app.py

# https://www.gradio.app/guides/using-hugging-face-integrations

import gradio as gr
from transformers import pipeline, Conversation

model = "mistralai/Mistral-7B-Instruct-v0.1"
model = "TinyLlama/TinyLlama-1.1B-Chat-v0.3"

title = "Shisa 7B"
description = "Test out Shisa 7B in either English or Japanese."
placeholder = "Type Here / ここに入力してください" 
examples = [
    "Hello, how are you?", 
    "こんにちは、元気ですか？",
    "おっす、元気？",
    "こんにちは、いかがお過ごしですか？",
]

# Docs: https://github.com/huggingface/transformers/blob/main/src/transformers/pipelines/conversational.py
conversation = Conversation()
chatbot = pipeline('conversational', model)
'''
conversation = Conversation("Going to the movies tonight - any suggestions?")
conversation.add_message({"role": "assistant", "content": "The Big lebowski."})
conversation.add_message({"role": "user", "content": "Is it good?"})
conversation.messages[:-1]
'''

def chat(input, history=[]):
    conversation.add_message({"role": "user", "content": input})
    # we do this shuffle so local shadow response doesn't get created
    response_conversation = chatbot(conversation)
    print(response_conversation)
    print(response_conversation.messages)
    print(response_conversation.messages[-1]["content"])

    conversation.add_message(response_conversation.messages[-1])
    response = conversation.messages[-1]["content"]
    return response, history

gr.ChatInterface(
    chat,
    chatbot=gr.Chatbot(height=400),
    textbox=gr.Textbox(placeholder=placeholder, container=False, scale=7),
    title=title,
    description=description,
    theme="soft",
    examples=examples,
    cache_examples=False,
    undo_btn="Delete Previous",
    clear_btn="Clear",
).launch()

'''
gr.Interface.load(
    "EleutherAI/gpt-j-6B",
    inputs=gr.Textbox(lines=5, label="Input Text"),
    title=title,
    description=description,
    article=article,
).launch()


# Doesn't support conversational pipelin
pipe = pipeline('conversational', model)
gr.Interface.from_pipeline(pipe).launch()


'''


# For async
# ).queue().launch()

'''
# Pipeline doesn't support conversational...
pipe = pipeline("conversational", model=model)
demo = gr.Interface.from_pipeline(pipe)
'''