maxent_grpo.training.rollout.local

Local HF generation helpers split from the vLLM adapter.

Functions

_env_flag(name[, default])

Classes

LocalGenerationMixin(ctx)

Handle prompt expansion, tokenization, and local HF sampling.

class maxent_grpo.training.rollout.local.LocalGenerationMixin(ctx)[source]

Bases: object

Handle prompt expansion, tokenization, and local HF sampling.

Parameters:

ctx (GenerationContext)

ctx: GenerationContext
describe()[source]

Expose the underlying generation configuration for logging.

Return type:

dict[str, Any]