Spaces:
Sleeping
Sleeping
File size: 10,271 Bytes
7644eac |
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 |
"""
Observability utilities for LLM monitoring and evaluation.
Provides centralized logging, tracing, and metrics tracking using LangSmith and W&B.
"""
import os
import time
import functools
from typing import Dict, Any, Optional, Callable
from datetime import datetime
import logging
# Import configuration
from src.utils.config import (
LANGCHAIN_TRACING_V2,
LANGCHAIN_API_KEY,
LANGCHAIN_PROJECT,
WANDB_API_KEY,
WANDB_PROJECT,
WANDB_ENTITY,
WANDB_MODE
)
logger = logging.getLogger(__name__)
class ObservabilityManager:
"""
Centralized manager for observability tools (LangSmith + W&B).
Handles initialization, tracing, and metrics logging.
"""
def __init__(self):
"""Initialize observability tools."""
self.langsmith_enabled = False
self.wandb_enabled = False
self.wandb_run = None
# Initialize LangSmith
self._init_langsmith()
# Initialize Weights & Biases
self._init_wandb()
def _init_langsmith(self):
"""Initialize LangSmith tracing."""
if LANGCHAIN_TRACING_V2 and LANGCHAIN_API_KEY:
try:
# Set environment variables for automatic LangChain tracing
os.environ["LANGCHAIN_TRACING_V2"] = "true"
os.environ["LANGCHAIN_API_KEY"] = LANGCHAIN_API_KEY
os.environ["LANGCHAIN_PROJECT"] = LANGCHAIN_PROJECT
self.langsmith_enabled = True
logger.info(f"✅ LangSmith tracing enabled for project: {LANGCHAIN_PROJECT}")
except Exception as e:
logger.warning(f"⚠️ Failed to initialize LangSmith: {e}")
self.langsmith_enabled = False
else:
logger.info("ℹ️ LangSmith tracing disabled (set LANGCHAIN_TRACING_V2=true and LANGCHAIN_API_KEY to enable)")
def _init_wandb(self):
"""Initialize Weights & Biases."""
if WANDB_API_KEY and WANDB_MODE != "disabled":
try:
import wandb
# Initialize W&B run
self.wandb_run = wandb.init(
project=WANDB_PROJECT,
entity=WANDB_ENTITY,
mode=WANDB_MODE,
config={
"framework": "langchain",
"application": "ai-learning-path-generator"
},
# Don't reinitialize if already running
reinit=False,
resume="allow"
)
self.wandb_enabled = True
logger.info(f"✅ W&B tracking enabled for project: {WANDB_PROJECT}")
except ImportError:
logger.warning("⚠️ wandb package not installed. Run: pip install wandb")
self.wandb_enabled = False
except Exception as e:
logger.warning(f"⚠️ Failed to initialize W&B: {e}")
self.wandb_enabled = False
else:
logger.info("ℹ️ W&B tracking disabled (set WANDB_API_KEY to enable)")
def log_llm_call(
self,
prompt: str,
response: str,
model: str,
metadata: Optional[Dict[str, Any]] = None,
latency_ms: Optional[float] = None,
token_count: Optional[int] = None,
cost: Optional[float] = None
):
"""
Log an LLM call to W&B Prompts.
Args:
prompt: The input prompt sent to the LLM
response: The LLM's response
model: Model name (e.g., 'gpt-4o-mini')
metadata: Additional metadata (user_id, topic, etc.)
latency_ms: Response time in milliseconds
token_count: Total tokens used
cost: Estimated cost in USD
"""
if not self.wandb_enabled:
return
try:
import wandb
# Log to W&B Prompts table
prompts_table = wandb.Table(
columns=[
"timestamp", "model", "prompt", "response",
"latency_ms", "token_count", "cost", "metadata"
]
)
prompts_table.add_data(
datetime.utcnow().isoformat(),
model,
prompt[:500], # Truncate for display
response[:500],
latency_ms,
token_count,
cost,
str(metadata or {})
)
wandb.log({"llm_calls": prompts_table})
# Also log metrics separately for easier aggregation
if latency_ms:
wandb.log({"llm_latency_ms": latency_ms})
if token_count:
wandb.log({"llm_tokens": token_count})
if cost:
wandb.log({"llm_cost_usd": cost})
except Exception as e:
logger.warning(f"Failed to log LLM call to W&B: {e}")
def log_metric(self, name: str, value: float, metadata: Optional[Dict] = None):
"""
Log a custom metric to W&B.
Args:
name: Metric name (e.g., 'path_generation_success')
value: Metric value
metadata: Additional context
"""
if not self.wandb_enabled:
return
try:
import wandb
log_data = {name: value}
if metadata:
# Flatten metadata into the log
for key, val in metadata.items():
log_data[f"{name}_{key}"] = val
wandb.log(log_data)
except Exception as e:
logger.warning(f"Failed to log metric to W&B: {e}")
def log_event(self, event_name: str, properties: Optional[Dict] = None):
"""
Log a custom event to W&B.
Args:
event_name: Name of the event (e.g., 'path_generated', 'validation_failed')
properties: Event properties
"""
if not self.wandb_enabled:
return
try:
import wandb
wandb.log({
"event": event_name,
"timestamp": datetime.utcnow().isoformat(),
"properties": properties or {}
})
except Exception as e:
logger.warning(f"Failed to log event to W&B: {e}")
def finish(self):
"""Clean up and finish W&B run."""
if self.wandb_enabled and self.wandb_run:
try:
import wandb
wandb.finish()
logger.info("✅ W&B run finished")
except Exception as e:
logger.warning(f"Failed to finish W&B run: {e}")
# Global observability manager instance
_observability_manager = None
def get_observability_manager() -> ObservabilityManager:
"""Get or create the global observability manager instance."""
global _observability_manager
if _observability_manager is None:
_observability_manager = ObservabilityManager()
return _observability_manager
def traceable(
name: Optional[str] = None,
metadata: Optional[Dict[str, Any]] = None,
log_to_wandb: bool = True
):
"""
Decorator to trace function execution with LangSmith and log to W&B.
Usage:
@traceable(name="generate_learning_path", metadata={"version": "v2"})
def generate_path(topic: str) -> dict:
...
Args:
name: Custom name for the trace (defaults to function name)
metadata: Additional metadata to attach to the trace
log_to_wandb: Whether to also log execution metrics to W&B
"""
def decorator(func: Callable) -> Callable:
@functools.wraps(func)
def wrapper(*args, **kwargs):
obs_manager = get_observability_manager()
trace_name = name or func.__name__
# Start timing
start_time = time.time()
success = False
error = None
result = None
try:
# Execute the function
result = func(*args, **kwargs)
success = True
return result
except Exception as e:
error = str(e)
raise
finally:
# Calculate latency
latency_ms = (time.time() - start_time) * 1000
# Log to W&B if enabled
if log_to_wandb and obs_manager.wandb_enabled:
obs_manager.log_metric(
f"{trace_name}_latency_ms",
latency_ms,
metadata={
"success": success,
"error": error,
**(metadata or {})
}
)
obs_manager.log_metric(
f"{trace_name}_success",
1.0 if success else 0.0
)
return wrapper
return decorator
def estimate_cost(model: str, input_tokens: int, output_tokens: int) -> float:
"""
Estimate the cost of an LLM call based on token usage.
Args:
model: Model name
input_tokens: Number of input tokens
output_tokens: Number of output tokens
Returns:
Estimated cost in USD
"""
# Pricing per 1M tokens (as of 2024)
pricing = {
"gpt-4o-mini": {"input": 0.15, "output": 0.60},
"gpt-4o": {"input": 5.00, "output": 15.00},
"gpt-3.5-turbo": {"input": 0.50, "output": 1.50},
"gpt-4": {"input": 30.00, "output": 60.00},
}
# Default to gpt-4o-mini pricing if model not found
model_pricing = pricing.get(model, pricing["gpt-4o-mini"])
input_cost = (input_tokens / 1_000_000) * model_pricing["input"]
output_cost = (output_tokens / 1_000_000) * model_pricing["output"]
return input_cost + output_cost
|