nemo_rl.experience.interfaces#
Module Contents#
Classes#
A single generated completion for one prompt. |
|
All completions for a single prompt, with prompt-level metadata. |
Data#
API#
- nemo_rl.experience.interfaces.NEMO_GYM_TASK_INDEX_KEY#
‘_ng_task_index’
- nemo_rl.experience.interfaces.NEMO_GYM_GROUP_ID_KEY#
‘_ng_group_id’
- nemo_rl.experience.interfaces.NEMO_GYM_GROUP_ATTEMPT_KEY#
‘_ng_group_attempt’
- nemo_rl.experience.interfaces.NEMO_GYM_ROLLOUT_INDEX_KEY#
‘_ng_rollout_index’
- nemo_rl.experience.interfaces.NEXT_NEMO_GYM_TASK_INDEX_KEY#
‘next_ng_task_index’
- nemo_rl.experience.interfaces.PENDING_PROMPTS_KEY#
‘pending_prompt_batch’
- nemo_rl.experience.interfaces.FRONTIER_ORDINAL_KEY#
‘frontier_ordinal’
- nemo_rl.experience.interfaces.RESUME_BASE_ORDINAL_KEY#
‘resume_base_ordinal’
- nemo_rl.experience.interfaces.RETAINED_TASK_INDICES_KEY#
‘retained_task_indices’
- nemo_rl.experience.interfaces.TRAINED_TASK_INDICES_KEY#
‘trained_task_indices’
- class nemo_rl.experience.interfaces.Completion#
A single generated completion for one prompt.
- message_log: nemo_rl.data.interfaces.LLMMessageLogType | nemo_rl.data.interfaces.VLMMessageLogType#
None
- env_extras: Optional[dict[str, Any]]#
None
- truncated: bool#
None
- reward: float#
None
- class nemo_rl.experience.interfaces.PromptGroupRecord#
All completions for a single prompt, with prompt-level metadata.
- prompt_idx: int#
None
- extra_env_info: Optional[dict[str, Any]]#
None
- metadata: dict[str, Any]#
None
- completions: list[nemo_rl.experience.interfaces.Completion]#
None
- rollout_metrics: dict[str, Any]#
None
- loss_multiplier: float#
1.0