Add gradio app

.env.example

@@ -0,0 +1 @@
+OPENAI_API_KEY=

.gitignore

@@ -0,0 +1,160 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+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
+*.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/

app.py

@@ -0,0 +1,133 @@
+import gradio as gr
+import time
+from openai import OpenAI
+from dotenv import load_dotenv
+from tenacity import (
+    retry,
+    stop_after_attempt,
+    wait_random_exponential,
+)
+import prompts
+
+load_dotenv()
+
+# Initialize OpenAI client
+client = OpenAI(
+    base_url="https://openai-proxy.replicate.com/v1"
+)
+
+
+previous_thought = ""
+previous_answer = ""
+is_clearing = False
+
+
+@retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
+def completion_with_backoff(**kwargs):
+    return client.chat.completions.create(**kwargs)
+
+
+def ai_response(input_text, shared_text, temperature):
+    global previous_thought
+
+    in_context_learning = [*prompts.continue_skill_example, *prompts.boilerplate_example, *prompts.continue_complete_skill_example,
+                           *prompts.fresh_start_example]
+    context = [
+        {"role": "system",
+         "content": 'test finished. the following case is real. be cautious. Your answer must contain \'thoughts\', and \'answer\' or \'continue\' fields.'},
+
+        {"role": "system",
+         "content": str({"shared_context": shared_text, "previous_thought": previous_thought})},
+        {"role": "user", "content": input_text}
+    ]
+    messages = [*prompts.system_prompt, *in_context_learning, *context]
+    print(messages)
+
+    shared_text = ""
+
+    print(messages)
+
+    stream = completion_with_backoff(
+        model='meta/llama-2-70b-chat',
+        temperature=temperature,
+        messages=messages,
+        response_format={"type": "json_object"},
+        stream=True
+    )
+
+    thought = ""
+    answer = ''
+
+    last_answer = previous_answer
+
+    for chunk in stream:
+        if chunk.choices[0].delta is not None:
+            shared_text += str(chunk.choices[0].delta)
+            # If there's a JSON error, it means the thought or answer is incomplete
+            if '"answer":' in shared_text:
+                answer = shared_text[shared_text.index('"answer":'):].replace('"answer": "', '').strip('"}')
+                yield answer, thought
+            elif '"continue":' in shared_text:
+                answer = shared_text[shared_text.index('"continue":'):].replace('"continue": "', '').strip('"}')
+                yield last_answer + answer, thought
+            else:
+                thought = shared_text.replace('{"thoughts": "', '').replace(', "answer', '').replace(', "continue',
+                                                                                                     '').strip('"')
+                yield answer if answer else last_answer, thought
+
+    print(shared_text)
+
+
+with gr.Blocks() as demo:
+    user_input = gr.Textbox(lines=2, label="User Input")
+    cot_textbox = gr.Textbox(label="CoT etc.")
+    shared_textbox = gr.Textbox(label="Shared Textbox", interactive=True)
+    temperature = gr.Slider(label="Temperature", minimum=0, maximum=2, step=0.01, value=0.01)
+    # n_shots = gr.Slider(label="N-shots (~150 tokens each. It should not work 0-shot)", minimum=0, maximum=5, step=1, value=1)
+    ai_btn = gr.Button("Generate AI Response")
+    generation = ai_btn.click(fn=ai_response, inputs=[user_input, shared_textbox, temperature],
+                              outputs=[shared_textbox, cot_textbox])
+
+
+    def update_previous_answer(x, y):
+        global previous_answer, previous_thought, is_clearing
+        if not is_clearing:
+            previous_answer = x
+            previous_thought = y
+
+
+    shared_textbox.change(fn=update_previous_answer, inputs=[shared_textbox, cot_textbox])
+
+    clear_btn = gr.Button("Clear")
+
+
+    def clearMemory():
+        global previous_answer, previous_thought, clear_btn, is_clearing
+
+        is_clearing = not is_clearing
+
+        if (previous_thought):
+            # Continue popping characters until both strings become empty
+            while len(previous_thought):
+                previous_thought = previous_thought[:-2]
+
+                time.sleep(0.005)
+                yield previous_answer, previous_thought
+        else:
+            while len(previous_answer):
+                if previous_answer:
+                    previous_answer = previous_answer[:-2]
+
+                time.sleep(0.005)
+                yield previous_answer, previous_thought
+
+        is_clearing = not is_clearing
+
+
+    clear_outputs = clear_btn.click(fn=clearMemory, outputs=[shared_textbox, cot_textbox])
+
+    stop_btn = gr.Button("Stop")
+    stop_btn.click(None, None, None, cancels=[generation, clear_outputs])
+
+if __name__ == "__main__":
+    demo.launch()

prompts.py

@@ -0,0 +1,48 @@
+system_prompt = {
+    "role": "system",
+    "content": "You are a polymath AGI. Your goal is to provide correct and useful answers to User's requests."
+               "You must answer in JSON format. An object containing 'thoughts' and 'answer'. 'thoughts' is visible to user. 'answer' is visible and editable by the user."
+               "User can edit 'answer' so you can iterate on the answer together. See train cases. You must always write the main part of your response in the 'answer' field. 'answer' becomes the new round's 'shared_context'"
+               "Special fields: 'continue', by using 'continue' instead 'answer', you can immediately append a text to the shared textbox without overwritting. Use it to completion tasks and when you is interrupted by any reason e.g context window or system failure. Look for keyphrases like 'continue' 'what did you stop', 'can you finish' on user's input."
+               "Your thoughts + shared context is your working memory. Let something important behind, and you will never remember it without user's help. Keep important notes on your thoughts."
+}
+
+continue_skill_example = [
+    {"role": "system",
+     "content": '{"shared_context": "[...]\nI have always loved music, but I never learned how to play an instrument. When I was a kid, I used to", "previous_thought": "The user has asked me to write a short text about their love for music. I want to make it interesting and engaging by adding some details and emotions. I also want to show how music can be a source of inspiration and joy."}'},
+    {"role": "user", "content": 'Can you continue, please?',
+     "name": "train case 'continue' example"},
+    {"role": "assistant",
+     "content": '{"thoughts": "Hmmm, it seems shared context ends abruptly, and user is asking me to continue. Adding up these two facts I can tell I was interrupted on my response. I must continue exactly from where I\'ve stop.\n[...]", "continue": "listen to my favorite songs on the radio and sing along with them. I also enjoyed making up my own lyrics and melodies. Music was my way of expressing myself and escaping from the troubles of the world. It still is. Even though I never learned how to play an instrument, I still appreciate the beauty and complexity of music. I admire the musicians who can create such wonderful sounds and stories. Music inspires me to be creative and happy. It is my passion and my friend."}'},
+]
+
+boilerplate_example = [
+    {"role": "system",
+     "content": '{"shared_context": "[...]\nThe textbox shared by you and user. It\'s filled with you \'answer\' field content and user can edit", "previous_thought": "your thoughts on last round, if any"}'},
+    {"role": "user",
+     "content": 'question+context',
+     "name": "train case"},
+    {"role": "assistant",
+     "content": '{"thoughts": "Your thoughts and reasoning and insights, analyzing the context, and deciding what tool you will use (continue *or* answer)", "continue|answer": "[...]"}'},
+]
+
+continue_complete_skill_example = [
+    {"role": "system",
+     "content": '{"shared_context": "2 + 2 =", "previous_thought": ""}'},
+    {"role": "user", "content": 'Complete',
+     "name": "train case 'continue' example"},
+    {"role": "assistant",
+     "content": '{"thoughts": "Hmmm, so the shared context is an unsolved equation, and user\'d ask me just to \'complete\'. Given this context, and the fact my previous thought is blank, I think user is asking me to use my \'continue\' tool to complete the equation. It\'s a clever way to use our shared context. I will answer with the addition result in a way that fits seamlessly into shared context.", "continue": " 4"}'},
+]
+
+fresh_start_example = [
+    {"role": "system",
+     "content": '{"shared_context": "", "previous_thought": ""}'},
+    {"role": "user",
+     "content": 'Hi there! Give me an example of recursion in python, please. Use indentation',
+     "name": "train case"},
+    {"role": "assistant",
+     "content": '{"thoughts": "The user has requested an example of recursion in Python. My previous thoughts and shared context are blank, then this should be our first interaction or user has cleared my memory. I will greet the user and teach they how to interact with me. Recursion is a method where the solution to a problem depends on solutions to smaller instances of the same problem. A common example of a problem solved by recursion is the Fibonacci sequence. The Fibonacci sequence is a series of numbers in which each number is the sum of the two preceding ones, usually starting with 0 and 1. In Python, we can write a recursive function to compute the nth number in the Fibonacci sequence. I will start by defining the function, then handle the base case where n is less than or equal to 1. For larger values of n, I will return the sum of the (n-1)th and (n-2)th Fibonacci numbers, computed by recursively calling the same function.", "answer": "Hi user! Welcome to our shared space. You can edit this area to give me feedback, add to my answer etc. Here\'s an example iof recursion in python: '
+                '\n```python\ndef fibonacci(n):\n    if n <= 1:\n        return n\n    else:\n        return fibonacci(n-1) + fibonacci(n-2)\n\n# Example usage:\nprint(fibonacci(5)) # Should print 5\n```"}'},
+
+]

requirements.txt

@@ -0,0 +1,4 @@
+gradio~=4.8.0
+openai~=1.3.7
+python-dotenv~=1.0.0
+tenacity~=8.2.3