Spaces:
Running
Running
File size: 9,027 Bytes
0e9afe0 | 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 | import hashlib
import math
from typing import Any
from helpers import files, history, skills, tokens
from helpers.llm_result import result_from_metadata
PARTS_KEY = "context_window_usage"
CACHE_KEY = "_context_window_usage_cache"
PROVIDER_USAGE_KEY = "context_window_provider_usage"
USAGE_KEYS = (
"messages",
"system_tools",
"skills",
"mcp_tools",
"system_prompt",
"extras",
)
MEASURED_KEYS = tuple(key for key in USAGE_KEYS if key != "system_prompt")
def reset(agent: Any) -> None:
params = _temporary_params(agent)
if params is not None:
params[PARTS_KEY] = {}
def discard(agent: Any) -> None:
params = _temporary_params(agent)
if params is not None:
params.pop(PARTS_KEY, None)
def record_prompt(agent: Any, key: str, prompt: Any) -> None:
parts = _parts(agent)
if parts is None or key not in MEASURED_KEYS:
return
text = files.remove_code_fences(str(prompt or ""), language="json")
parts[key] = _cached_tokens(agent, f"prompt:{key}", text)
def capture_context(agent: Any, loop_data: Any) -> None:
parts = _parts(agent)
if parts is None or loop_data is None:
return
output = list(getattr(loop_data, "history_output", None) or [])
parts["_history_output"] = output
skill_output = [message for message in output if skills.skill_instruction_name(message)]
skill_tokens = _output_tokens(agent, "history_skills", skill_output)
parts["messages"] = max(_history_tokens(agent, output) - skill_tokens, 0)
parts["skills"] = parts.get("skills", 0) + skill_tokens
protocol_values = {
**getattr(loop_data, "protocol_persistent", {}),
**getattr(loop_data, "protocol_temporary", {}),
}
extras_values = {
**getattr(loop_data, "extras_persistent", {}),
**getattr(loop_data, "extras_temporary", {}),
}
protocol = agent._build_context_message(
"agent.context.protocol.md",
"protocol",
protocol_values,
include_empty=False,
)
extras = agent._build_context_message(
"agent.context.extras.md",
"extras",
extras_values,
include_empty=True,
)
parts["extras"] = _output_tokens(agent, "extras", protocol + extras)
def finalize(agent: Any) -> None:
params = _temporary_params(agent)
parts = params.pop(PARTS_KEY, None) if params is not None else None
window = agent.get_data(agent.DATA_NAME_CTX_WINDOW) if agent else None
if not isinstance(parts, dict) or not isinstance(window, dict):
return
history_output = parts.pop("_history_output", None)
total = _non_negative_int(window.get("tokens"))
usage = {key: _non_negative_int(parts.get(key)) for key in MEASURED_KEYS}
measured_total = sum(usage.values())
if total and measured_total >= total and isinstance(history_output, list):
message_output = [
message
for message in history_output
if not skills.skill_instruction_name(message)
]
usage["messages"] = _output_tokens(
agent, "history_messages", message_output
)
measured_total = sum(usage.values())
if measured_total > total and measured_total:
usage = _scale_to_total(usage, total, measured_total)
measured_total = total
usage["system_prompt"] = total - measured_total
usage = {key: usage.get(key, 0) for key in USAGE_KEYS}
updated = dict(window)
updated["usage"] = usage
agent.set_data(agent.DATA_NAME_CTX_WINDOW, updated)
def usage_snapshot(value: Any) -> dict[str, int]:
if not isinstance(value, dict):
return {}
return {key: _non_negative_int(value.get(key)) for key in USAGE_KEYS}
def capture_provider_usage(agent: Any, result: Any) -> None:
if agent is None or result is None or not hasattr(result, "usage"):
return
snapshot = provider_usage_snapshot(getattr(result, "usage", None))
agent.set_data(
PROVIDER_USAGE_KEY,
snapshot if snapshot else {"available": False},
)
def latest_provider_usage(agent: Any) -> dict[str, int | float]:
data = getattr(agent, "data", None)
if isinstance(data, dict) and PROVIDER_USAGE_KEY in data:
stored = data.get(PROVIDER_USAGE_KEY)
if isinstance(stored, dict) and stored.get("available") is False:
return {}
return provider_usage_snapshot(stored)
all_messages = getattr(getattr(agent, "history", None), "all_messages", None)
if not callable(all_messages):
return {}
for message in reversed(all_messages()):
if not getattr(message, "ai", False):
continue
result = result_from_metadata(getattr(message, "metadata", None))
if result:
return provider_usage_snapshot(result.usage)
return {}
def provider_usage_snapshot(value: Any) -> dict[str, int | float]:
if not isinstance(value, dict):
return {}
input_details = {
**_mapping(value.get("prompt_tokens_details")),
**_mapping(value.get("input_tokens_details")),
}
result: dict[str, int | float] = {}
fields = {
"input_tokens": (value.get("input_tokens"), value.get("prompt_tokens")),
"cached_tokens": (
input_details.get("cached_tokens"),
input_details.get("cache_read_tokens"),
value.get("cache_read_input_tokens"),
value.get("cached_tokens"),
),
"output_tokens": (
value.get("output_tokens"),
value.get("completion_tokens"),
),
}
for key, values in fields.items():
number = _optional_non_negative_int(*values)
if number is not None:
result[key] = number
cost = _optional_non_negative_float(
value.get("cost"), value.get("response_cost")
)
if cost is not None:
result["cost"] = cost
return result
def _parts(agent: Any) -> dict[str, Any] | None:
params = _temporary_params(agent)
value = params.get(PARTS_KEY) if params is not None else None
return value if isinstance(value, dict) else None
def _temporary_params(agent: Any) -> dict[str, Any] | None:
loop_data = getattr(agent, "loop_data", None)
params = getattr(loop_data, "params_temporary", None)
return params if isinstance(params, dict) else None
def _history_tokens(agent: Any, output: list[history.OutputMessage]) -> int:
get_tokens = getattr(getattr(agent, "history", None), "get_tokens", None)
if callable(get_tokens):
return _non_negative_int(get_tokens())
return _count_output_tokens(output)
def _output_tokens(
agent: Any, cache_key: str, output: list[history.OutputMessage]
) -> int:
text = history.output_text(output, ai_label="assistant", human_label="user")
return _cached_tokens(agent, cache_key, text)
def _count_output_tokens(output: list[history.OutputMessage]) -> int:
text = history.output_text(output, ai_label="assistant", human_label="user")
return tokens.approximate_prompt_tokens(text)
def _cached_tokens(agent: Any, key: str, text: str) -> int:
cache = _cache(agent)
digest = hashlib.sha256(text.encode("utf-8")).hexdigest()
cached = cache.get(key) if cache is not None else None
if isinstance(cached, tuple) and len(cached) == 2 and cached[0] == digest:
return _non_negative_int(cached[1])
count = tokens.approximate_prompt_tokens(text)
if cache is not None:
cache[key] = (digest, count)
return count
def _cache(agent: Any) -> dict[str, tuple[str, int]] | None:
data = getattr(agent, "data", None)
if not isinstance(data, dict):
return None
cache = data.get(CACHE_KEY)
if not isinstance(cache, dict):
cache = {}
data[CACHE_KEY] = cache
return cache
def _non_negative_int(value: Any) -> int:
try:
return max(int(value or 0), 0)
except (TypeError, ValueError):
return 0
def _mapping(value: Any) -> dict[str, Any]:
return value if isinstance(value, dict) else {}
def _optional_non_negative_int(*values: Any) -> int | None:
for value in values:
if value is None:
continue
try:
return max(int(value), 0)
except (TypeError, ValueError):
continue
return None
def _optional_non_negative_float(*values: Any) -> float | None:
for value in values:
if value is None:
continue
try:
number = float(value)
except (TypeError, ValueError):
continue
if math.isfinite(number):
return max(number, 0)
return None
def _scale_to_total(values: dict[str, int], total: int, current: int) -> dict[str, int]:
scaled = {key: value * total // current for key, value in values.items()}
remainder = total - sum(scaled.values())
order = sorted(values, key=lambda key: values[key] * total % current, reverse=True)
for key in order[:remainder]:
scaled[key] += 1
return scaled
|