stormlog.infer.vllm_hook.gate

Decide, from vLLM’s configured objects, whether the hook may record.

The gate runs where the configuration is known: in the scheduler’s constructor on the engine side and after Worker.init_device on the worker side. Both sides apply the same rules to the same vllm_config, so they agree. Anything outside the supported matrix in docs/vllm_execution.md is refused with a reason, and vLLM then runs untouched.

Functions

check(vllm_config, *[, scheduler, runner])	Apply the supported matrix; scheduler and runner are instances.
config_summary(vllm_config, *[, scheduler, ...])
vllm_version()

Classes

GateResult(enabled, refused[, config])

Whether to record, why not, and what was configured.

class stormlog.infer.vllm_hook.gate.GateResult(enabled, refused, config=<factory>)

Bases: object

Whether to record, why not, and what was configured.

Parameters:

• enabled (bool)

• refused (str | None)

• config (dict[str, Any])

enabled: bool

refused: str | None

config: dict[str, Any]

stormlog.infer.vllm_hook.gate.check(vllm_config, *, scheduler=None, runner=None)

Apply the supported matrix; scheduler and runner are instances.

Parameters:

• vllm_config (Any)

• scheduler (Any)

• runner (Any)

Return type:

GateResult

stormlog.infer.vllm_hook.gate.config_summary(vllm_config, *, scheduler=None, runner=None)

Parameters:

• vllm_config (Any)

• scheduler (Any)

• runner (Any)

Return type:

dict[str, Any]

stormlog.infer.vllm_hook.gate.vllm_version()

Return type:

str | None