Spaces:
Runtime error
Runtime error
Better error message, add back up API key
Browse files
app.py
CHANGED
|
@@ -25,6 +25,10 @@ character_avatars = {
|
|
| 25 |
}
|
| 26 |
|
| 27 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
predefined_characters = ["Harry Potter", "Hermione Granger", "poor Ph.D. student", "Donald Trump", "a super cute red panda"]
|
| 29 |
|
| 30 |
def get_character(dropdown_value, custom_value):
|
|
@@ -47,71 +51,105 @@ for character, filename in character_avatars.items():
|
|
| 47 |
else:
|
| 48 |
pass
|
| 49 |
|
| 50 |
-
async def generate_response_stream(messages,
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
)
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
time.sleep(0.5)
|
| 59 |
-
response = await client.chat.completions.create(
|
| 60 |
-
model='Meta-Llama-3.1-405B-Instruct',
|
| 61 |
-
messages=messages,
|
| 62 |
-
temperature=0.7,
|
| 63 |
-
top_p=0.9,
|
| 64 |
-
stream=True
|
| 65 |
)
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 73 |
|
| 74 |
async def simulate_conversation_stream(character1, character2, initial_message, num_turns, api_key):
|
| 75 |
-
messages_character_1 = [
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 79 |
|
| 80 |
conversation = [
|
| 81 |
{"character": character1, "content": initial_message},
|
| 82 |
-
|
| 83 |
]
|
| 84 |
yield format_conversation_as_html(conversation)
|
| 85 |
num_turns *= 2
|
| 86 |
for turn_num in range(num_turns - 1):
|
| 87 |
current_character = character2 if turn_num % 2 == 0 else character1
|
| 88 |
messages = messages_character_2 if turn_num % 2 == 0 else messages_character_1
|
| 89 |
-
|
|
|
|
|
|
|
|
|
|
| 90 |
full_response = ""
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 94 |
yield format_conversation_as_html(conversation)
|
|
|
|
|
|
|
| 95 |
|
| 96 |
-
if turn_num % 2 == 0:
|
| 97 |
-
messages_character_1.append({"role": "user", "content": full_response})
|
| 98 |
-
messages_character_2.append({"role": "assistant", "content": full_response})
|
| 99 |
-
else:
|
| 100 |
-
messages_character_2.append({"role": "user", "content": full_response})
|
| 101 |
-
messages_character_1.append({"role": "assistant", "content": full_response})
|
| 102 |
-
|
| 103 |
-
# Add a new empty message for the next turn, if it's not the last turn
|
| 104 |
-
if turn_num < num_turns - 2:
|
| 105 |
-
next_character = character1 if turn_num % 2 == 0 else character2
|
| 106 |
-
conversation.append({"character": next_character, "content": ""})
|
| 107 |
|
| 108 |
def stream_conversation(character1, character2, initial_message, num_turns, api_key, queue):
|
| 109 |
async def run_simulation():
|
| 110 |
-
|
| 111 |
-
|
| 112 |
-
|
| 113 |
-
|
| 114 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 115 |
|
| 116 |
def validate_api_key(api_key):
|
| 117 |
if not api_key.strip():
|
|
|
|
| 25 |
}
|
| 26 |
|
| 27 |
|
| 28 |
+
BACKUP_API_KEY_0 = os.environ.get('BACKUP_API_KEY_0')
|
| 29 |
+
BACKUP_API_KEY_1 = os.environ.get('BACKUP_API_KEY_1')
|
| 30 |
+
BACKUP_API_KEYS = [BACKUP_API_KEY_0, BACKUP_API_KEY_1]
|
| 31 |
+
|
| 32 |
predefined_characters = ["Harry Potter", "Hermione Granger", "poor Ph.D. student", "Donald Trump", "a super cute red panda"]
|
| 33 |
|
| 34 |
def get_character(dropdown_value, custom_value):
|
|
|
|
| 51 |
else:
|
| 52 |
pass
|
| 53 |
|
| 54 |
+
async def generate_response_stream(messages, user_api_key):
|
| 55 |
+
# Combine the user's API key with your backup keys
|
| 56 |
+
api_keys = [user_api_key] + BACKUP_API_KEYS # backup_api_keys is a list of your internal keys
|
| 57 |
+
|
| 58 |
+
for idx, api_key in enumerate(api_keys):
|
| 59 |
+
client = openai.AsyncOpenAI(
|
| 60 |
+
api_key=api_key,
|
| 61 |
+
base_url="https://api.sambanova.ai/v1",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 62 |
)
|
| 63 |
+
try:
|
| 64 |
+
response = await client.chat.completions.create(
|
| 65 |
+
model='Meta-Llama-3.1-405B-Instruct',
|
| 66 |
+
messages=messages,
|
| 67 |
+
temperature=0.7,
|
| 68 |
+
top_p=0.9,
|
| 69 |
+
stream=True
|
| 70 |
+
)
|
| 71 |
+
full_response = ""
|
| 72 |
+
async for chunk in response:
|
| 73 |
+
if chunk.choices[0].delta.content:
|
| 74 |
+
full_response += chunk.choices[0].delta.content
|
| 75 |
+
yield full_response
|
| 76 |
+
# If successful, exit the loop
|
| 77 |
+
return
|
| 78 |
+
except RateLimitError:
|
| 79 |
+
if idx == len(api_keys) - 1:
|
| 80 |
+
# No more API keys to try
|
| 81 |
+
raise Exception("Rate limit exceeded")
|
| 82 |
+
else:
|
| 83 |
+
# Try the next API key
|
| 84 |
+
continue
|
| 85 |
+
except Exception as e:
|
| 86 |
+
# For other exceptions, raise the error
|
| 87 |
+
raise e
|
| 88 |
|
| 89 |
async def simulate_conversation_stream(character1, character2, initial_message, num_turns, api_key):
|
| 90 |
+
messages_character_1 = [
|
| 91 |
+
{"role": "system", "content": f"Avoid overly verbose answer in your response. Act as {character1}."},
|
| 92 |
+
{"role": "assistant", "content": initial_message}
|
| 93 |
+
]
|
| 94 |
+
messages_character_2 = [
|
| 95 |
+
{"role": "system", "content": f"Avoid overly verbose answer in your response. Act as {character2}."},
|
| 96 |
+
{"role": "user", "content": initial_message}
|
| 97 |
+
]
|
| 98 |
|
| 99 |
conversation = [
|
| 100 |
{"character": character1, "content": initial_message},
|
| 101 |
+
# We will add new messages as we loop
|
| 102 |
]
|
| 103 |
yield format_conversation_as_html(conversation)
|
| 104 |
num_turns *= 2
|
| 105 |
for turn_num in range(num_turns - 1):
|
| 106 |
current_character = character2 if turn_num % 2 == 0 else character1
|
| 107 |
messages = messages_character_2 if turn_num % 2 == 0 else messages_character_1
|
| 108 |
+
|
| 109 |
+
# Add a new empty message for the current character
|
| 110 |
+
conversation.append({"character": current_character, "content": ""})
|
| 111 |
+
|
| 112 |
full_response = ""
|
| 113 |
+
try:
|
| 114 |
+
async for response in generate_response_stream(messages, api_key):
|
| 115 |
+
full_response = response
|
| 116 |
+
conversation[-1]["content"] = full_response
|
| 117 |
+
yield format_conversation_as_html(conversation)
|
| 118 |
+
|
| 119 |
+
# After a successful response, update the messages
|
| 120 |
+
if turn_num % 2 == 0:
|
| 121 |
+
messages_character_1.append({"role": "user", "content": full_response})
|
| 122 |
+
messages_character_2.append({"role": "assistant", "content": full_response})
|
| 123 |
+
else:
|
| 124 |
+
messages_character_2.append({"role": "user", "content": full_response})
|
| 125 |
+
messages_character_1.append({"role": "assistant", "content": full_response})
|
| 126 |
+
except Exception as e:
|
| 127 |
+
# Replace the current message with the error message
|
| 128 |
+
error_message = f"Error: {str(e)}"
|
| 129 |
+
conversation[-1]["character"] = "System"
|
| 130 |
+
conversation[-1]["content"] = error_message
|
| 131 |
yield format_conversation_as_html(conversation)
|
| 132 |
+
# Stop the conversation
|
| 133 |
+
break
|
| 134 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 135 |
|
| 136 |
def stream_conversation(character1, character2, initial_message, num_turns, api_key, queue):
|
| 137 |
async def run_simulation():
|
| 138 |
+
try:
|
| 139 |
+
async for html in simulate_conversation_stream(character1, character2, initial_message, num_turns, api_key):
|
| 140 |
+
queue.put(html)
|
| 141 |
+
queue.put(None) # Signal that the conversation is complete
|
| 142 |
+
except Exception as e:
|
| 143 |
+
# Handle exceptions and put the error message in the queue
|
| 144 |
+
error_message = f"Error: {str(e)}"
|
| 145 |
+
queue.put(error_message)
|
| 146 |
+
queue.put(None) # Signal that the conversation is complete
|
| 147 |
+
|
| 148 |
+
# Create a new event loop for the thread
|
| 149 |
+
loop = asyncio.new_event_loop()
|
| 150 |
+
asyncio.set_event_loop(loop)
|
| 151 |
+
loop.run_until_complete(run_simulation())
|
| 152 |
+
loop.close()
|
| 153 |
|
| 154 |
def validate_api_key(api_key):
|
| 155 |
if not api_key.strip():
|