Skip to content

[codex] Honor max-connections above HTTP client cap #3

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
ywang96 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[codex] Honor max-connections above HTTP client cap #3

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 18 additions & 4 deletions kimi_model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,10 +22,12 @@

from inspect_ai.log import transcript
from inspect_ai.model import GenerateConfig, modelapi
from inspect_ai.model._openai import OpenAIAsyncHttpxClient
from inspect_ai.model._providers.openai_compatible import OpenAICompatibleAPI

# Unlimited read timeout for thinking mode (model may think for a long time)
STREAM_TIMEOUT = httpx.Timeout(timeout=None, connect=60.0)
STREAM_LIMITS = httpx.Limits(max_connections=None, max_keepalive_connections=100)

RETRYABLE_READ_ERRORS = (
HttpcoreReadError,
Expand All @@ -35,6 +37,21 @@
)


def _default_http_client() -> httpx.AsyncClient:
"""Use inspect_ai/OpenAI client defaults without a hidden pool cap.

inspect_ai already enforces `max_connections` with its own semaphore, so the
underlying HTTP client should not impose a lower default limit than the user
requested. We keep OpenAI's proxy/follow-redirect behavior while using an
unlimited active connection pool and a conservative keepalive pool.
"""

return OpenAIAsyncHttpxClient(
timeout=STREAM_TIMEOUT,
limits=STREAM_LIMITS,
)


def _get_file_logger() -> logging.Logger:
"""Get or create file logger for retry events."""
logger = logging.getLogger("kimi_retry")
Expand Down Expand Up @@ -117,10 +134,7 @@ def __init__(
**model_args,
) -> None:
if "http_client" not in model_args:
model_args["http_client"] = httpx.AsyncClient(
timeout=STREAM_TIMEOUT,
# http2=True,
)
model_args["http_client"] = _default_http_client()

super().__init__(
model_name=model_name,
Expand Down