]
-            graph.edge(' '.join(chatbot_text), ' '.join(user_text))
-            try:
-                graph.edge(' '.join(user_text), ' '.join([word + '\n' if i % 5 == 0 and i > 0 else word for i, word in enumerate(st.session_state.generated[x + 1].split(' '))]))
-            except:
-                pass
-        st.graphviz_chart(graph)
+        system = [utterance for utterance in st.session_state.generated][-3:]
+        user = [utterance for utterance in st.session_state.past][-2:]
+        graph = ""
+        for i, utterance in enumerate(system):
+            utterance = utterance.strip('\n')
+            utterance = " ".join([word + '<br>' if i % 5 == 0 and i > 0 else word for i, word in enumerate(utterance.split(" "))])
+            utterance = utterance.replace('\"', '')
+            if i < len(user):
+                user[i] = user[i].strip('\n')
+                user[i] = user[i].replace('\"', '')
+                user[i] = " ".join([word + '<br>' if i % 5 == 0 and i > 0 else word for i, word in enumerate(user[i].split(' '))])
+                graph += f"{string.ascii_uppercase[i]}(\"{utterance}\") --> |{user[i]}| {string.ascii_uppercase[i+1]};"
+            else:
+                graph += f"{string.ascii_uppercase[i]}(\"{utterance}\") --> {string.ascii_uppercase[i+1]}(...);style {string.ascii_uppercase[i+1]} fill:none,color:white;"
+    graph = graph.replace('\n', ' ')#replace(')','').replace('(','')
+    #print(graph)
+    return graph
 
 
 def main() -> None:
     c1, c2 = st.columns(2)
     with c1, c2:
-        st.session_state.input_kind = c2.radio(
-            label=st.session_state.locale.input_kind,
-            options=(st.session_state.locale.input_kind_1, st.session_state.locale.input_kind_2),
-            horizontal=True,
-        )
-        role_kind = c1.radio(
-            label=st.session_state.locale.radio_placeholder,
-            options=(st.session_state.locale.radio_text1, st.session_state.locale.radio_text2),
-            horizontal=True,
-        )
-        if role_kind == st.session_state.locale.radio_text1:
-            c2.selectbox(label=st.session_state.locale.select_placeholder2, key="role",
-                             options=st.session_state.locale.ai_role_options)
-        elif role_kind == st.session_state.locale.radio_text2: 
-            c2.text_input(label=st.session_state.locale.select_placeholder3, key="role")
+        character_type = c1.selectbox(label=st.session_state.locale.select_placeholder2, key="role",
+                         options=st.session_state.locale.ai_role_options)
+        st.session_state.dp.character = character_type
+        if character_type == 'default':
+            st.session_state.dp.llm_rephrasing = False
+        else:
+            st.session_state.dp.llm_rephrasing = True
         
     get_user_input()
     show_chat_buttons()
     
     show_conversation()
     with st.sidebar:
-        show_graph()
+        mermaid(graph())
+        #show_graph()
 
 
 if __name__ == "__main__":

ai_talks/src/utils/constants.py

@@ -1,16 +1,6 @@
 AI_ROLE_OPTIONS_EN: list[str] = [
-    "helpful assistant",
-    "code assistant",
-    "code reviewer",
-    "text improver",
-    "cinema expert",
-    "sport expert",
-    "online games expert",
-    "food recipes expert",
-    "English grammar expert",
-    "friendly and helpful teaching assistant",
-    "laconic assistant",
-    "helpful, pattern-following assistant",
-    "translate corporate jargon into plain English",
+    "default",
+    "helpful_chef",
+    "ramsay",
 ]
 

ai_talks/src/utils/conversation.py

@@ -12,7 +12,7 @@ from AMUseBotBackend.src.NLU.nlu import NLU
 import os
 from dotenv import load_dotenv
 
-load_dotenv()
+load_dotenv('.env_template')
 
 INTENT_DICT_PATH = os.getenv('INTENT_DICT_PATH')
 MODEL_IDENTIFIER_PATH = os.getenv('MODEL_IDENTIFIER_PATH')
@@ -23,7 +23,8 @@ def get_nlu_model(intent_dict_path = INTENT_DICT_PATH, model_identifier_path = M
                             model_identifier_path=model_identifier_path)
 
 def clear_chat() -> None:
-    st.session_state.generated = []
+    st.session_state.generated = ["Hello! I'm AMUseBot, a virtual cooking assistant. Please tell me the name of the dish that you'd like to prepare today."]
+    st.session_state.dst.restart()
     st.session_state.past = []
     st.session_state.messages = []
     st.session_state.user_text = ""

ai_talks/src/utils/lang.py

@@ -20,15 +20,8 @@ class Locale:
     chat_clear_btn: str
     chat_save_btn: str
     speak_btn: str
-    input_kind: str
-    input_kind_1: str
-    input_kind_2: str
     select_placeholder1: str
     select_placeholder2: str
-    select_placeholder3: str
-    radio_placeholder: str
-    radio_text1: str
-    radio_text2: str
     stt_placeholder: str
     footer_title: str
     footer_option0: str
@@ -55,15 +48,8 @@ en = Locale(
     chat_clear_btn="Clear",
     chat_save_btn="Save",
     speak_btn="Push to Speak",
-    input_kind="Input Kind",
-    input_kind_1="Text",
-    input_kind_2="Voice [test mode]",
     select_placeholder1="Select Model",
     select_placeholder2="Select Role",
-    select_placeholder3="Create Role",
-    radio_placeholder="Role Interaction",
-    radio_text1="Select",
-    radio_text2="Create",
     stt_placeholder="To Hear The Voice Of AI Press Play",
     footer_title="Support & Feedback",
     footer_option0="Chat",

characters_dict.json

@@ -0,0 +1,21 @@
+{
+  "task_paraphrase": "You're currently reading a step of a recipe, paraphrase it so that it matches your character: ",
+  "task_substitute": "A user has just asked for a substitute for a missing ingredient, answer him according to your character in one short sentence with at most 3 alternatives: ",
+  "model": "gpt-3.5-turbo-0613", 
+  "characters": {
+    "default": {
+      "prompt": "",
+      "task_specification": ""
+
+    },
+    "helpful_chef": {
+      "prompt": "You're a master chef known for treating everyone like your equal. ",
+      "task_specification": " Give your answer as a natural sounding, full English sentence. Keep the sentence length similar and do not make the language flowery."
+
+    },
+    "ramsay": {
+      "prompt": "You're Gordon Ramsay, a famous British chef known for his short temper and routinely insulting people. ",
+      "task_specification": ""
+    }
+  }
+}