Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,10 +1,9 @@
|
|
| 1 |
-
###################### (원본코드 시작) ######################
|
| 2 |
import gradio as gr
|
| 3 |
from huggingface_hub import InferenceClient
|
| 4 |
import openai
|
| 5 |
-
import anthropic
|
| 6 |
import os
|
| 7 |
|
|
|
|
| 8 |
MODELS = {
|
| 9 |
"Zephyr 7B Beta": "HuggingFaceH4/zephyr-7b-beta",
|
| 10 |
"Meta Llama 3.1 8B": "meta-llama/Meta-Llama-3.1-8B-Instruct",
|
|
@@ -15,6 +14,7 @@ MODELS = {
|
|
| 15 |
"Aya-23-35B": "CohereForAI/aya-23-35B"
|
| 16 |
}
|
| 17 |
|
|
|
|
| 18 |
COHERE_MODEL = "CohereForAI/c4ai-command-r-plus-08-2024"
|
| 19 |
|
| 20 |
def get_client(model_name):
|
|
@@ -29,7 +29,6 @@ def get_client(model_name):
|
|
| 29 |
raise ValueError("유효하지 않은 모델 이름입니다.")
|
| 30 |
return InferenceClient(model_id, token=hf_token)
|
| 31 |
|
| 32 |
-
|
| 33 |
def respond(
|
| 34 |
message,
|
| 35 |
chat_history,
|
|
@@ -39,16 +38,12 @@ def respond(
|
|
| 39 |
top_p,
|
| 40 |
system_message,
|
| 41 |
):
|
| 42 |
-
"""
|
| 43 |
-
기존 모델(Zephyr, Meta Llama 등) / Cohere Command R+ 모델에 대응하는 함수
|
| 44 |
-
"""
|
| 45 |
try:
|
| 46 |
client = get_client(model_name)
|
| 47 |
except ValueError as e:
|
| 48 |
chat_history.append((message, str(e)))
|
| 49 |
return chat_history
|
| 50 |
|
| 51 |
-
# Gradio에서 유지하던 형식
|
| 52 |
messages = [{"role": "system", "content": system_message}]
|
| 53 |
for human, assistant in chat_history:
|
| 54 |
messages.append({"role": "user", "content": human})
|
|
@@ -56,8 +51,8 @@ def respond(
|
|
| 56 |
messages.append({"role": "user", "content": message})
|
| 57 |
|
| 58 |
try:
|
| 59 |
-
# Cohere 모델(비스트리밍) vs 기타 모델(스트리밍) 구분
|
| 60 |
if model_name == "Cohere Command R+":
|
|
|
|
| 61 |
response = client.chat_completion(
|
| 62 |
messages,
|
| 63 |
max_tokens=max_tokens,
|
|
@@ -68,6 +63,7 @@ def respond(
|
|
| 68 |
chat_history.append((message, assistant_message))
|
| 69 |
return chat_history
|
| 70 |
else:
|
|
|
|
| 71 |
stream = client.chat_completion(
|
| 72 |
messages,
|
| 73 |
max_tokens=max_tokens,
|
|
@@ -89,7 +85,6 @@ def respond(
|
|
| 89 |
chat_history.append((message, error_message))
|
| 90 |
yield chat_history
|
| 91 |
|
| 92 |
-
|
| 93 |
def cohere_respond(
|
| 94 |
message,
|
| 95 |
chat_history,
|
|
@@ -98,9 +93,6 @@ def cohere_respond(
|
|
| 98 |
temperature,
|
| 99 |
top_p,
|
| 100 |
):
|
| 101 |
-
"""
|
| 102 |
-
Cohere Command R+ 전용 응답 함수
|
| 103 |
-
"""
|
| 104 |
model_name = "Cohere Command R+"
|
| 105 |
try:
|
| 106 |
client = get_client(model_name)
|
|
@@ -108,11 +100,7 @@ def cohere_respond(
|
|
| 108 |
chat_history.append((message, str(e)))
|
| 109 |
return chat_history
|
| 110 |
|
| 111 |
-
messages = []
|
| 112 |
-
# system_message는 messages 내 system 키로 추가하지 않고, 아래와 같이 적용해도 문제는 없음
|
| 113 |
-
# 이 부분은 기존 구조를 유지
|
| 114 |
-
messages.append({"role": "system", "content": system_message})
|
| 115 |
-
|
| 116 |
for human, assistant in chat_history:
|
| 117 |
if human:
|
| 118 |
messages.append({"role": "user", "content": human})
|
|
@@ -121,7 +109,10 @@ def cohere_respond(
|
|
| 121 |
|
| 122 |
messages.append({"role": "user", "content": message})
|
| 123 |
|
|
|
|
|
|
|
| 124 |
try:
|
|
|
|
| 125 |
response_full = client.chat_completion(
|
| 126 |
messages,
|
| 127 |
max_tokens=max_tokens,
|
|
@@ -136,7 +127,6 @@ def cohere_respond(
|
|
| 136 |
chat_history.append((message, error_message))
|
| 137 |
return chat_history
|
| 138 |
|
| 139 |
-
|
| 140 |
def chatgpt_respond(
|
| 141 |
message,
|
| 142 |
chat_history,
|
|
@@ -145,15 +135,11 @@ def chatgpt_respond(
|
|
| 145 |
temperature,
|
| 146 |
top_p,
|
| 147 |
):
|
| 148 |
-
"""
|
| 149 |
-
ChatGPT (OpenAI) 전용 응답 함수
|
| 150 |
-
"""
|
| 151 |
openai.api_key = os.getenv("OPENAI_API_KEY")
|
| 152 |
if not openai.api_key:
|
| 153 |
chat_history.append((message, "OPENAI_API_KEY 환경 변수가 필요합니다."))
|
| 154 |
return chat_history
|
| 155 |
|
| 156 |
-
# ChatGPT는 system role을 messages 내에서 허용하므로 현재 방식 OK
|
| 157 |
messages = [{"role": "system", "content": system_message}]
|
| 158 |
for human, assistant in chat_history:
|
| 159 |
messages.append({"role": "user", "content": human})
|
|
@@ -162,7 +148,7 @@ def chatgpt_respond(
|
|
| 162 |
|
| 163 |
try:
|
| 164 |
response = openai.ChatCompletion.create(
|
| 165 |
-
model="gpt-
|
| 166 |
messages=messages,
|
| 167 |
max_tokens=max_tokens,
|
| 168 |
temperature=temperature,
|
|
@@ -176,63 +162,13 @@ def chatgpt_respond(
|
|
| 176 |
chat_history.append((message, error_message))
|
| 177 |
return chat_history
|
| 178 |
|
| 179 |
-
|
| 180 |
-
def claude_respond(
|
| 181 |
-
message,
|
| 182 |
-
chat_history,
|
| 183 |
-
system_message,
|
| 184 |
-
max_tokens,
|
| 185 |
-
temperature,
|
| 186 |
-
top_p,
|
| 187 |
-
):
|
| 188 |
-
"""
|
| 189 |
-
Anthropic Claude 전용 응답 함수 (Messages API)
|
| 190 |
-
- "system" role은 messages 목록에 넣지 않고,
|
| 191 |
-
최상위 system 인자로 넘겨야 한다.
|
| 192 |
-
"""
|
| 193 |
-
anthropic_api_key = os.getenv("ANTHROPIC_API_KEY")
|
| 194 |
-
if not anthropic_api_key:
|
| 195 |
-
chat_history.append((message, "ANTHROPIC_API_KEY 환경 변수가 필요합니다."))
|
| 196 |
-
return chat_history
|
| 197 |
-
|
| 198 |
-
client = anthropic.Anthropic(api_key=anthropic_api_key)
|
| 199 |
-
|
| 200 |
-
# Anthropic Messages API에 맞게 "system" role은 top-level 파라미터로,
|
| 201 |
-
# 나머지는 user/assistant만 messages에 넣음
|
| 202 |
-
anthro_messages = []
|
| 203 |
-
for human, assistant in chat_history:
|
| 204 |
-
if human:
|
| 205 |
-
anthro_messages.append({"role": "user", "content": human})
|
| 206 |
-
if assistant:
|
| 207 |
-
anthro_messages.append({"role": "assistant", "content": assistant})
|
| 208 |
-
anthro_messages.append({"role": "user", "content": message})
|
| 209 |
-
|
| 210 |
-
try:
|
| 211 |
-
response = client.messages.create(
|
| 212 |
-
model="claude-3-haiku-20240307", # Claude 모델 예시 (사용 가능 모델 확인)
|
| 213 |
-
system=system_message, # ← 여기서 system_message를 전달
|
| 214 |
-
messages=anthro_messages, # user/assistant만 포함
|
| 215 |
-
max_tokens_to_sample=max_tokens,
|
| 216 |
-
temperature=temperature,
|
| 217 |
-
top_p=top_p,
|
| 218 |
-
)
|
| 219 |
-
assistant_message = response["completion"]
|
| 220 |
-
chat_history.append((message, assistant_message))
|
| 221 |
-
return chat_history
|
| 222 |
-
except Exception as e:
|
| 223 |
-
error_message = f"오류가 발생했습니다: {str(e)}"
|
| 224 |
-
chat_history.append((message, error_message))
|
| 225 |
-
return chat_history
|
| 226 |
-
|
| 227 |
-
|
| 228 |
def clear_conversation():
|
| 229 |
return []
|
| 230 |
|
| 231 |
-
|
| 232 |
with gr.Blocks() as demo:
|
| 233 |
gr.Markdown("# Prompting AI Chatbot")
|
| 234 |
gr.Markdown("언어모델별 프롬프트 테스트 챗봇입니다.")
|
| 235 |
-
|
| 236 |
with gr.Tab("일반 모델"):
|
| 237 |
with gr.Row():
|
| 238 |
with gr.Column(scale=1):
|
|
@@ -263,7 +199,7 @@ with gr.Blocks() as demo:
|
|
| 263 |
msg.submit(respond, [msg, chatbot, model_name, max_tokens, temperature, top_p, system_message], chatbot)
|
| 264 |
submit_button.click(respond, [msg, chatbot, model_name, max_tokens, temperature, top_p, system_message], chatbot)
|
| 265 |
clear_button.click(clear_conversation, outputs=chatbot, queue=False)
|
| 266 |
-
|
| 267 |
with gr.Tab("Cohere Command R+"):
|
| 268 |
with gr.Row():
|
| 269 |
cohere_system_message = gr.Textbox(
|
|
@@ -283,7 +219,7 @@ with gr.Blocks() as demo:
|
|
| 283 |
step=0.05,
|
| 284 |
label="Top-P",
|
| 285 |
)
|
| 286 |
-
|
| 287 |
cohere_chatbot = gr.Chatbot(height=600)
|
| 288 |
cohere_msg = gr.Textbox(label="메세지를 입력하세요")
|
| 289 |
with gr.Row():
|
|
@@ -301,7 +237,7 @@ with gr.Blocks() as demo:
|
|
| 301 |
cohere_chatbot
|
| 302 |
)
|
| 303 |
cohere_clear_button.click(clear_conversation, outputs=cohere_chatbot, queue=False)
|
| 304 |
-
|
| 305 |
with gr.Tab("ChatGPT"):
|
| 306 |
with gr.Row():
|
| 307 |
chatgpt_system_message = gr.Textbox(
|
|
@@ -321,7 +257,7 @@ with gr.Blocks() as demo:
|
|
| 321 |
step=0.05,
|
| 322 |
label="Top-P",
|
| 323 |
)
|
| 324 |
-
|
| 325 |
chatgpt_chatbot = gr.Chatbot(height=600)
|
| 326 |
chatgpt_msg = gr.Textbox(label="메세지를 입력하세요")
|
| 327 |
with gr.Row():
|
|
@@ -340,55 +276,5 @@ with gr.Blocks() as demo:
|
|
| 340 |
)
|
| 341 |
chatgpt_clear_button.click(clear_conversation, outputs=chatgpt_chatbot, queue=False)
|
| 342 |
|
| 343 |
-
###################### (원본코드 끝) ######################
|
| 344 |
-
|
| 345 |
-
|
| 346 |
-
# ------------------------- 여기부터 "클로드 전용 탭" 추가 -------------------------
|
| 347 |
-
# - 원본코드에 이미 claude_respond 함수가 있으므로 그대로 활용
|
| 348 |
-
# - 탭만 추가하여 UI 구성
|
| 349 |
-
|
| 350 |
-
with demo: # 원본코드에서 "with gr.Blocks() as demo:" 를 이미 열었음
|
| 351 |
-
with gr.Tab("Claude"):
|
| 352 |
-
with gr.Row():
|
| 353 |
-
claude_system_message = gr.Textbox(
|
| 354 |
-
value="""반드시 한글로 답변할 것.
|
| 355 |
-
너는 Claude, Anthropic에서 개발한 언어 모델이다.
|
| 356 |
-
내가 요구하는 것을 최대한 자세하고 정확하게 답변하라.
|
| 357 |
-
""",
|
| 358 |
-
label="System Message",
|
| 359 |
-
lines=3
|
| 360 |
-
)
|
| 361 |
-
claude_max_tokens = gr.Slider(
|
| 362 |
-
minimum=1, maximum=2048, value=512, step=1, label="Max Tokens"
|
| 363 |
-
)
|
| 364 |
-
claude_temperature = gr.Slider(
|
| 365 |
-
minimum=0.1, maximum=2.0, value=0.7, step=0.05, label="Temperature"
|
| 366 |
-
)
|
| 367 |
-
claude_top_p = gr.Slider(
|
| 368 |
-
minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-P",
|
| 369 |
-
)
|
| 370 |
-
|
| 371 |
-
claude_chatbot = gr.Chatbot(height=600)
|
| 372 |
-
claude_msg = gr.Textbox(label="메세지를 입력하세요")
|
| 373 |
-
|
| 374 |
-
with gr.Row():
|
| 375 |
-
claude_submit_button = gr.Button("전송")
|
| 376 |
-
claude_clear_button = gr.Button("대화 내역 지우기")
|
| 377 |
-
|
| 378 |
-
# 이미 정의된 claude_respond 함수를 그대로 사용
|
| 379 |
-
claude_msg.submit(
|
| 380 |
-
claude_respond,
|
| 381 |
-
[claude_msg, claude_chatbot, claude_system_message, claude_max_tokens, claude_temperature, claude_top_p],
|
| 382 |
-
claude_chatbot
|
| 383 |
-
)
|
| 384 |
-
claude_submit_button.click(
|
| 385 |
-
claude_respond,
|
| 386 |
-
[claude_msg, claude_chatbot, claude_system_message, claude_max_tokens, claude_temperature, claude_top_p],
|
| 387 |
-
claude_chatbot
|
| 388 |
-
)
|
| 389 |
-
claude_clear_button.click(clear_conversation, outputs=claude_chatbot, queue=False)
|
| 390 |
-
|
| 391 |
-
|
| 392 |
-
# 원본 코드의 데모 런치 부분도 수정 금지
|
| 393 |
if __name__ == "__main__":
|
| 394 |
demo.launch()
|
|
|
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
from huggingface_hub import InferenceClient
|
| 3 |
import openai
|
|
|
|
| 4 |
import os
|
| 5 |
|
| 6 |
+
# 제거할 모델들을 MODELS 사전에서 제외
|
| 7 |
MODELS = {
|
| 8 |
"Zephyr 7B Beta": "HuggingFaceH4/zephyr-7b-beta",
|
| 9 |
"Meta Llama 3.1 8B": "meta-llama/Meta-Llama-3.1-8B-Instruct",
|
|
|
|
| 14 |
"Aya-23-35B": "CohereForAI/aya-23-35B"
|
| 15 |
}
|
| 16 |
|
| 17 |
+
# Cohere Command R+ 모델 ID 정의
|
| 18 |
COHERE_MODEL = "CohereForAI/c4ai-command-r-plus-08-2024"
|
| 19 |
|
| 20 |
def get_client(model_name):
|
|
|
|
| 29 |
raise ValueError("유효하지 않은 모델 이름입니다.")
|
| 30 |
return InferenceClient(model_id, token=hf_token)
|
| 31 |
|
|
|
|
| 32 |
def respond(
|
| 33 |
message,
|
| 34 |
chat_history,
|
|
|
|
| 38 |
top_p,
|
| 39 |
system_message,
|
| 40 |
):
|
|
|
|
|
|
|
|
|
|
| 41 |
try:
|
| 42 |
client = get_client(model_name)
|
| 43 |
except ValueError as e:
|
| 44 |
chat_history.append((message, str(e)))
|
| 45 |
return chat_history
|
| 46 |
|
|
|
|
| 47 |
messages = [{"role": "system", "content": system_message}]
|
| 48 |
for human, assistant in chat_history:
|
| 49 |
messages.append({"role": "user", "content": human})
|
|
|
|
| 51 |
messages.append({"role": "user", "content": message})
|
| 52 |
|
| 53 |
try:
|
|
|
|
| 54 |
if model_name == "Cohere Command R+":
|
| 55 |
+
# Cohere Command R+ 모델을 위한 비스트리밍 처리
|
| 56 |
response = client.chat_completion(
|
| 57 |
messages,
|
| 58 |
max_tokens=max_tokens,
|
|
|
|
| 63 |
chat_history.append((message, assistant_message))
|
| 64 |
return chat_history
|
| 65 |
else:
|
| 66 |
+
# 다른 모델들을 위한 스트리밍 처리
|
| 67 |
stream = client.chat_completion(
|
| 68 |
messages,
|
| 69 |
max_tokens=max_tokens,
|
|
|
|
| 85 |
chat_history.append((message, error_message))
|
| 86 |
yield chat_history
|
| 87 |
|
|
|
|
| 88 |
def cohere_respond(
|
| 89 |
message,
|
| 90 |
chat_history,
|
|
|
|
| 93 |
temperature,
|
| 94 |
top_p,
|
| 95 |
):
|
|
|
|
|
|
|
|
|
|
| 96 |
model_name = "Cohere Command R+"
|
| 97 |
try:
|
| 98 |
client = get_client(model_name)
|
|
|
|
| 100 |
chat_history.append((message, str(e)))
|
| 101 |
return chat_history
|
| 102 |
|
| 103 |
+
messages = [{"role": "system", "content": system_message}]
|
|
|
|
|
|
|
|
|
|
|
|
|
| 104 |
for human, assistant in chat_history:
|
| 105 |
if human:
|
| 106 |
messages.append({"role": "user", "content": human})
|
|
|
|
| 109 |
|
| 110 |
messages.append({"role": "user", "content": message})
|
| 111 |
|
| 112 |
+
response = ""
|
| 113 |
+
|
| 114 |
try:
|
| 115 |
+
# Cohere Command R+ 모델을 위한 비스트리밍 처리
|
| 116 |
response_full = client.chat_completion(
|
| 117 |
messages,
|
| 118 |
max_tokens=max_tokens,
|
|
|
|
| 127 |
chat_history.append((message, error_message))
|
| 128 |
return chat_history
|
| 129 |
|
|
|
|
| 130 |
def chatgpt_respond(
|
| 131 |
message,
|
| 132 |
chat_history,
|
|
|
|
| 135 |
temperature,
|
| 136 |
top_p,
|
| 137 |
):
|
|
|
|
|
|
|
|
|
|
| 138 |
openai.api_key = os.getenv("OPENAI_API_KEY")
|
| 139 |
if not openai.api_key:
|
| 140 |
chat_history.append((message, "OPENAI_API_KEY 환경 변수가 필요합니다."))
|
| 141 |
return chat_history
|
| 142 |
|
|
|
|
| 143 |
messages = [{"role": "system", "content": system_message}]
|
| 144 |
for human, assistant in chat_history:
|
| 145 |
messages.append({"role": "user", "content": human})
|
|
|
|
| 148 |
|
| 149 |
try:
|
| 150 |
response = openai.ChatCompletion.create(
|
| 151 |
+
model="gpt-4o-mini", # 또는 다른 모델 ID 사용
|
| 152 |
messages=messages,
|
| 153 |
max_tokens=max_tokens,
|
| 154 |
temperature=temperature,
|
|
|
|
| 162 |
chat_history.append((message, error_message))
|
| 163 |
return chat_history
|
| 164 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 165 |
def clear_conversation():
|
| 166 |
return []
|
| 167 |
|
|
|
|
| 168 |
with gr.Blocks() as demo:
|
| 169 |
gr.Markdown("# Prompting AI Chatbot")
|
| 170 |
gr.Markdown("언어모델별 프롬프트 테스트 챗봇입니다.")
|
| 171 |
+
|
| 172 |
with gr.Tab("일반 모델"):
|
| 173 |
with gr.Row():
|
| 174 |
with gr.Column(scale=1):
|
|
|
|
| 199 |
msg.submit(respond, [msg, chatbot, model_name, max_tokens, temperature, top_p, system_message], chatbot)
|
| 200 |
submit_button.click(respond, [msg, chatbot, model_name, max_tokens, temperature, top_p, system_message], chatbot)
|
| 201 |
clear_button.click(clear_conversation, outputs=chatbot, queue=False)
|
| 202 |
+
|
| 203 |
with gr.Tab("Cohere Command R+"):
|
| 204 |
with gr.Row():
|
| 205 |
cohere_system_message = gr.Textbox(
|
|
|
|
| 219 |
step=0.05,
|
| 220 |
label="Top-P",
|
| 221 |
)
|
| 222 |
+
|
| 223 |
cohere_chatbot = gr.Chatbot(height=600)
|
| 224 |
cohere_msg = gr.Textbox(label="메세지를 입력하세요")
|
| 225 |
with gr.Row():
|
|
|
|
| 237 |
cohere_chatbot
|
| 238 |
)
|
| 239 |
cohere_clear_button.click(clear_conversation, outputs=cohere_chatbot, queue=False)
|
| 240 |
+
|
| 241 |
with gr.Tab("ChatGPT"):
|
| 242 |
with gr.Row():
|
| 243 |
chatgpt_system_message = gr.Textbox(
|
|
|
|
| 257 |
step=0.05,
|
| 258 |
label="Top-P",
|
| 259 |
)
|
| 260 |
+
|
| 261 |
chatgpt_chatbot = gr.Chatbot(height=600)
|
| 262 |
chatgpt_msg = gr.Textbox(label="메세지를 입력하세요")
|
| 263 |
with gr.Row():
|
|
|
|
| 276 |
)
|
| 277 |
chatgpt_clear_button.click(clear_conversation, outputs=chatgpt_chatbot, queue=False)
|
| 278 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 279 |
if __name__ == "__main__":
|
| 280 |
demo.launch()
|