first commit openai simulators

.gitignore

@@ -0,0 +1,192 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+env/
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+.pytest_cache
+test-reports/
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/#use-with-ide
+.pdm.toml
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/
+
+# Database
+*.db
+*.rdb
+
+# Pycharm
+.idea
+
+# VS Code
+.vscode/
+
+# Spyder
+.spyproject/
+
+# Jupyter NB Checkpoints
+.ipynb_checkpoints/
+
+# exclude data from source control by default
+/data/
+
+# Mac OS-specific storage files
+.DS_Store
+
+# vim
+*.swp
+*.swo
+
+# Mypy cache
+.mypy_cache/

README.md

@@ -5,7 +5,7 @@ colorFrom: pink
 colorTo: yellow
 sdk: streamlit
 sdk_version: 1.26.0
-app_file: app.py
+app_file: convosim.py
 pinned: false
 ---
 

convosim.py

@@ -0,0 +1,38 @@
+import openai 
+import os
+import streamlit as st
+from langchain.schema.messages import HumanMessage
+
+from utils import create_memory_add_initial_message, clear_memory, get_chain
+
+openai_api_key = os.environ['OPENAI_API_KEY'] 
+memories = ['memory']
+
+with st.sidebar:
+    temperature = st.slider("Temperature", 0., 1., value=0.8, step=0.1)
+    issue = st.selectbox("Select an Issue", ['Anxiety','Suicide'], index=0,
+                            on_change=clear_memory, args=(memories,)
+                        )
+    supported_languages = ['English', "Spanish"] if issue == "Anxiety" else ['English']
+    language = st.selectbox("Select a Language", supported_languages, index=0,
+                            on_change=clear_memory, args=(memories,)
+                        )
+                            
+    source = st.selectbox("Select a source Model A", ['OpenAI GPT3.5','Finetuned OpenAI'], index=1,
+                            on_change=clear_memory, args=(memories,)
+                        )
+
+create_memory_add_initial_message(memories, language)
+llm_chain = get_chain(issue, language, source, st.session_state[memories[0]], temperature)
+
+st.title("💬 Simulator") 
+
+for msg in st.session_state[memories[0]].buffer_as_messages:
+    role = "user" if type(msg) == HumanMessage else "assistant"
+    st.chat_message(role).write(msg.content)
+
+if prompt := st.chat_input():
+    st.chat_message("user").write(prompt)
+    response = llm_chain.predict(input=prompt, stop="helper:")
+    # response = update_memory_completion(prompt, st.session_state["memory"], OA_engine, temperature)
+    st.chat_message("assistant").write(response)

models/custom_parsers.py

@@ -0,0 +1,16 @@
+from typing import List
+from langchain.schema import BaseOutputParser
+
+class CustomStringOutputParser(BaseOutputParser[List[str]]):
+    """Parse the output of an LLM call to a list."""
+
+    @property
+    def _type(self) -> str:
+        return "str"
+
+    def parse(self, text: str) -> str:
+        """Parse the output of an LLM call."""
+        text = text.split("texter:")[0]
+        text = text.rstrip("\n")
+        text = text.strip()
+        return text

models/openai/finetuned_models.py

@@ -0,0 +1,75 @@
+import openai
+from models.custom_parsers import CustomStringOutputParser
+from langchain.chains import LLMChain
+from langchain.llms import OpenAI
+from langchain.prompts import PromptTemplate
+import logging
+
+finetuned_models = {
+    # "olivia_babbage_engine": "babbage:ft-crisis-text-line:exp-olivia-babbage-2023-02-23-19-57-19",
+    "Anxiety-English": "curie:ft-crisis-text-line:exp-olivia-curie-2-2023-02-24-00-25-13",
+    # "olivia_davinci_engine": "davinci:ft-crisis-text-line:exp-olivia-davinci-2023-02-24-00-02-41",
+    # "olivia_augmented_babbage_engine": "babbage:ft-crisis-text-line:exp-olivia-augmented-babbage-2023-02-24-18-35-42",
+    # "Olivia-Augmented": "curie:ft-crisis-text-line:exp-olivia-augmented-curie-2023-02-24-20-13-33",
+    # "olivia_augmented_davinci_engine": "davinci:ft-crisis-text-line:exp-olivia-augmented-davinci-2023-02-24-23-57-08",
+    # "kit_babbage_engine": "babbage:ft-crisis-text-line:exp-kit-babbage-2023-03-06-21-34-10",
+    # "kit_curie_engine": "curie:ft-crisis-text-line:exp-kit-curie-2023-03-06-22-01-29",
+    "Suicide-English": "curie:ft-crisis-text-line:exp-kit-curie-2-2023-03-08-16-26-48",
+    # "kit_davinci_engine": "davinci:ft-crisis-text-line:exp-kit-davinci-2023-03-06-23-09-15",
+    # "olivia_es_davinci_engine": "davinci:ft-crisis-text-line:es-olivia-davinci-2023-04-25-17-07-44",
+    "Anxiety-Spanish": "curie:ft-crisis-text-line:es-olivia-curie-2023-04-27-15-02-42",
+    # "olivia_curie_engine": "curie:ft-crisis-text-line:exp-olivia-curie-2-2023-02-24-00-25-13",
+    # "Oscar-Spanish": "curie:ft-crisis-text-line:es-oscar-curie-2023-05-03-21-55-06",
+    # "oscar_es_davinci_engine": "davinci:ft-crisis-text-line:es-oscar-davinci-2023-05-03-21-39-29",
+}
+
+# def generate_next_response(completion_engine, context, temperature=0.8):
+    
+#     completion = openai.Completion.create(
+#         engine=completion_engine, 
+#         prompt=context,
+#         temperature=temperature,
+#         max_tokens=150,
+#         stop="helper:"
+#     )
+    
+#     completion_text = completion['choices'][0]['text']
+
+#     return completion_text
+
+# def update_memory_completion(helper_input, memory, OA_engine, temperature=0.8):
+#     memory.chat_memory.add_user_message(helper_input)
+#     context = "## BEGIN ## \n" + memory.load_memory_variables({})['history'] + "\ntexter:"
+#     print(context)
+#     response = generate_next_response(OA_engine, context, temperature).strip().replace("\n","")
+#     response = response.split("texter:")[0]
+#     memory.chat_memory.add_ai_message(response)
+#     return response
+
+
+def get_finetuned_chain(model_name, memory, temperature=0.8):
+
+    _TEXTER_TEMPLATE_ = """The following is a friendly conversation between a volunter and a person in crisis;
+    Current conversation:
+    {history}
+    helper: {input}
+    texter:"""
+
+    PROMPT = PromptTemplate(
+        input_variables=['history', 'input'],
+        template=_TEXTER_TEMPLATE_
+    )
+    llm = OpenAI(
+        temperature=temperature, 
+        model=model_name,
+        max_tokens=150,
+    )
+    llm_chain = LLMChain(
+        llm=llm,
+        prompt=PROMPT,
+        verbose=True,
+        memory=memory,
+        output_parser = CustomStringOutputParser()
+    )
+    logging.debug(f"loaded fine tuned model {model_name}")
+    return llm_chain

models/openai/role_models.py

@@ -0,0 +1,57 @@
+import logging
+from models.custom_parsers import CustomStringOutputParser
+from langchain.chains import ConversationChain
+from langchain.llms import OpenAI
+from langchain.prompts import PromptTemplate
+
+
+_ANXIETY_TEMPLATE_ = """The following is a friendly conversation between a volunter and a person in crisis;
+Your name is Olivia and you are a person in crisis because you are experiencing an anxiety attack.
+Do not disclose your name unless asked.
+Current conversation:
+{history}
+helper: {input}
+texter:"""
+
+_ANSIEDAD_TEMPLATE_ = """La siguiente es una conversacion entre un voluntario  y una persona en crisis;
+Tu nombre es Olivia y eres una persona en crisis porque experimentas un episodio de ansiedad.
+No reveles tu nombre a menos que te lo pregunten
+Current conversation:
+{history}
+helper: {input}
+texter:"""
+
+_SUICIDE_TEMPLATE_ = """The following is a friendly conversation between a volunter and a person in crisis;
+Your name is Kit and you are a person in crisis because you are experiencing suicidal thoughts.
+Do not disclose your name unless asked.
+Current conversation:
+{history}
+helper: {input}
+texter:"""
+
+role_templates = {
+    'Anxiety-English': _ANXIETY_TEMPLATE_,
+    'Anxiety-Spanish': _ANSIEDAD_TEMPLATE_,
+    'Suicide-English': _SUICIDE_TEMPLATE_,
+}
+
+
+def get_role_chain(template, memory, temperature=0.8):
+
+    PROMPT = PromptTemplate(
+        input_variables=['history', 'input'],
+        template=template
+    )
+    llm = OpenAI(
+        temperature=temperature, 
+        max_tokens=150,
+    )
+    llm_chain = ConversationChain(
+        llm=llm,
+        prompt=PROMPT,
+        verbose=True,
+        memory=memory,
+        output_parser=CustomStringOutputParser()
+    )
+    logging.debug(f"loaded GPT3.5 model")
+    return llm_chain

pages/comparisor.py

@@ -0,0 +1,127 @@
+import openai 
+import os
+import streamlit as st
+from langchain.schema.messages import HumanMessage
+import logging
+
+from utils import create_memory_add_initial_message, clear_memory, get_chain
+
+openai_api_key = os.environ['OPENAI_API_KEY'] 
+memories = ['memoryA', 'memoryB', 'commonMemory']
+
+
+def delete_last_message(memory):
+    last_prompt = memory.chat_memory.messages[-2].content
+    memory.chat_memory.messages = memory.chat_memory.messages[:-2]
+    return last_prompt
+
+def replace_last_message(memory, new_message):
+    memory.chat_memory.messages = memory.chat_memory.messages[:-1]
+    memory.chat_memory.add_ai_message(new_message)
+
+def regenerateA():
+    last_prompt = delete_last_message(st.session_state[memories[0]])
+    new_response = llm_chainA.predict(input=last_prompt, stop="helper:")
+    col1.chat_message("user").write(last_prompt)
+    col1.chat_message("assistant").write(new_response)
+
+def regenerateB():
+    last_prompt = delete_last_message(st.session_state[memories[1]])
+    new_response = llm_chainB.predict(input=last_prompt, stop="helper:")
+    col2.chat_message("user").write(last_prompt)
+    col2.chat_message("assistant").write(new_response)
+
+def replaceA():
+    last_prompt = st.session_state[memories[1]].chat_memory.messages[-2].content
+    new_message = st.session_state[memories[1]].chat_memory.messages[-1].content
+    replace_last_message(st.session_state[memories[0]], new_message)
+    st.session_state['commonMemory'].save_context({"inputs":last_prompt}, {"outputs":new_message})
+
+def replaceB():
+    last_prompt = st.session_state[memories[0]].chat_memory.messages[-2].content
+    new_message = st.session_state[memories[0]].chat_memory.messages[-1].content
+    replace_last_message(st.session_state[memories[1]], new_message)
+    st.session_state['commonMemory'].save_context({"inputs":last_prompt}, {"outputs":new_message})
+
+def regenerateBoth():
+    regenerateA()
+    regenerateB()
+
+def bothGood():
+    if len(st.session_state['memoryA'].buffer_as_messages) == 1:
+        pass
+    else:
+        last_prompt = st.session_state[memories[0]].chat_memory.messages[-2].content
+        last_reponse = st.session_state[memories[0]].chat_memory.messages[-1].content
+        st.session_state['commonMemory'].save_context({"inputs":last_prompt}, {"outputs":last_reponse})
+
+with st.sidebar:
+    issue = st.selectbox("Select an Issue", ['Anxiety','Suicide'], index=0,
+                            on_change=clear_memory, args=(memories,)
+                        )
+    supported_languages = ['English', "Spanish"] if issue == "Anxiety" else ['English']
+    language = st.selectbox("Select a Language", supported_languages, index=0,
+                            on_change=clear_memory, args=(memories,)
+                        )
+    
+    with st.expander("Model A"):
+        temperatureA = st.slider("Temperature Model A", 0., 1., value=0.8, step=0.1)
+        sourceA = st.selectbox("Select a source Model A", ['OpenAI GPT3.5','Finetuned OpenAI'], index=0,
+                                on_change=clear_memory, args=(memories,)
+                            )
+    with st.expander("Model B"):
+        temperatureB = st.slider("Temperature Model B", 0., 1., value=0.8, step=0.1)
+        sourceB = st.selectbox("Select a source Model B", ['OpenAI GPT3.5','Finetuned OpenAI'], index=1,
+                                on_change=clear_memory, args=(memories,)
+                            )
+        
+    sbcol1, sbcol2 = st.columns(2)
+    beta = sbcol1.button("A is better", on_click=replaceB)
+    betb = sbcol2.button("B is better", on_click=replaceA)
+
+    same = sbcol1.button("Tie", on_click=bothGood)
+    bbad = sbcol2.button("Both are bad", on_click=regenerateBoth)
+    
+    # regenA = sbcol1.button("Regenerate A", on_click=regenerateA)
+    # regenB = sbcol2.button("Regenerate B", on_click=regenerateB)
+    clear = st.button("Clear History", on_click=clear_memory, args=(memories,))
+
+create_memory_add_initial_message(memories, language)
+llm_chainA = get_chain(issue, language, sourceA, st.session_state[memories[0]], temperatureA)
+llm_chainB = get_chain(issue, language, sourceB, st.session_state[memories[1]], temperatureB)
+
+st.title(f"💬 History") 
+for msg in st.session_state['commonMemory'].buffer_as_messages:
+    role = "user" if type(msg) == HumanMessage else "assistant"
+    st.chat_message(role).write(msg.content)
+
+
+col1, col2 = st.columns(2)
+col1.title(f"💬 Simulator A") 
+col2.title(f"💬 Simulator B") 
+
+def reset_buttons():
+    buttons = [beta, betb, same, bbad, 
+               #regenA, regenB
+               ]
+    for but in buttons:
+        but = False
+
+def disable_chat():
+    buttons = [beta, betb, same, bbad]
+    if any(buttons):
+        return False
+    else:
+        return True
+
+if prompt := st.chat_input(disabled=disable_chat()):
+    col1.chat_message("user").write(prompt)
+    col2.chat_message("user").write(prompt)
+
+    responseA = llm_chainA.predict(input=prompt, stop="helper:")
+    responseB = llm_chainB.predict(input=prompt, stop="helper:")
+
+    col1.chat_message("assistant").write(responseA)
+    col2.chat_message("assistant").write(responseB)
+
+    reset_buttons()

requirements.txt

@@ -0,0 +1,3 @@
+scipy==1.11.1
+openai==0.28.0
+langchain==0.0.281

utils.py

@@ -0,0 +1,34 @@
+import streamlit as st
+from langchain.memory import ConversationBufferMemory
+
+from models.openai.finetuned_models import finetuned_models, get_finetuned_chain
+from models.openai.role_models import get_role_chain, role_templates
+
+def add_initial_message(model_name, memory):
+    if "Spanish" in model_name:
+        memory.chat_memory.add_ai_message("Hola necesito ayuda") 
+    else:
+        memory.chat_memory.add_ai_message("Hi I need help")
+
+def clear_memory(memories):
+    for memory in memories:
+        if memory not in st.session_state:
+            st.session_state[memory] = ConversationBufferMemory(ai_prefix='texter', human_prefix='helper')
+        st.session_state[memory].clear()
+
+def create_memory_add_initial_message(memories, language):
+    for memory in memories:
+        if memory not in st.session_state:
+            st.session_state[memory] = ConversationBufferMemory(ai_prefix='texter', human_prefix='helper')
+            add_initial_message(language, st.session_state[memory])
+        if len(st.session_state[memory].buffer_as_messages) < 1:
+            add_initial_message(language, st.session_state[memory])
+
+
+def get_chain(issue, language, source, memory, temperature):
+    if source in ("Finetuned OpenAI"):
+        OA_engine = finetuned_models[f"{issue}-{language}"]
+        return get_finetuned_chain(OA_engine, memory, temperature)
+    if source in ('OpenAI GPT3.5'):
+        template = role_templates[f"{issue}-{language}"]
+        return get_role_chain(template, memory, temperature)