Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -4,7 +4,6 @@ import openai
|
|
| 4 |
import anthropic
|
| 5 |
import os
|
| 6 |
|
| 7 |
-
# 제거할 모델들을 MODELS 사전에서 제외
|
| 8 |
MODELS = {
|
| 9 |
"Zephyr 7B Beta": "HuggingFaceH4/zephyr-7b-beta",
|
| 10 |
"Meta Llama 3.1 8B": "meta-llama/Meta-Llama-3.1-8B-Instruct",
|
|
@@ -15,7 +14,6 @@ MODELS = {
|
|
| 15 |
"Aya-23-35B": "CohereForAI/aya-23-35B"
|
| 16 |
}
|
| 17 |
|
| 18 |
-
# Cohere Command R+ 모델 ID 정의
|
| 19 |
COHERE_MODEL = "CohereForAI/c4ai-command-r-plus-08-2024"
|
| 20 |
|
| 21 |
def get_client(model_name):
|
|
@@ -30,6 +28,7 @@ def get_client(model_name):
|
|
| 30 |
raise ValueError("유효하지 않은 모델 이름입니다.")
|
| 31 |
return InferenceClient(model_id, token=hf_token)
|
| 32 |
|
|
|
|
| 33 |
def respond(
|
| 34 |
message,
|
| 35 |
chat_history,
|
|
@@ -39,12 +38,16 @@ def respond(
|
|
| 39 |
top_p,
|
| 40 |
system_message,
|
| 41 |
):
|
|
|
|
|
|
|
|
|
|
| 42 |
try:
|
| 43 |
client = get_client(model_name)
|
| 44 |
except ValueError as e:
|
| 45 |
chat_history.append((message, str(e)))
|
| 46 |
return chat_history
|
| 47 |
|
|
|
|
| 48 |
messages = [{"role": "system", "content": system_message}]
|
| 49 |
for human, assistant in chat_history:
|
| 50 |
messages.append({"role": "user", "content": human})
|
|
@@ -52,8 +55,8 @@ def respond(
|
|
| 52 |
messages.append({"role": "user", "content": message})
|
| 53 |
|
| 54 |
try:
|
|
|
|
| 55 |
if model_name == "Cohere Command R+":
|
| 56 |
-
# Cohere Command R+ 모델을 위한 비스트리밍 처리
|
| 57 |
response = client.chat_completion(
|
| 58 |
messages,
|
| 59 |
max_tokens=max_tokens,
|
|
@@ -64,7 +67,6 @@ def respond(
|
|
| 64 |
chat_history.append((message, assistant_message))
|
| 65 |
return chat_history
|
| 66 |
else:
|
| 67 |
-
# 다른 모델들을 위한 스트리밍 처리
|
| 68 |
stream = client.chat_completion(
|
| 69 |
messages,
|
| 70 |
max_tokens=max_tokens,
|
|
@@ -86,6 +88,7 @@ def respond(
|
|
| 86 |
chat_history.append((message, error_message))
|
| 87 |
yield chat_history
|
| 88 |
|
|
|
|
| 89 |
def cohere_respond(
|
| 90 |
message,
|
| 91 |
chat_history,
|
|
@@ -94,6 +97,9 @@ def cohere_respond(
|
|
| 94 |
temperature,
|
| 95 |
top_p,
|
| 96 |
):
|
|
|
|
|
|
|
|
|
|
| 97 |
model_name = "Cohere Command R+"
|
| 98 |
try:
|
| 99 |
client = get_client(model_name)
|
|
@@ -101,7 +107,11 @@ def cohere_respond(
|
|
| 101 |
chat_history.append((message, str(e)))
|
| 102 |
return chat_history
|
| 103 |
|
| 104 |
-
messages = [
|
|
|
|
|
|
|
|
|
|
|
|
|
| 105 |
for human, assistant in chat_history:
|
| 106 |
if human:
|
| 107 |
messages.append({"role": "user", "content": human})
|
|
@@ -111,7 +121,6 @@ def cohere_respond(
|
|
| 111 |
messages.append({"role": "user", "content": message})
|
| 112 |
|
| 113 |
try:
|
| 114 |
-
# Cohere Command R+ 모델을 위한 비스트리밍 처리
|
| 115 |
response_full = client.chat_completion(
|
| 116 |
messages,
|
| 117 |
max_tokens=max_tokens,
|
|
@@ -126,6 +135,7 @@ def cohere_respond(
|
|
| 126 |
chat_history.append((message, error_message))
|
| 127 |
return chat_history
|
| 128 |
|
|
|
|
| 129 |
def chatgpt_respond(
|
| 130 |
message,
|
| 131 |
chat_history,
|
|
@@ -134,11 +144,15 @@ def chatgpt_respond(
|
|
| 134 |
temperature,
|
| 135 |
top_p,
|
| 136 |
):
|
|
|
|
|
|
|
|
|
|
| 137 |
openai.api_key = os.getenv("OPENAI_API_KEY")
|
| 138 |
if not openai.api_key:
|
| 139 |
chat_history.append((message, "OPENAI_API_KEY 환경 변수가 필요합니다."))
|
| 140 |
return chat_history
|
| 141 |
|
|
|
|
| 142 |
messages = [{"role": "system", "content": system_message}]
|
| 143 |
for human, assistant in chat_history:
|
| 144 |
messages.append({"role": "user", "content": human})
|
|
@@ -147,7 +161,7 @@ def chatgpt_respond(
|
|
| 147 |
|
| 148 |
try:
|
| 149 |
response = openai.ChatCompletion.create(
|
| 150 |
-
model="gpt-4", #
|
| 151 |
messages=messages,
|
| 152 |
max_tokens=max_tokens,
|
| 153 |
temperature=temperature,
|
|
@@ -161,6 +175,7 @@ def chatgpt_respond(
|
|
| 161 |
chat_history.append((message, error_message))
|
| 162 |
return chat_history
|
| 163 |
|
|
|
|
| 164 |
def claude_respond(
|
| 165 |
message,
|
| 166 |
chat_history,
|
|
@@ -169,33 +184,39 @@ def claude_respond(
|
|
| 169 |
temperature,
|
| 170 |
top_p,
|
| 171 |
):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 172 |
anthropic_api_key = os.getenv("ANTHROPIC_API_KEY")
|
| 173 |
if not anthropic_api_key:
|
| 174 |
chat_history.append((message, "ANTHROPIC_API_KEY 환경 변수가 필요합니다."))
|
| 175 |
return chat_history
|
| 176 |
|
| 177 |
-
client = anthropic.Anthropic(
|
| 178 |
-
api_key=anthropic_api_key,
|
| 179 |
-
)
|
| 180 |
|
| 181 |
-
#
|
| 182 |
-
|
|
|
|
|
|
|
| 183 |
for human, assistant in chat_history:
|
| 184 |
if human:
|
| 185 |
-
|
| 186 |
if assistant:
|
| 187 |
-
|
| 188 |
-
|
| 189 |
|
| 190 |
try:
|
| 191 |
response = client.messages.create(
|
| 192 |
-
model="claude-3-haiku-20240307", # 사용 가능
|
| 193 |
-
|
|
|
|
|
|
|
| 194 |
temperature=temperature,
|
| 195 |
top_p=top_p,
|
| 196 |
-
messages=messages,
|
| 197 |
)
|
| 198 |
-
assistant_message = response[
|
| 199 |
chat_history.append((message, assistant_message))
|
| 200 |
return chat_history
|
| 201 |
except Exception as e:
|
|
@@ -203,13 +224,15 @@ def claude_respond(
|
|
| 203 |
chat_history.append((message, error_message))
|
| 204 |
return chat_history
|
| 205 |
|
|
|
|
| 206 |
def clear_conversation():
|
| 207 |
return []
|
| 208 |
|
|
|
|
| 209 |
with gr.Blocks() as demo:
|
| 210 |
gr.Markdown("# Prompting AI Chatbot")
|
| 211 |
gr.Markdown("언어모델별 프롬프트 테스트 챗봇입니다.")
|
| 212 |
-
|
| 213 |
with gr.Tab("일반 모델"):
|
| 214 |
with gr.Row():
|
| 215 |
with gr.Column(scale=1):
|
|
@@ -240,7 +263,7 @@ with gr.Blocks() as demo:
|
|
| 240 |
msg.submit(respond, [msg, chatbot, model_name, max_tokens, temperature, top_p, system_message], chatbot)
|
| 241 |
submit_button.click(respond, [msg, chatbot, model_name, max_tokens, temperature, top_p, system_message], chatbot)
|
| 242 |
clear_button.click(clear_conversation, outputs=chatbot, queue=False)
|
| 243 |
-
|
| 244 |
with gr.Tab("Cohere Command R+"):
|
| 245 |
with gr.Row():
|
| 246 |
cohere_system_message = gr.Textbox(
|
|
@@ -260,7 +283,7 @@ with gr.Blocks() as demo:
|
|
| 260 |
step=0.05,
|
| 261 |
label="Top-P",
|
| 262 |
)
|
| 263 |
-
|
| 264 |
cohere_chatbot = gr.Chatbot(height=600)
|
| 265 |
cohere_msg = gr.Textbox(label="메세지를 입력하세요")
|
| 266 |
with gr.Row():
|
|
@@ -278,7 +301,7 @@ with gr.Blocks() as demo:
|
|
| 278 |
cohere_chatbot
|
| 279 |
)
|
| 280 |
cohere_clear_button.click(clear_conversation, outputs=cohere_chatbot, queue=False)
|
| 281 |
-
|
| 282 |
with gr.Tab("ChatGPT"):
|
| 283 |
with gr.Row():
|
| 284 |
chatgpt_system_message = gr.Textbox(
|
|
@@ -298,7 +321,7 @@ with gr.Blocks() as demo:
|
|
| 298 |
step=0.05,
|
| 299 |
label="Top-P",
|
| 300 |
)
|
| 301 |
-
|
| 302 |
chatgpt_chatbot = gr.Chatbot(height=600)
|
| 303 |
chatgpt_msg = gr.Textbox(label="메세지를 입력하세요")
|
| 304 |
with gr.Row():
|
|
@@ -316,7 +339,7 @@ with gr.Blocks() as demo:
|
|
| 316 |
chatgpt_chatbot
|
| 317 |
)
|
| 318 |
chatgpt_clear_button.click(clear_conversation, outputs=chatgpt_chatbot, queue=False)
|
| 319 |
-
|
| 320 |
with gr.Tab("Claude"):
|
| 321 |
with gr.Row():
|
| 322 |
claude_system_message = gr.Textbox(
|
|
@@ -336,13 +359,14 @@ with gr.Blocks() as demo:
|
|
| 336 |
step=0.05,
|
| 337 |
label="Top-P",
|
| 338 |
)
|
| 339 |
-
|
| 340 |
claude_chatbot = gr.Chatbot(height=600)
|
| 341 |
claude_msg = gr.Textbox(label="메세지를 입력하세요")
|
| 342 |
with gr.Row():
|
| 343 |
claude_submit_button = gr.Button("전송")
|
| 344 |
claude_clear_button = gr.Button("대화 내역 지우기")
|
| 345 |
-
|
|
|
|
| 346 |
claude_msg.submit(
|
| 347 |
claude_respond,
|
| 348 |
[claude_msg, claude_chatbot, claude_system_message, claude_max_tokens, claude_temperature, claude_top_p],
|
|
|
|
| 4 |
import anthropic
|
| 5 |
import os
|
| 6 |
|
|
|
|
| 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_MODEL = "CohereForAI/c4ai-command-r-plus-08-2024"
|
| 18 |
|
| 19 |
def get_client(model_name):
|
|
|
|
| 28 |
raise ValueError("유효하지 않은 모델 이름입니다.")
|
| 29 |
return InferenceClient(model_id, token=hf_token)
|
| 30 |
|
| 31 |
+
|
| 32 |
def respond(
|
| 33 |
message,
|
| 34 |
chat_history,
|
|
|
|
| 38 |
top_p,
|
| 39 |
system_message,
|
| 40 |
):
|
| 41 |
+
"""
|
| 42 |
+
기존 모델(Zephyr, Meta Llama 등) / Cohere Command R+ 모델에 대응하는 함수
|
| 43 |
+
"""
|
| 44 |
try:
|
| 45 |
client = get_client(model_name)
|
| 46 |
except ValueError as e:
|
| 47 |
chat_history.append((message, str(e)))
|
| 48 |
return chat_history
|
| 49 |
|
| 50 |
+
# Gradio에서 유지하던 형식
|
| 51 |
messages = [{"role": "system", "content": system_message}]
|
| 52 |
for human, assistant in chat_history:
|
| 53 |
messages.append({"role": "user", "content": human})
|
|
|
|
| 55 |
messages.append({"role": "user", "content": message})
|
| 56 |
|
| 57 |
try:
|
| 58 |
+
# Cohere 모델(비스트리밍) vs 기타 모델(스트리밍) 구분
|
| 59 |
if model_name == "Cohere Command R+":
|
|
|
|
| 60 |
response = client.chat_completion(
|
| 61 |
messages,
|
| 62 |
max_tokens=max_tokens,
|
|
|
|
| 67 |
chat_history.append((message, assistant_message))
|
| 68 |
return chat_history
|
| 69 |
else:
|
|
|
|
| 70 |
stream = client.chat_completion(
|
| 71 |
messages,
|
| 72 |
max_tokens=max_tokens,
|
|
|
|
| 88 |
chat_history.append((message, error_message))
|
| 89 |
yield chat_history
|
| 90 |
|
| 91 |
+
|
| 92 |
def cohere_respond(
|
| 93 |
message,
|
| 94 |
chat_history,
|
|
|
|
| 97 |
temperature,
|
| 98 |
top_p,
|
| 99 |
):
|
| 100 |
+
"""
|
| 101 |
+
Cohere Command R+ 전용 응답 함수
|
| 102 |
+
"""
|
| 103 |
model_name = "Cohere Command R+"
|
| 104 |
try:
|
| 105 |
client = get_client(model_name)
|
|
|
|
| 107 |
chat_history.append((message, str(e)))
|
| 108 |
return chat_history
|
| 109 |
|
| 110 |
+
messages = []
|
| 111 |
+
# system_message는 messages 내 system 키로 추가하지 않고, 아래와 같이 적용해도 문제는 없음
|
| 112 |
+
# 이 부분은 기존 구조를 유지
|
| 113 |
+
messages.append({"role": "system", "content": system_message})
|
| 114 |
+
|
| 115 |
for human, assistant in chat_history:
|
| 116 |
if human:
|
| 117 |
messages.append({"role": "user", "content": human})
|
|
|
|
| 121 |
messages.append({"role": "user", "content": message})
|
| 122 |
|
| 123 |
try:
|
|
|
|
| 124 |
response_full = client.chat_completion(
|
| 125 |
messages,
|
| 126 |
max_tokens=max_tokens,
|
|
|
|
| 135 |
chat_history.append((message, error_message))
|
| 136 |
return chat_history
|
| 137 |
|
| 138 |
+
|
| 139 |
def chatgpt_respond(
|
| 140 |
message,
|
| 141 |
chat_history,
|
|
|
|
| 144 |
temperature,
|
| 145 |
top_p,
|
| 146 |
):
|
| 147 |
+
"""
|
| 148 |
+
ChatGPT (OpenAI) 전용 응답 함수
|
| 149 |
+
"""
|
| 150 |
openai.api_key = os.getenv("OPENAI_API_KEY")
|
| 151 |
if not openai.api_key:
|
| 152 |
chat_history.append((message, "OPENAI_API_KEY 환경 변수가 필요합니다."))
|
| 153 |
return chat_history
|
| 154 |
|
| 155 |
+
# ChatGPT는 system role을 messages 내에서 허용하므로 현재 방식 OK
|
| 156 |
messages = [{"role": "system", "content": system_message}]
|
| 157 |
for human, assistant in chat_history:
|
| 158 |
messages.append({"role": "user", "content": human})
|
|
|
|
| 161 |
|
| 162 |
try:
|
| 163 |
response = openai.ChatCompletion.create(
|
| 164 |
+
model="gpt-4", # 적절한 모델 ID를 사용
|
| 165 |
messages=messages,
|
| 166 |
max_tokens=max_tokens,
|
| 167 |
temperature=temperature,
|
|
|
|
| 175 |
chat_history.append((message, error_message))
|
| 176 |
return chat_history
|
| 177 |
|
| 178 |
+
|
| 179 |
def claude_respond(
|
| 180 |
message,
|
| 181 |
chat_history,
|
|
|
|
| 184 |
temperature,
|
| 185 |
top_p,
|
| 186 |
):
|
| 187 |
+
"""
|
| 188 |
+
Anthropic Claude 전용 응답 함수 (Messages API)
|
| 189 |
+
- "system" role은 messages 목록에 넣지 않고,
|
| 190 |
+
최상위 system 인자로 넘겨야 한다.
|
| 191 |
+
"""
|
| 192 |
anthropic_api_key = os.getenv("ANTHROPIC_API_KEY")
|
| 193 |
if not anthropic_api_key:
|
| 194 |
chat_history.append((message, "ANTHROPIC_API_KEY 환경 변수가 필요합니다."))
|
| 195 |
return chat_history
|
| 196 |
|
| 197 |
+
client = anthropic.Anthropic(api_key=anthropic_api_key)
|
|
|
|
|
|
|
| 198 |
|
| 199 |
+
# Anthropic Messages API에 맞게 "system" role은 top-level 파라미터로,
|
| 200 |
+
# 나머지는 user/assistant만 messages에 넣음
|
| 201 |
+
# (messages 내 "system" role이 있으면 오류 발생)
|
| 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:
|
|
|
|
| 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 |
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 |
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 |
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 |
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():
|
|
|
|
| 339 |
chatgpt_chatbot
|
| 340 |
)
|
| 341 |
chatgpt_clear_button.click(clear_conversation, outputs=chatgpt_chatbot, queue=False)
|
| 342 |
+
|
| 343 |
with gr.Tab("Claude"):
|
| 344 |
with gr.Row():
|
| 345 |
claude_system_message = gr.Textbox(
|
|
|
|
| 359 |
step=0.05,
|
| 360 |
label="Top-P",
|
| 361 |
)
|
| 362 |
+
|
| 363 |
claude_chatbot = gr.Chatbot(height=600)
|
| 364 |
claude_msg = gr.Textbox(label="메세지를 입력하세요")
|
| 365 |
with gr.Row():
|
| 366 |
claude_submit_button = gr.Button("전송")
|
| 367 |
claude_clear_button = gr.Button("대화 내역 지우기")
|
| 368 |
+
|
| 369 |
+
# Claude 전용 함수 호출
|
| 370 |
claude_msg.submit(
|
| 371 |
claude_respond,
|
| 372 |
[claude_msg, claude_chatbot, claude_system_message, claude_max_tokens, claude_temperature, claude_top_p],
|