Spaces:
Sleeping
Sleeping
File size: 14,704 Bytes
225a75e |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 |
#!/usr/bin/env python3
"""
HuggingFace Qwen 2.5 Model Client
Handles inference for router, main, and complex models with cost tracking
"""
import os
import time
import logging
from typing import Dict, Any, List, Optional
from dataclasses import dataclass
from enum import Enum
from huggingface_hub import InferenceClient
from langchain_huggingface import HuggingFaceEndpoint
from langchain_core.language_models.llms import LLM
# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
class ModelTier(Enum):
"""Model complexity tiers for cost optimization"""
ROUTER = "router" # 3B - Fast, cheap routing decisions
MAIN = "main" # 14B - Balanced performance
COMPLEX = "complex" # 32B - Best performance for hard tasks
@dataclass
class ModelConfig:
"""Configuration for each Qwen model"""
name: str
tier: ModelTier
max_tokens: int
temperature: float
cost_per_token: float # Estimated cost per token
timeout: int
@dataclass
class InferenceResult:
"""Result of model inference with metadata"""
response: str
model_used: str
tokens_used: int
cost_estimate: float
response_time: float
success: bool
error: Optional[str] = None
class QwenClient:
"""HuggingFace client for Qwen 2.5 model family"""
def __init__(self, hf_token: Optional[str] = None):
"""Initialize the Qwen client with HuggingFace token"""
self.hf_token = hf_token or os.getenv("HUGGINGFACE_TOKEN")
if not self.hf_token:
logger.warning("No HuggingFace token provided. API access may be limited.")
# Define model configurations - Updated with best available models
self.models = {
ModelTier.ROUTER: ModelConfig(
name="Qwen/Qwen2.5-7B-Instruct", # Fast router for classification
tier=ModelTier.ROUTER,
max_tokens=512,
temperature=0.1,
cost_per_token=0.0003, # 7B model
timeout=15
),
ModelTier.MAIN: ModelConfig(
name="Qwen/Qwen2.5-32B-Instruct", # 4.5x more powerful for main tasks
tier=ModelTier.MAIN,
max_tokens=1024,
temperature=0.1,
cost_per_token=0.0008, # Higher cost for 32B
timeout=25
),
ModelTier.COMPLEX: ModelConfig(
name="Qwen/Qwen2.5-72B-Instruct", # 10x more powerful for complex reasoning!
tier=ModelTier.COMPLEX,
max_tokens=2048,
temperature=0.1,
cost_per_token=0.0015, # Premium for 72B model
timeout=35
)
}
# Initialize clients
self.inference_clients = {}
self.langchain_clients = {}
self._initialize_clients()
# Cost tracking
self.total_cost = 0.0
self.request_count = 0
self.budget_limit = 0.10 # $0.10 total budget
def _initialize_clients(self):
"""Initialize HuggingFace clients for each model"""
for tier, config in self.models.items():
try:
# HuggingFace InferenceClient for direct API calls
self.inference_clients[tier] = InferenceClient(
model=config.name,
token=self.hf_token
)
# LangChain wrapper for integration
self.langchain_clients[tier] = HuggingFaceEndpoint(
repo_id=config.name,
max_new_tokens=config.max_tokens,
temperature=config.temperature,
huggingfacehub_api_token=self.hf_token,
timeout=config.timeout
)
logger.info(f"✅ Initialized {tier.value} model: {config.name}")
except Exception as e:
logger.error(f"❌ Failed to initialize {tier.value} model: {e}")
self.inference_clients[tier] = None
self.langchain_clients[tier] = None
def get_model_status(self) -> Dict[str, bool]:
"""Check which models are available"""
status = {}
for tier in ModelTier:
status[tier.value] = (
self.inference_clients.get(tier) is not None and
self.langchain_clients.get(tier) is not None
)
return status
def select_model_tier(self, complexity: str = "medium", budget_conscious: bool = True, question_text: str = "") -> ModelTier:
"""Smart model selection based on task complexity, budget, and question analysis"""
# Check budget constraints
budget_used_percent = (self.total_cost / self.budget_limit) * 100
if budget_conscious and budget_used_percent > 80:
logger.warning(f"Budget critical ({budget_used_percent:.1f}% used), forcing router model")
return ModelTier.ROUTER
elif budget_conscious and budget_used_percent > 60:
logger.warning(f"Budget warning ({budget_used_percent:.1f}% used), limiting complex model usage")
complexity = "simple" if complexity == "complex" else complexity
# Enhanced complexity analysis based on question content
if question_text:
question_lower = question_text.lower()
# Indicators for complex reasoning (use 72B model)
complex_indicators = [
"analyze", "explain why", "reasoning", "logic", "complex", "difficult",
"multi-step", "calculate and explain", "compare and contrast",
"what is the relationship", "how does", "why is", "prove that",
"step by step", "detailed analysis", "comprehensive"
]
# Indicators for simple tasks (use 7B model)
simple_indicators = [
"what is", "who is", "when", "where", "simple", "quick",
"yes or no", "true or false", "list", "name", "find"
]
# Math and coding indicators (use 32B model - good balance)
math_indicators = [
"calculate", "compute", "solve", "equation", "formula", "math",
"number", "total", "sum", "average", "percentage", "code", "program"
]
# File processing indicators (use 32B+ models)
file_indicators = [
"image", "picture", "photo", "audio", "sound", "video", "file",
"document", "excel", "csv", "data", "chart", "graph"
]
# Count indicators
complex_score = sum(1 for indicator in complex_indicators if indicator in question_lower)
simple_score = sum(1 for indicator in simple_indicators if indicator in question_lower)
math_score = sum(1 for indicator in math_indicators if indicator in question_lower)
file_score = sum(1 for indicator in file_indicators if indicator in question_lower)
# Auto-detect complexity based on content
if complex_score >= 2 or len(question_text) > 200:
complexity = "complex"
elif file_score >= 1 or math_score >= 2:
complexity = "medium"
elif simple_score >= 2 and complex_score == 0:
complexity = "simple"
# Select based on complexity with budget awareness
if complexity == "complex" and budget_used_percent < 70:
selected_tier = ModelTier.COMPLEX
elif complexity == "simple" or budget_used_percent > 75:
selected_tier = ModelTier.ROUTER
else:
selected_tier = ModelTier.MAIN
# Fallback if selected model unavailable
if not self.inference_clients.get(selected_tier):
logger.warning(f"Selected model {selected_tier.value} unavailable, falling back")
for fallback in [ModelTier.MAIN, ModelTier.ROUTER, ModelTier.COMPLEX]:
if self.inference_clients.get(fallback):
selected_tier = fallback
break
else:
raise RuntimeError("No models available")
# Log selection reasoning
logger.info(f"Selected {selected_tier.value} model (complexity: {complexity}, budget: {budget_used_percent:.1f}%)")
return selected_tier
async def generate_async(self,
prompt: str,
tier: Optional[ModelTier] = None,
max_tokens: Optional[int] = None) -> InferenceResult:
"""Async text generation with the specified model tier"""
if tier is None:
tier = self.select_model_tier()
config = self.models[tier]
client = self.inference_clients.get(tier)
if not client:
return InferenceResult(
response="",
model_used=config.name,
tokens_used=0,
cost_estimate=0.0,
response_time=0.0,
success=False,
error=f"Model {tier.value} not available"
)
start_time = time.time()
try:
# Use specified max_tokens or model default
tokens = max_tokens or config.max_tokens
# Use chat completion API for conversational models
messages = [{"role": "user", "content": prompt}]
response = client.chat_completion(
messages=messages,
model=config.name,
max_tokens=tokens,
temperature=config.temperature
)
response_time = time.time() - start_time
# Extract response from chat completion
if response and response.choices:
response_text = response.choices[0].message.content
else:
raise ValueError("No response received from model")
# Estimate tokens used (rough approximation)
estimated_tokens = len(prompt.split()) + len(response_text.split())
cost_estimate = estimated_tokens * config.cost_per_token
# Update tracking
self.total_cost += cost_estimate
self.request_count += 1
logger.info(f"✅ Generated response using {tier.value} model in {response_time:.2f}s")
return InferenceResult(
response=response_text,
model_used=config.name,
tokens_used=estimated_tokens,
cost_estimate=cost_estimate,
response_time=response_time,
success=True
)
except Exception as e:
response_time = time.time() - start_time
logger.error(f"❌ Generation failed with {tier.value} model: {e}")
return InferenceResult(
response="",
model_used=config.name,
tokens_used=0,
cost_estimate=0.0,
response_time=response_time,
success=False,
error=str(e)
)
def generate(self,
prompt: str,
tier: Optional[ModelTier] = None,
max_tokens: Optional[int] = None) -> InferenceResult:
"""Synchronous text generation (wrapper for async)"""
import asyncio
# Create event loop if needed
try:
loop = asyncio.get_event_loop()
except RuntimeError:
loop = asyncio.new_event_loop()
asyncio.set_event_loop(loop)
return loop.run_until_complete(
self.generate_async(prompt, tier, max_tokens)
)
def get_langchain_llm(self, tier: ModelTier) -> Optional[LLM]:
"""Get LangChain LLM instance for agent integration"""
return self.langchain_clients.get(tier)
def get_usage_stats(self) -> Dict[str, Any]:
"""Get current usage and cost statistics"""
return {
"total_cost": self.total_cost,
"request_count": self.request_count,
"budget_limit": self.budget_limit,
"budget_remaining": self.budget_limit - self.total_cost,
"budget_used_percent": (self.total_cost / self.budget_limit) * 100,
"average_cost_per_request": self.total_cost / max(self.request_count, 1),
"models_available": self.get_model_status()
}
def reset_usage_tracking(self):
"""Reset usage statistics (for testing/development)"""
self.total_cost = 0.0
self.request_count = 0
logger.info("Usage tracking reset")
# Test functions
def test_model_connection(client: QwenClient, tier: ModelTier):
"""Test connection to a specific model tier"""
test_prompt = "Hello! Please respond with 'Connection successful' if you can read this."
logger.info(f"Testing {tier.value} model...")
result = client.generate(test_prompt, tier=tier, max_tokens=50)
if result.success:
logger.info(f"✅ {tier.value} model test successful: {result.response[:50]}...")
logger.info(f" Response time: {result.response_time:.2f}s")
logger.info(f" Cost estimate: ${result.cost_estimate:.6f}")
else:
logger.error(f"❌ {tier.value} model test failed: {result.error}")
return result.success
def test_all_models():
"""Test all available models"""
logger.info("🧪 Testing all Qwen models...")
client = QwenClient()
results = {}
for tier in ModelTier:
results[tier] = test_model_connection(client, tier)
logger.info("📊 Test Results Summary:")
for tier, success in results.items():
status = "✅ PASS" if success else "❌ FAIL"
logger.info(f" {tier.value:8}: {status}")
logger.info("💰 Usage Statistics:")
stats = client.get_usage_stats()
for key, value in stats.items():
if key != "models_available":
logger.info(f" {key}: {value}")
return results
if __name__ == "__main__":
# Load environment variables for testing
from dotenv import load_dotenv
load_dotenv()
# Run tests when script executed directly
test_all_models() |