Update README.md

.env_template

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

@@ -52,7 +52,7 @@ class NLG:
 
     def llm_substitute_product(character, user_message):
 
-        input = st.session_state.characters_dict['task_substitute'] + f'"{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)

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

@@ -12,7 +12,9 @@ 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
 
@@ -28,11 +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)}
 
@@ -42,7 +39,7 @@ 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')
@@ -60,8 +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:
@@ -79,39 +74,48 @@ if __name__ == '__main__':
         with open(CHARACTERS_DICT) as f:
             st.session_state.characters_dict = json.load(f)
         
-def show_graph():
+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 graph():
     # Create a graphlib graph object
     if st.session_state.generated:
-        user, chatbot = [], []
+        system = [utterance for utterance in st.session_state.generated][-3:]
-        graph = graphviz.Digraph()
+        user = [utterance for utterance in st.session_state.past][-2:]
-        chatbot = copy.deepcopy(st.session_state.generated)
+        graph = ""
-        user = copy.deepcopy(st.session_state.past)
+        for i, utterance in enumerate(system):
-        for x in range(len(user)):
+            utterance = utterance.strip('\n')
-            chatbot_text = [word + '\n' if i % 5 == 0 and i > 0 else word for i, word in enumerate(chatbot[x].split(' '))]
+            utterance = " ".join([word + '<br>' if i % 5 == 0 and i > 0 else word for i, word in enumerate(utterance.split(" "))])
-            user_text    = [word + '\n' if i % 5 == 0 and i > 0 else word for i, word in enumerate(user[x].split(' '))]
+            utterance = utterance.replace('\"', '')
-            graph.edge(' '.join(chatbot_text), ' '.join(user_text))
+            if i < len(user):
-        try:
+                user[i] = user[i].strip('\n')
-            graph.edge(' '.join(user_text), ' '.join([word + '\n' if i % 5 == 0 and i > 0 else word for i, word in enumerate(chatbot[x + 1].split(' '))]))
+                user[i] = user[i].replace('\"', '')
-        except:
+                user[i] = " ".join([word + '<br>' if i % 5 == 0 and i > 0 else word for i, word in enumerate(user[i].split(' '))])
-            pass
+                graph += f"{string.ascii_uppercase[i]}(\"{utterance}\") --> |{user[i]}| {string.ascii_uppercase[i+1]};"
-        st.graphviz_chart(graph)
+            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(
+        character_type = c1.selectbox(label=st.session_state.locale.select_placeholder2, key="role",
-            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:
-            character_type = c2.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':
@@ -119,15 +123,13 @@ def main() -> None:
         else:
             st.session_state.dp.llm_rephrasing = True
         
-        elif role_kind == st.session_state.locale.radio_text2: 
-            c2.text_input(label=st.session_state.locale.select_placeholder3, key="role")
-        
     get_user_input()
     show_chat_buttons()
     
     show_conversation()
     with st.sidebar:
-        show_graph()
+        mermaid(graph())
+        #show_graph()
 
 
 if __name__ == "__main__":

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')

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

@@ -1,6 +1,6 @@
 {
-  "task_paraphrase": "You're currently reading a step of a recipe, paraphrese it so that it matches your charater: ",
+  "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: ", 
+  "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": {
@@ -10,7 +10,7 @@
     },
     "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."
+      "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": {