Skip to content

[Bugfix][Rollout] Restore generation with vLLM 0.30 #1

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

Draft
0z5a wants to merge 2 commits into
base: stack/vime-dev-h100-20260928
Choose a base branch
from
Draft

[Bugfix][Rollout] Restore generation with vLLM 0.30 #1

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e8f2a3e
fix(rollout): use vLLM 0.30 CLI paths and enable generation endpoints
0z5a Sep 28, 2026
555a862
Keep implementation and remove validation artifacts
0z5a Oct 1, 2026
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
14 changes: 7 additions & 7 deletions tests/_unit_stubs.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -273,9 +273,9 @@ def add_cli_args(cls, parser): # noqa: ARG003
# vllm.entrypoints stubs (used by arguments.add_vllm_arguments and vllm_engine._vllm_server_field_names)
entrypoints_mod = types.ModuleType("vllm.entrypoints")
entrypoints_mod.__path__ = []
openai_mod = types.ModuleType("vllm.entrypoints.openai")
openai_mod.__path__ = []
cli_args_mod = types.ModuleType("vllm.entrypoints.openai.cli_args")
launchers_mod = types.ModuleType("vllm.entrypoints.launchers")
launchers_mod.__path__ = []
cli_args_mod = types.ModuleType("vllm.entrypoints.launchers.cli_args")

import dataclasses as _dc

Expand All @@ -288,8 +288,8 @@ def add_cli_args(cls, parser): # noqa: ARG003
cli_args_mod.FrontendArgs = FrontendArgs
cli_args_mod.make_arg_parser = lambda parser=None: parser
cli_args_mod.validate_parsed_serve_args = lambda args: args
openai_mod.cli_args = cli_args_mod
entrypoints_mod.openai = openai_mod
launchers_mod.cli_args = cli_args_mod
entrypoints_mod.launchers = launchers_mod
vllm_mod.entrypoints = entrypoints_mod

cli_mod = types.ModuleType("vllm.entrypoints.cli")
Expand All @@ -314,8 +314,8 @@ class ServeSubcommand:
sys.modules["vllm.engine"] = engine_mod
sys.modules["vllm.engine.arg_utils"] = arg_utils
sys.modules["vllm.entrypoints"] = entrypoints_mod
sys.modules["vllm.entrypoints.openai"] = openai_mod
sys.modules["vllm.entrypoints.openai.cli_args"] = cli_args_mod
sys.modules["vllm.entrypoints.launchers"] = launchers_mod
sys.modules["vllm.entrypoints.launchers.cli_args"] = cli_args_mod
sys.modules["vllm.entrypoints.cli"] = cli_mod
sys.modules["vllm.entrypoints.cli.serve"] = serve_mod

Expand Down
2 changes: 1 addition & 1 deletion vime/backends/vllm_utils/arguments.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -99,7 +99,7 @@ def patched_add_argument_group(*g_args, **g_kwargs):
parser.add_argument = _wrap_add_argument(old_add_argument)
parser.add_argument_group = patched_add_argument_group
AsyncEngineArgs.add_cli_args(parser)
from vllm.entrypoints.openai.cli_args import FrontendArgs
from vllm.entrypoints.launchers.cli_args import FrontendArgs

FrontendArgs.add_cli_args(parser)
parser.add_argument = old_add_argument
Expand Down
5 changes: 3 additions & 2 deletions vime/backends/vllm_utils/vllm_engine.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -95,7 +95,7 @@ def _run_vllm_server(kwargs: dict, env: dict) -> None:
os.environ.update(env)

from vllm.entrypoints.cli.serve import ServeSubcommand
from vllm.entrypoints.openai.cli_args import make_arg_parser, validate_parsed_serve_args
from vllm.entrypoints.launchers.cli_args import make_arg_parser, validate_parsed_serve_args
from vllm.utils.argparse_utils import FlexibleArgumentParser

ns = argparse.Namespace(**kwargs)
Expand Down Expand Up @@ -682,6 +682,7 @@ def _compute_server_args(
"logprobs_mode": "processed_logprobs",
"enable_prompt_tokens_details": True,
"enable_server_load_tracking": True,
"enable_scale_out": True,
}

if pp > 1:
Expand Down Expand Up @@ -794,7 +795,7 @@ def _compute_server_args(
def _vllm_server_field_names() -> frozenset[str]:
"""Return the vLLM fields accepted by CLI generation and config overrides."""
from vllm.engine.arg_utils import AsyncEngineArgs
from vllm.entrypoints.openai.cli_args import FrontendArgs
from vllm.entrypoints.launchers.cli_args import FrontendArgs

return frozenset(f.name for f in (*dataclasses.fields(AsyncEngineArgs), *dataclasses.fields(FrontendArgs)))

Expand Down