mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-11-06 05:17:15 +00:00
unsupportedpastels
a212421dbd
The per-route record held only the latest lookup, so a second session that detected the same window on a route cleared the label of an agent still running on its earlier default. The compressor now keeps the source of its own lookup (last_defaulted_context_length, read right after it resolves) and the agent label reads that. A nested lookup that defaults (a MoA preset's aggregator) now marks the outer resolution too, so the preset's window keeps "fallback". Addresses Greptile review on #148801 (f6e7d756).
72 lines
3.2 KiB
Python
72 lines
3.2 KiB
Python
"""Which context-window resolutions ended on a static default, recorded by the resolver itself.
|
|
|
|
``get_model_context_length`` has many sources (config, endpoint metadata, /models, provider catalogs,
|
|
local probes, Bedrock's table and probe). Only the resolver knows whether its answer came from one of
|
|
them or from a default (the generic 256K, Bedrock's 128K), so the "fallback" label reads this record
|
|
instead of re-deriving the answer. Keyed by profile home: a multiplexed gateway resolves for several
|
|
profiles in one process.
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
import contextvars
|
|
import functools
|
|
import inspect
|
|
from typing import Callable, Optional
|
|
|
|
# Set while a recorded resolution runs; the default branches flip it via mark_context_defaulted().
|
|
_DEFAULTED: contextvars.ContextVar[Optional[list]] = contextvars.ContextVar("context_defaulted", default=None)
|
|
_RECORDS: dict[tuple[str, str, str], int] = {} # (profile home, model, base_url) -> the defaulted window
|
|
# The calling context's last resolution: its defaulted window, or None when a source knew it. A compressor reads this
|
|
# right after its own lookup and keeps it, so a later lookup on the same route cannot relabel its window.
|
|
_LAST_DEFAULTED: contextvars.ContextVar[Optional[int]] = contextvars.ContextVar("last_context_defaulted", default=None)
|
|
_MAX_RECORDS = 512
|
|
|
|
|
|
def _key(model, base_url) -> tuple[str, str, str]:
|
|
from hermes_constants import hermes_home_key
|
|
return hermes_home_key(), str(model or "").strip(), str(base_url or "").strip().rstrip("/")
|
|
|
|
|
|
def mark_context_defaulted() -> None:
|
|
"""Called by a resolver branch that returns a static default instead of a known window."""
|
|
flag = _DEFAULTED.get()
|
|
if flag is not None:
|
|
flag[0] = True
|
|
|
|
|
|
def records_context_defaults(resolve: Callable[..., int]) -> Callable[..., int]:
|
|
"""Wrap the resolver: remember a defaulted answer for its (model, base_url), forget it otherwise."""
|
|
signature = inspect.signature(resolve)
|
|
|
|
@functools.wraps(resolve)
|
|
def wrapper(*args, **kwargs):
|
|
outer, flag = _DEFAULTED.get(), [False]
|
|
token = _DEFAULTED.set(flag)
|
|
try:
|
|
context_length = resolve(*args, **kwargs)
|
|
finally:
|
|
_DEFAULTED.reset(token)
|
|
if flag[0] and outer is not None:
|
|
outer[0] = True # a nested lookup (a MoA preset's aggregator) defaulted the answer the outer call returns
|
|
_LAST_DEFAULTED.set(context_length if flag[0] else None)
|
|
call = signature.bind_partial(*args, **kwargs).arguments
|
|
key = _key(call.get("model"), call.get("base_url"))
|
|
if flag[0]:
|
|
if len(_RECORDS) >= _MAX_RECORDS:
|
|
_RECORDS.clear()
|
|
_RECORDS[key] = context_length
|
|
else:
|
|
_RECORDS.pop(key, None)
|
|
return context_length
|
|
return wrapper
|
|
|
|
|
|
def last_defaulted_context_length() -> Optional[int]:
|
|
"""The window this context's most recent resolution defaulted to, or None when a source knew it."""
|
|
return _LAST_DEFAULTED.get()
|
|
|
|
|
|
def was_context_defaulted(context_length, model, base_url="") -> bool:
|
|
"""True when the last resolution of ``model`` on ``base_url`` in this profile defaulted to ``context_length``."""
|
|
return _RECORDS.get(_key(model, base_url)) == context_length
|