Spaces:
Sleeping
Sleeping
File size: 9,913 Bytes
af383cf | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 | """LLM client utilities for async API calls."""
import asyncio
import logging
import re
from typing import Optional, Any
from abc import ABC, abstractmethod
import json
from openai import AsyncOpenAI, OpenAI
from openai import RateLimitError as OpenAIRateLimitError
from anthropic import Anthropic, AsyncAnthropic
from tenacity import (
retry,
retry_if_exception_type,
wait_exponential,
stop_after_attempt,
before_sleep_log,
)
from ..config import get_settings
logger = logging.getLogger(__name__)
def _parse_retry_after(exc: Exception) -> float:
"""Extract 'retry after N seconds' from a Groq 429 error message."""
msg = str(exc)
match = re.search(r"try again in ([\d\.]+)s", msg)
if match:
return min(float(match.group(1)) + 1.0, 60.0)
return 5.0
async def _groq_retry_wait(retry_state):
"""Custom wait that honours the 'retry after Xs' hint in the error."""
exc = retry_state.outcome.exception()
secs = _parse_retry_after(exc) if exc else 5.0
logger.warning(f"Rate limited — waiting {secs:.1f}s before retry")
await asyncio.sleep(secs)
_groq_retry = retry(
retry=retry_if_exception_type(OpenAIRateLimitError),
wait=wait_exponential(multiplier=1, min=2, max=60),
stop=stop_after_attempt(8),
before_sleep=before_sleep_log(logger, logging.WARNING),
reraise=False,
)
class LLMClient(ABC):
"""Abstract base class for LLM clients."""
@abstractmethod
async def generate(
self,
system_prompt: str,
user_message: str,
temperature: float = 0.1,
top_p: float = 0.9,
max_tokens: int = 2000,
) -> str:
"""Generate text from the LLM."""
pass
@abstractmethod
async def generate_json(
self,
system_prompt: str,
user_message: str,
schema: dict,
temperature: float = 0.1,
top_p: float = 0.9,
max_tokens: int = 2000,
) -> dict:
"""Generate structured JSON from the LLM."""
pass
class OpenAIClient(LLMClient):
"""Async OpenAI client wrapper."""
def __init__(self, api_key: Optional[str] = None, model: Optional[str] = None):
settings = get_settings()
self.client = AsyncOpenAI(api_key=api_key or settings.openai_api_key)
self.model = model or settings.openai_model
async def generate(
self,
system_prompt: str,
user_message: str,
temperature: float = 0.1,
top_p: float = 0.9,
max_tokens: int = 2000,
) -> str:
"""Generate text using OpenAI API."""
try:
response = await self.client.chat.completions.create(
model=self.model,
messages=[
{"role": "system", "content": system_prompt},
{"role": "user", "content": user_message},
],
temperature=temperature,
top_p=top_p,
max_tokens=max_tokens,
)
return response.choices[0].message.content or ""
except Exception as e:
logger.error(f"OpenAI API error: {e}")
raise
async def generate_json(
self,
system_prompt: str,
user_message: str,
schema: dict,
temperature: float = 0.1,
top_p: float = 0.9,
max_tokens: int = 2000,
) -> dict:
"""Generate structured JSON using OpenAI API."""
try:
response = await self.client.chat.completions.create(
model=self.model,
messages=[
{"role": "system", "content": system_prompt},
{"role": "user", "content": user_message},
],
temperature=temperature,
top_p=top_p,
max_tokens=max_tokens,
response_format={
"type": "json_schema",
"json_schema": {
"name": "evaluation_result",
"schema": schema,
"strict": True,
},
},
)
content = response.choices[0].message.content or "{}"
return json.loads(content)
except Exception as e:
logger.error(f"OpenAI JSON generation error: {e}")
raise
class AnthropicClient(LLMClient):
"""Async Anthropic client wrapper."""
def __init__(self, api_key: Optional[str] = None, model: Optional[str] = None):
settings = get_settings()
self.client = AsyncAnthropic(api_key=api_key or settings.anthropic_api_key)
self.model = model or settings.anthropic_model
async def generate(
self,
system_prompt: str,
user_message: str,
temperature: float = 0.1,
top_p: float = 0.9,
max_tokens: int = 2000,
) -> str:
"""Generate text using Anthropic API."""
try:
response = await self.client.messages.create(
model=self.model,
max_tokens=max_tokens,
system=system_prompt,
messages=[
{"role": "user", "content": user_message},
],
temperature=temperature,
top_p=top_p,
)
return response.content[0].text
except Exception as e:
logger.error(f"Anthropic API error: {e}")
raise
async def generate_json(
self,
system_prompt: str,
user_message: str,
schema: dict,
temperature: float = 0.1,
top_p: float = 0.9,
max_tokens: int = 2000,
) -> dict:
"""Generate structured JSON using Anthropic API."""
# Anthropic doesn't have native JSON mode, so we format the request carefully
json_instructions = f"""
Please respond with valid JSON matching this schema:
{json.dumps(schema, indent=2)}
Only return the JSON object, no other text.
"""
try:
response = await self.client.messages.create(
model=self.model,
max_tokens=max_tokens,
system=system_prompt + "\n" + json_instructions,
messages=[
{"role": "user", "content": user_message},
],
temperature=temperature,
top_p=top_p,
)
content = response.content[0].text
# Try to parse JSON from the response
return json.loads(content)
except Exception as e:
logger.error(f"Anthropic JSON generation error: {e}")
raise
class GroqClient(LLMClient):
"""Async Groq client wrapper.
Groq exposes an OpenAI-compatible REST API, so we reuse AsyncOpenAI with a
custom base_url. Note: Groq does NOT support the ``response_format`` JSON
schema mode, so ``generate_json`` falls back to prompt-level instructions
and normal JSON parsing (same approach as AnthropicClient).
"""
def __init__(self, api_key: Optional[str] = None, model: Optional[str] = None):
settings = get_settings()
self.client = AsyncOpenAI(
api_key=api_key or settings.groq_api_key,
base_url="https://api.groq.com/openai/v1",
)
self.model = model or settings.groq_model
@_groq_retry
async def generate(
self,
system_prompt: str,
user_message: str,
temperature: float = 0.1,
top_p: float = 0.9,
max_tokens: int = 2000,
) -> str:
"""Generate text using Groq API (auto-retries on 429)."""
response = await self.client.chat.completions.create(
model=self.model,
messages=[
{"role": "system", "content": system_prompt},
{"role": "user", "content": user_message},
],
temperature=temperature,
top_p=top_p,
max_tokens=max_tokens,
)
return response.choices[0].message.content or ""
@_groq_retry
async def generate_json(
self,
system_prompt: str,
user_message: str,
schema: dict,
temperature: float = 0.1,
top_p: float = 0.9,
max_tokens: int = 2000,
) -> dict:
"""Generate structured JSON using Groq API (prompt-based, auto-retries on 429)."""
json_instructions = (
f"\nRespond with valid JSON matching this schema:\n"
f"{json.dumps(schema, indent=2)}\n"
"Only return the JSON object, no other text."
)
response = await self.client.chat.completions.create(
model=self.model,
messages=[
{"role": "system", "content": system_prompt + json_instructions},
{"role": "user", "content": user_message},
],
temperature=temperature,
top_p=top_p,
max_tokens=max_tokens,
)
content = response.choices[0].message.content or "{}"
return json.loads(content)
def get_llm_client(provider: str = "groq", model: Optional[str] = None) -> LLMClient:
"""Get an LLM client based on provider name.
Args:
provider: One of "openai", "anthropic", "groq".
model: Optional model override. If None, uses the provider's default
from settings. Pass a specific model name to use a different
model than the default (e.g. a cheaper judge model).
"""
if provider.lower() == "openai":
return OpenAIClient(model=model)
elif provider.lower() == "anthropic":
return AnthropicClient(model=model)
elif provider.lower() == "groq":
return GroqClient(model=model)
else:
raise ValueError(f"Unknown provider: {provider}. Choose from: openai, anthropic, groq")
|