Kims12's picture
Update app.py
6d1479a verified
raw
history blame
14.3 kB
import gradio as gr
from huggingface_hub import InferenceClient
import openai
import anthropic
import os
MODELS = {
"Zephyr 7B Beta": "HuggingFaceH4/zephyr-7b-beta",
"Meta Llama 3.1 8B": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"Meta-Llama 3.1 70B-Instruct": "meta-llama/Meta-Llama-3.1-70B-Instruct",
"Microsoft": "microsoft/Phi-3-mini-4k-instruct",
"Mixtral 8x7B": "mistralai/Mistral-7B-Instruct-v0.3",
"Mixtral Nous-Hermes": "NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO",
"Aya-23-35B": "CohereForAI/aya-23-35B"
}
COHERE_MODEL = "CohereForAI/c4ai-command-r-plus-08-2024"
def get_client(model_name):
hf_token = os.getenv("HF_TOKEN")
if not hf_token:
raise ValueError("HF_TOKEN 환경 변수가 필요합니다.")
if model_name in MODELS:
model_id = MODELS[model_name]
elif model_name == "Cohere Command R+":
model_id = COHERE_MODEL
else:
raise ValueError("유효하지 않은 모델 이름입니다.")
return InferenceClient(model_id, token=hf_token)
def respond(
message,
chat_history,
model_name,
max_tokens,
temperature,
top_p,
system_message,
):
"""
기존 모델(Zephyr, Meta Llama 등) / Cohere Command R+ 모델에 대응하는 함수
"""
try:
client = get_client(model_name)
except ValueError as e:
chat_history.append((message, str(e)))
return chat_history
# Gradio에서 유지하던 형식
messages = [{"role": "system", "content": system_message}]
for human, assistant in chat_history:
messages.append({"role": "user", "content": human})
messages.append({"role": "assistant", "content": assistant})
messages.append({"role": "user", "content": message})
try:
# Cohere 모델(비스트리밍) vs 기타 모델(스트리밍) 구분
if model_name == "Cohere Command R+":
response = client.chat_completion(
messages,
max_tokens=max_tokens,
temperature=temperature,
top_p=top_p,
)
assistant_message = response.choices[0].message.content
chat_history.append((message, assistant_message))
return chat_history
else:
stream = client.chat_completion(
messages,
max_tokens=max_tokens,
temperature=temperature,
top_p=top_p,
stream=True,
)
partial_message = ""
for response in stream:
if response.choices[0].delta.content is not None:
partial_message += response.choices[0].delta.content
if len(chat_history) > 0 and chat_history[-1][0] == message:
chat_history[-1] = (message, partial_message)
else:
chat_history.append((message, partial_message))
yield chat_history
except Exception as e:
error_message = f"오류가 발생했습니다: {str(e)}"
chat_history.append((message, error_message))
yield chat_history
def cohere_respond(
message,
chat_history,
system_message,
max_tokens,
temperature,
top_p,
):
"""
Cohere Command R+ 전용 응답 함수
"""
model_name = "Cohere Command R+"
try:
client = get_client(model_name)
except ValueError as e:
chat_history.append((message, str(e)))
return chat_history
messages = []
# system_message는 messages 내 system 키로 추가하지 않고, 아래와 같이 적용해도 문제는 없음
# 이 부분은 기존 구조를 유지
messages.append({"role": "system", "content": system_message})
for human, assistant in chat_history:
if human:
messages.append({"role": "user", "content": human})
if assistant:
messages.append({"role": "assistant", "content": assistant})
messages.append({"role": "user", "content": message})
try:
response_full = client.chat_completion(
messages,
max_tokens=max_tokens,
temperature=temperature,
top_p=top_p,
)
assistant_message = response_full.choices[0].message.content
chat_history.append((message, assistant_message))
return chat_history
except Exception as e:
error_message = f"오류가 발생했습니다: {str(e)}"
chat_history.append((message, error_message))
return chat_history
def chatgpt_respond(
message,
chat_history,
system_message,
max_tokens,
temperature,
top_p,
):
"""
ChatGPT (OpenAI) 전용 응답 함수
"""
openai.api_key = os.getenv("OPENAI_API_KEY")
if not openai.api_key:
chat_history.append((message, "OPENAI_API_KEY 환경 변수가 필요합니다."))
return chat_history
# ChatGPT는 system role을 messages 내에서 허용하므로 현재 방식 OK
messages = [{"role": "system", "content": system_message}]
for human, assistant in chat_history:
messages.append({"role": "user", "content": human})
messages.append({"role": "assistant", "content": assistant})
messages.append({"role": "user", "content": message})
try:
response = openai.ChatCompletion.create(
model="gpt-4", # 적절한 모델 ID를 사용
messages=messages,
max_tokens=max_tokens,
temperature=temperature,
top_p=top_p,
)
assistant_message = response.choices[0].message['content']
chat_history.append((message, assistant_message))
return chat_history
except Exception as e:
error_message = f"오류가 발생했습니다: {str(e)}"
chat_history.append((message, error_message))
return chat_history
def claude_respond(
message,
chat_history,
system_message,
max_tokens,
temperature,
top_p,
):
"""
Anthropic Claude 전용 응답 함수 (Messages API)
- "system" role은 messages 목록에 넣지 않고,
최상위 system 인자로 넘겨야 한다.
"""
anthropic_api_key = os.getenv("ANTHROPIC_API_KEY")
if not anthropic_api_key:
chat_history.append((message, "ANTHROPIC_API_KEY 환경 변수가 필요합니다."))
return chat_history
client = anthropic.Anthropic(api_key=anthropic_api_key)
# Anthropic Messages API에 맞게 "system" role은 top-level 파라미터로,
# 나머지는 user/assistant만 messages에 넣음
# (messages 내 "system" role이 있으면 오류 발생)
anthro_messages = []
for human, assistant in chat_history:
if human:
anthro_messages.append({"role": "user", "content": human})
if assistant:
anthro_messages.append({"role": "assistant", "content": assistant})
anthro_messages.append({"role": "user", "content": message})
try:
response = client.messages.create(
model="claude-3-haiku-20240307", # Claude 모델 예시 (사용 가능 모델 확인)
system=system_message, # ← 여기서 system_message를 전달
messages=anthro_messages, # user/assistant만 포함
max_tokens_to_sample=max_tokens,
temperature=temperature,
top_p=top_p,
)
assistant_message = response["completion"]
chat_history.append((message, assistant_message))
return chat_history
except Exception as e:
error_message = f"오류가 발생했습니다: {str(e)}"
chat_history.append((message, error_message))
return chat_history
def clear_conversation():
return []
with gr.Blocks() as demo:
gr.Markdown("# Prompting AI Chatbot")
gr.Markdown("언어모델별 프롬프트 테스트 챗봇입니다.")
with gr.Tab("일반 모델"):
with gr.Row():
with gr.Column(scale=1):
model_name = gr.Radio(
choices=list(MODELS.keys()),
label="Language Model",
value="Zephyr 7B Beta"
)
max_tokens = gr.Slider(minimum=0, maximum=2000, value=500, step=100, label="Max Tokens")
temperature = gr.Slider(minimum=0.1, maximum=2.0, value=0.7, step=0.05, label="Temperature")
top_p = gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p")
system_message = gr.Textbox(
value="""반드시 한글로 답변할 것.
너는 최고의 비서이다.
내가 요구하는것들을 최대한 자세하고 정확하게 답변하라.
""",
label="System Message",
lines=3
)
with gr.Column(scale=2):
chatbot = gr.Chatbot()
msg = gr.Textbox(label="메세지를 입력하세요")
with gr.Row():
submit_button = gr.Button("전송")
clear_button = gr.Button("대화 내역 지우기")
msg.submit(respond, [msg, chatbot, model_name, max_tokens, temperature, top_p, system_message], chatbot)
submit_button.click(respond, [msg, chatbot, model_name, max_tokens, temperature, top_p, system_message], chatbot)
clear_button.click(clear_conversation, outputs=chatbot, queue=False)
with gr.Tab("Cohere Command R+"):
with gr.Row():
cohere_system_message = gr.Textbox(
value="""반드시 한글로 답변할 것.
너는 최고의 비서이다.
내가 요구하는것들을 최대한 자세하고 정확하게 답변하라.
""",
label="System Message",
lines=3
)
cohere_max_tokens = gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens")
cohere_temperature = gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature")
cohere_top_p = gr.Slider(
minimum=0.1,
maximum=1.0,
value=0.95,
step=0.05,
label="Top-P",
)
cohere_chatbot = gr.Chatbot(height=600)
cohere_msg = gr.Textbox(label="메세지를 입력하세요")
with gr.Row():
cohere_submit_button = gr.Button("전송")
cohere_clear_button = gr.Button("대화 내역 지우기")
cohere_msg.submit(
cohere_respond,
[cohere_msg, cohere_chatbot, cohere_system_message, cohere_max_tokens, cohere_temperature, cohere_top_p],
cohere_chatbot
)
cohere_submit_button.click(
cohere_respond,
[cohere_msg, cohere_chatbot, cohere_system_message, cohere_max_tokens, cohere_temperature, cohere_top_p],
cohere_chatbot
)
cohere_clear_button.click(clear_conversation, outputs=cohere_chatbot, queue=False)
with gr.Tab("ChatGPT"):
with gr.Row():
chatgpt_system_message = gr.Textbox(
value="""반드시 한글로 답변할 것.
너는 ChatGPT, OpenAI에서 개발한 언어 모델이다.
내가 요구하는 것을 최대한 자세하고 정확하게 답변하라.
""",
label="System Message",
lines=3
)
chatgpt_max_tokens = gr.Slider(minimum=1, maximum=4096, value=1024, step=1, label="Max Tokens")
chatgpt_temperature = gr.Slider(minimum=0.1, maximum=2.0, value=0.7, step=0.05, label="Temperature")
chatgpt_top_p = gr.Slider(
minimum=0.1,
maximum=1.0,
value=0.95,
step=0.05,
label="Top-P",
)
chatgpt_chatbot = gr.Chatbot(height=600)
chatgpt_msg = gr.Textbox(label="메세지를 입력하세요")
with gr.Row():
chatgpt_submit_button = gr.Button("전송")
chatgpt_clear_button = gr.Button("대화 내역 지우기")
chatgpt_msg.submit(
chatgpt_respond,
[chatgpt_msg, chatgpt_chatbot, chatgpt_system_message, chatgpt_max_tokens, chatgpt_temperature, chatgpt_top_p],
chatgpt_chatbot
)
chatgpt_submit_button.click(
chatgpt_respond,
[chatgpt_msg, chatgpt_chatbot, chatgpt_system_message, chatgpt_max_tokens, chatgpt_temperature, chatgpt_top_p],
chatgpt_chatbot
)
chatgpt_clear_button.click(clear_conversation, outputs=chatgpt_chatbot, queue=False)
with gr.Tab("Claude"):
with gr.Row():
claude_system_message = gr.Textbox(
value="""반드시 한글로 답변할 것.
너는 Claude, Anthropic에서 개발한 언어 모델이다.
내가 요구하는 것을 최대한 자세하고 정확하게 답변하라.
""",
label="System Message",
lines=3
)
claude_max_tokens = gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max Tokens")
claude_temperature = gr.Slider(minimum=0.1, maximum=2.0, value=0.7, step=0.05, label="Temperature")
claude_top_p = gr.Slider(
minimum=0.1,
maximum=1.0,
value=0.95,
step=0.05,
label="Top-P",
)
claude_chatbot = gr.Chatbot(height=600)
claude_msg = gr.Textbox(label="메세지를 입력하세요")
with gr.Row():
claude_submit_button = gr.Button("전송")
claude_clear_button = gr.Button("대화 내역 지우기")
# Claude 전용 함수 호출
claude_msg.submit(
claude_respond,
[claude_msg, claude_chatbot, claude_system_message, claude_max_tokens, claude_temperature, claude_top_p],
claude_chatbot
)
claude_submit_button.click(
claude_respond,
[claude_msg, claude_chatbot, claude_system_message, claude_max_tokens, claude_temperature, claude_top_p],
claude_chatbot
)
claude_clear_button.click(clear_conversation, outputs=claude_chatbot, queue=False)
if __name__ == "__main__":
demo.launch()