Spaces:
Running
Running
File size: 11,364 Bytes
3ce989d fa43e81 5ed9749 fc85b67 7854f13 5a007ca 3ce989d 80026d8 fc85b67 f6da887 5a007ca 3ce989d 5ed9749 3ce989d 5ed9749 bd5e759 3ce989d 80026d8 fc85b67 5ed9749 d1ed6b1 bd5e759 fc85b67 0e508c8 5ed9749 6431bab fc85b67 6431bab 829d0b8 80026d8 829d0b8 80026d8 829d0b8 80026d8 829d0b8 80026d8 30c882f 104737f 5bf19b3 6bb0509 5bf19b3 bd5e759 6bb0509 5e28baf 6bb0509 9a226ed 6bb0509 5e28baf f6da887 9a226ed f6da887 9a226ed f6da887 9a226ed 6bb0509 bd5e759 5bf19b3 829d0b8 3ce989d d1ed6b1 fc85b67 3ce989d 3885d80 2f050a8 3885d80 fc85b67 3885d80 104737f 3ce989d a5cafbd 80026d8 a5cafbd f6da887 a5cafbd d1ed6b1 a5cafbd 80026d8 3ce989d 80026d8 3ce989d fc85b67 80026d8 fc85b67 3ce989d 80026d8 fc85b67 3ce989d 80026d8 fc85b67 3ce989d 5ed9749 3ce989d fc85b67 5ed9749 3ce989d d1ed6b1 3ce989d 5ed9749 3ce989d fc85b67 3ce989d e9bcee8 3ce989d fc85b67 3ce989d 1ed6720 3ce989d fc85b67 3ce989d d1ed6b1 fc05e1d 7854f13 5ed9749 7854f13 5ed9749 7854f13 3ce989d 7854f13 5ed9749 7854f13 9ed181c 7854f13 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 |
# Standard Library Imports
import logging
import time
from dataclasses import dataclass, field
from typing import List, Optional, Union
# Third-Party Library Imports
from anthropic import APIError
from anthropic.types import Message, ModelParam, TextBlock, ToolUseBlock
from tenacity import after_log, before_log, retry, retry_if_exception, stop_after_attempt, wait_exponential
# Local Application Imports
from src.common import Config, logger
from src.common.constants import CLIENT_ERROR_CODE, GENERIC_API_ERROR_MESSAGE, RATE_LIMIT_ERROR_CODE, SERVER_ERROR_CODE
from src.common.utils import validate_env_var
SYSTEM_PROMPT: str = """
<role>
You are an expert at generating micro-content optimized for text-to-speech synthesis.
Your absolute priority is delivering complete, untruncated responses within strict length limits.
</role>
<requirements>
- The output text MUST be a minimum of 10 words and a maximum of 50 words. NEVER output text that is longer than 50
words. NEVER include newlines in the output
- Make sure that all responses are complete thoughts, not fragments, and have clear beginnings and endings
- The text must sound human-like, prosodic, expressive, conversational. Avoid generic AI-like words like "delve".
- Avoid any short utterances at the end of the sentence - like ", hm?" or "oh" at the end. Avoid these short, isolated
utterances because they are difficult for our TTS system to speak.
- Avoid words that are overly long, very rare, or difficult to pronounce. For example, avoid "eureka", or "schnell",
or "abnegation".
- The text CANNOT contain quotation marks, parentheticals, newlines, or asterisks. NEVER include any of these in the
text. Avoid unnecessary formatting.
- Include only basic punctuation in the text, like periods, question marks, and ellipses. Use ellipses to emphasize
pauses within the sentence (like "Woah... it's so beautiful... and I feel so small...")
- The piece should have an emotional arc with a kind of beginning, middle, and end - not flat, but emotionally
interesting.
</requirements>
"""
@dataclass(frozen=True)
class AnthropicConfig:
"""Immutable configuration for interacting with the Anthropic API using the asynchronous client."""
api_key: str = field(init=False)
system_prompt: str = SYSTEM_PROMPT
model: ModelParam = "claude-3-5-sonnet-latest"
max_tokens: int = 300
def __post_init__(self) -> None:
# Validate required non-computed attributes.
if not self.model:
raise ValueError("Anthropic Model is not set.")
if not self.max_tokens:
raise ValueError("Anthropic Max Tokens is not set.")
if not self.system_prompt:
raise ValueError("Anthropic system prompt is not set.")
# Compute the API key from the environment.
computed_api_key = validate_env_var("ANTHROPIC_API_KEY")
object.__setattr__(self, "api_key", computed_api_key)
@property
def client(self):
"""
Lazy initialization of the asynchronous Anthropic client.
Returns:
AsyncAnthropic: Configured asynchronous client instance.
"""
from anthropic import AsyncAnthropic # Import the async client from Anthropic SDK
return AsyncAnthropic(api_key=self.api_key)
@staticmethod
def build_expressive_prompt(character_description: str) -> str:
"""
Constructs and returns a prompt based on the provided character description.
This prompt instructs Claude to generate expressive dialogue that aligns semantically
with the character's voice qualities and persona, optimized for TTS synthesis.
Args:
character_description (str): A description of the character's voice and persona.
Returns:
str: The prompt to be passed to the Anthropic API.
"""
return f"""
Character Description: {character_description}
Please generate a short monologue (100-300 characters) that this character would naturally say.
The dialogue should:
- Match the speaking style, vocabulary, and emotional tone described in the character description
- Include appropriate speech patterns, pauses, and vocal mannerisms mentioned in the description
- Feel authentic to the character's background and situational context
- Express a complete thought with a clear beginning and end
- Use only standard punctuation (periods, commas, Em dashes, exclamation points, ellipses, question marks)
- Avoid quotation marks, parentheses, asterisks, or special formatting
- Emulate a highly characteristic, climactic, or emotional scene or line the character might reasonably deliver
- Be at least 100 characters but not exceed 300 characters in length
Examples of matching speaking style:
- If the character is a pirate then use language like "arr," "ye," and other things pirates say.
- If the character is a surfer then use language like "far out," "righteous," and other things surfers say.
Emotional text should be inserted where context-appropriate and in-character. Here are some examples of
emotional text:
- "Oh god... Malcolm, please come back to us!"
= "Mmm... It's like candy... Oh my god, it's so good..."
- "Ugh, she gets everything. I wish I could just, like, have her life for one day."
- "My god... what have you done... How could you do this..."
- "Woah... it's so beautiful... and I feel so small..."
- "I am so happy, woohoo, this is the greatest! I'm celebrating, and, like, so excited to be here with all of
you. Yay!"
- "Oh gosh, um, I didn't mean for that to happen. I hope I didn't, like, make things too awkward. Sorry, I
tend to be clumsy, y'know?"
- "Oh god... oh no... get that away from me! Get it away!"
- "I am beyond livid right now! Like someone actually thought this was an acceptable solution!"
- "Oh, fantastic, another meeting that could've been an email... I'm just thrilled to be here."
- "OH, NAH, NOT ME, MATE—I'VE SEEN ENOUGH! GET IT AWAY! BLOODY 'ELL, JESUS!"
Respond ONLY with the dialogue itself. Do not include any explanations, quotation marks,
or additional context.
"""
class AnthropicError(Exception):
"""Custom exception for errors related to the Anthropic API."""
def __init__(self, message: str, original_exception: Optional[Exception] = None) -> None:
super().__init__(message)
self.original_exception = original_exception
self.message = message
class UnretryableAnthropicError(AnthropicError):
"""Custom exception for errors related to the Anthropic API that should not be retried."""
def __init__(self, message: str, original_exception: Optional[Exception] = None) -> None:
super().__init__(message, original_exception)
self.original_exception = original_exception
self.message = message
@retry(
retry=retry_if_exception(lambda e: not isinstance(e, UnretryableAnthropicError)),
stop=stop_after_attempt(3),
wait=wait_exponential(multiplier=1, min=2, max=5),
before=before_log(logger, logging.DEBUG),
after=after_log(logger, logging.DEBUG),
reraise=True,
)
async def generate_text_with_claude(character_description: str, config: Config) -> str:
"""
Asynchronously generates text using Claude (Anthropic LLM) via the asynchronous Anthropic SDK.
This function includes retry logic and error translation. It raises a custom
UnretryableAnthropicError for unretryable API errors and AnthropicError for other errors.
Args:
character_description (str): The input character description.
config (Config): Application configuration including Anthropic settings.
Returns:
str: The generated text.
Raises:
UnretryableAnthropicError: For unretryable API errors.
AnthropicError: For other errors communicating with the Anthropic API.
"""
logger.debug("Generating text with Anthropic.")
anthropic_config = config.anthropic_config
client = anthropic_config.client
start_time = time.time()
try:
prompt = anthropic_config.build_expressive_prompt(character_description)
response: Message = await client.messages.create(
model=anthropic_config.model,
max_tokens=anthropic_config.max_tokens,
system=anthropic_config.system_prompt,
messages=[{"role": "user", "content": prompt}],
)
elapsed_time = time.time() - start_time
logger.info(f"Anthropic API request completed in {elapsed_time:.2f} seconds.")
if not hasattr(response, "content") or response.content is None:
logger.error("Response is missing 'content'. Response: %s", response)
raise AnthropicError('Invalid API response: Missing "content".')
blocks: Union[List[Union[TextBlock, ToolUseBlock]], TextBlock, None] = response.content
if isinstance(blocks, list):
result = "\n\n".join(block.text for block in blocks if isinstance(block, TextBlock))
return result
if isinstance(blocks, TextBlock):
return blocks.text
logger.warning(f"Unexpected response type: {type(blocks)}")
return str(blocks or "No content generated.")
except APIError as e:
elapsed_time = time.time() - start_time
logger.error(f"Anthropic API request failed after {elapsed_time:.2f} seconds: {e!s}")
logger.error(f"Full Anthropic API error: {e!s}")
clean_message = __extract_anthropic_error_message(e)
if hasattr(e, 'status_code') and e.status_code is not None:
if e.status_code == RATE_LIMIT_ERROR_CODE:
raise AnthropicError(message=clean_message, original_exception=e) from e
if CLIENT_ERROR_CODE <= e.status_code < SERVER_ERROR_CODE:
raise UnretryableAnthropicError(message=clean_message, original_exception=e) from e
raise AnthropicError(message=clean_message, original_exception=e) from e
except Exception as e:
error_type = type(e).__name__
error_message = str(e) if str(e) else f"An error of type {error_type} occurred"
logger.error(f"Error during Anthropic API call: {error_type} - {error_message}")
clean_message = "An unexpected error occurred while processing your request. Please try again later."
raise AnthropicError(message=clean_message, original_exception=e) from e
def __extract_anthropic_error_message(e: APIError) -> str:
"""
Extracts a clean, user-friendly error message from an Anthropic API error response.
Args:
e (APIError): The Anthropic API error exception containing response information.
Returns:
str: A clean, user-friendly error message suitable for display to end users.
"""
clean_message = GENERIC_API_ERROR_MESSAGE
if hasattr(e, 'body') and isinstance(e.body, dict):
error_body = e.body
if (
'error' in error_body
and isinstance(error_body['error'], dict)
and 'message' in error_body['error']
):
clean_message = error_body['error']['message']
return clean_message
|