Spaces:
Running
Running
File size: 11,029 Bytes
3193174 | 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 | """
Management of token, request, and time budgets.
Provides cost control at the graph level and at individual node level.
"""
from collections.abc import Callable
from datetime import UTC, datetime
from typing import Any
from pydantic import BaseModel, ConfigDict
__all__ = [
"Budget",
"BudgetConfig",
"BudgetTracker",
"NodeBudget",
]
class Budget(BaseModel):
"""Tracks a resource limit accounting for used and reserved amounts."""
limit: float
used: float = 0.0
reserved: float = 0.0
@property
def available(self) -> float:
"""Remaining available resource (excluding reserved amount)."""
return max(0.0, self.limit - self.used - self.reserved)
@property
def remaining(self) -> float:
"""Remaining resource ignoring the reserve (limit - used)."""
return max(0.0, self.limit - self.used)
@property
def usage_ratio(self) -> float:
"""Fraction of the resource limit that has been consumed."""
if self.limit <= 0:
return 0.0
return self.used / self.limit
@property
def is_exhausted(self) -> bool:
"""True if no available resource remains."""
return self.available <= 0
def can_spend(self, amount: float) -> bool:
"""Check whether the available resource is sufficient for the given amount."""
return self.available >= amount
def spend(self, amount: float) -> bool:
"""Consume the resource if available; return True on success."""
if not self.can_spend(amount):
return False
self.used += amount
return True
def reserve(self, amount: float) -> bool:
"""Reserve resource for a future operation."""
if self.available < amount:
return False
self.reserved += amount
return True
def release_reservation(self, amount: float) -> None:
"""Release a portion of the reservation."""
self.reserved = max(0.0, self.reserved - amount)
def commit_reservation(self, amount: float) -> None:
"""Move up to the given amount from reservation into usage."""
actual = min(amount, self.reserved)
self.reserved -= actual
self.used += actual
def reset(self) -> None:
"""Reset used and reserved amounts to zero."""
self.used = 0.0
self.reserved = 0.0
def to_dict(self) -> dict[str, Any]:
"""Serialize the budget to a dictionary."""
return {
"limit": self.limit,
"used": self.used,
"reserved": self.reserved,
"available": self.available,
"usage_ratio": self.usage_ratio,
}
class NodeBudget(BaseModel):
"""Per-node limits for tokens, requests, time, and message lengths."""
node_id: str
tokens: Budget | None = None
requests: Budget | None = None
time_seconds: Budget | None = None
max_prompt_length: int | None = None
max_response_length: int | None = None
def can_execute(self, estimated_tokens: int = 0) -> tuple[bool, str | None]:
"""Check whether a step can be executed given the estimated token count."""
if self.tokens and not self.tokens.can_spend(estimated_tokens):
return False, f"Token budget exhausted for node {self.node_id}"
if self.requests and not self.requests.can_spend(1):
return False, f"Request budget exhausted for node {self.node_id}"
return True, None
def record_usage(
self,
tokens: int = 0,
time_seconds: float = 0.0,
) -> None:
"""Record actual resource consumption for the node."""
if self.tokens:
self.tokens.spend(tokens)
if self.requests:
self.requests.spend(1)
if self.time_seconds:
self.time_seconds.spend(time_seconds)
def to_dict(self) -> dict[str, Any]:
"""Serialize the node budget to a dictionary."""
return {
"node_id": self.node_id,
"tokens": self.tokens.to_dict() if self.tokens else None,
"requests": self.requests.to_dict() if self.requests else None,
"time_seconds": self.time_seconds.to_dict() if self.time_seconds else None,
"limits": {
"max_prompt_length": self.max_prompt_length,
"max_response_length": self.max_response_length,
},
}
class BudgetConfig(BaseModel):
model_config = ConfigDict(arbitrary_types_allowed=True)
"""Configuration for global and per-component execution limits."""
total_token_limit: int | None = None
total_request_limit: int | None = None
total_time_limit_seconds: float | None = None
node_token_limit: int | None = None
node_request_limit: int | None = None
node_time_limit_seconds: float | None = None
max_prompt_length: int | None = None
max_response_length: int | None = None
warn_at_usage_ratio: float = 0.8
on_budget_warning: Callable[[str, Budget], None] | None = None
on_budget_exceeded: Callable[[str, Budget], None] | None = None
class BudgetTracker:
"""Tracks global and per-node budgets and issues warnings when thresholds are approached."""
def __init__(self, config: BudgetConfig | None = None):
self.config = config or BudgetConfig()
self._global_tokens = Budget(limit=float(self.config.total_token_limit or float("inf")))
self._global_requests = Budget(limit=float(self.config.total_request_limit or float("inf")))
self._global_time = Budget(limit=self.config.total_time_limit_seconds or float("inf"))
self._node_budgets: dict[str, NodeBudget] = {}
self._start_time: datetime | None = None
def start(self) -> None:
"""Record the start time for time-budget tracking."""
self._start_time = datetime.now(UTC)
def get_elapsed_seconds(self) -> float:
"""Return elapsed seconds since start() was called."""
if self._start_time is None:
return 0.0
return (datetime.now(UTC) - self._start_time).total_seconds()
def get_node_budget(self, node_id: str) -> NodeBudget:
"""Return (or create) the budget for the given node."""
if node_id not in self._node_budgets:
self._node_budgets[node_id] = NodeBudget(
node_id=node_id,
tokens=Budget(limit=float(self.config.node_token_limit or float("inf")))
if self.config.node_token_limit
else None,
requests=Budget(limit=float(self.config.node_request_limit or float("inf")))
if self.config.node_request_limit
else None,
time_seconds=Budget(limit=self.config.node_time_limit_seconds or float("inf"))
if self.config.node_time_limit_seconds
else None,
max_prompt_length=self.config.max_prompt_length,
max_response_length=self.config.max_response_length,
)
return self._node_budgets[node_id]
def can_execute(
self,
node_id: str,
estimated_tokens: int = 0,
) -> tuple[bool, str | None]:
"""Check whether a step can be executed considering both global and node-level limits."""
if self._global_time.is_exhausted:
elapsed = self.get_elapsed_seconds()
time_limit = self.config.total_time_limit_seconds
if time_limit is not None and elapsed >= time_limit:
return False, f"Time budget exhausted: {elapsed:.1f}s"
if not self._global_tokens.can_spend(estimated_tokens):
return (
False,
f"Global token budget exhausted: {self._global_tokens.used}/{self._global_tokens.limit}",
)
if not self._global_requests.can_spend(1):
return (
False,
f"Global request budget exhausted: {self._global_requests.used}/{self._global_requests.limit}",
)
node_budget = self.get_node_budget(node_id)
can, reason = node_budget.can_execute(estimated_tokens)
if not can:
return False, reason
return True, None
def record_usage(
self,
node_id: str,
prompt_tokens: int = 0,
completion_tokens: int = 0,
latency_seconds: float = 0.0,
) -> None:
"""Record actual consumption for a node and update global counters."""
total_tokens = prompt_tokens + completion_tokens
self._global_tokens.spend(total_tokens)
self._global_requests.spend(1)
node_budget = self.get_node_budget(node_id)
node_budget.record_usage(tokens=total_tokens, time_seconds=latency_seconds)
self._check_warnings()
def truncate_prompt(self, prompt: str) -> str:
"""Truncate prompt to the configured limit and append a truncation marker."""
if self.config.max_prompt_length and len(prompt) > self.config.max_prompt_length:
return prompt[: self.config.max_prompt_length] + "\n[TRUNCATED]"
return prompt
def truncate_response(self, response: str) -> str:
"""Truncate response to the configured limit and append a truncation marker."""
if self.config.max_response_length and len(response) > self.config.max_response_length:
return response[: self.config.max_response_length] + "\n[TRUNCATED]"
return response
def _check_warnings(self) -> None:
"""Invoke warning callbacks if the warn_at_usage_ratio threshold has been reached."""
if self.config.on_budget_warning:
if self._global_tokens.usage_ratio >= self.config.warn_at_usage_ratio:
self.config.on_budget_warning("tokens", self._global_tokens)
if self._global_requests.usage_ratio >= self.config.warn_at_usage_ratio:
self.config.on_budget_warning("requests", self._global_requests)
@property
def global_tokens(self) -> Budget:
return self._global_tokens
@property
def global_requests(self) -> Budget:
return self._global_requests
@property
def global_time(self) -> Budget:
return self._global_time
def get_summary(self) -> dict[str, Any]:
"""Return a summary of global and per-node budget usage."""
return {
"global": {
"tokens": self._global_tokens.to_dict(),
"requests": self._global_requests.to_dict(),
"time": self._global_time.to_dict(),
"elapsed_seconds": self.get_elapsed_seconds(),
},
"nodes": {node_id: budget.to_dict() for node_id, budget in self._node_budgets.items()},
}
def reset(self) -> None:
"""Reset all budgets and the start time."""
self._global_tokens.reset()
self._global_requests.reset()
self._global_time.reset()
self._node_budgets.clear()
self._start_time = None
|