Index

_

__all__ (in module nemo_rl.data.datasets) (in module nemo_rl.data.datasets.eval_datasets) (in module nemo_rl.data.datasets.preference_datasets) (in module nemo_rl.data.datasets.response_datasets) (in module nemo_rl.data.packing) (in module nemo_rl.models.generation.vllm) __call__() (nemo_rl.algorithms.interfaces.LossFunction method) (nemo_rl.algorithms.loss_functions.ClippedPGLossFn method) (nemo_rl.algorithms.loss_functions.DistillationLossFn method) (nemo_rl.algorithms.loss_functions.DPOLossFn method) (nemo_rl.algorithms.loss_functions.NLLLoss method) (nemo_rl.algorithms.loss_functions.PreferenceLoss method) (nemo_rl.algorithms.loss_functions.SequencePackingLossWrapper method) (nemo_rl.data.interfaces.TaskDataProcessFnCallable method) (nemo_rl.distributed.worker_groups.RayWorkerBuilder method) __contact_emails__ (in module nemo_rl.package_info) __contact_names__ (in module nemo_rl.package_info) __del__() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) (nemo_rl.environments.reward_model_environment.RewardModelEnvironment method) (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.utils.logger.Logger method) (nemo_rl.utils.logger.MLflowLogger method) __description__ (in module nemo_rl.package_info) __download_url__ (in module nemo_rl.package_info) __eq__() (nemo_rl.distributed.named_sharding.NamedSharding method) __extra__ (nemo_rl.algorithms.loss_functions.ClippedPGLossDataDict attribute) (nemo_rl.data.interfaces.DatumSpec attribute) (nemo_rl.models.generation.interfaces.GenerationDatumSpec attribute) (nemo_rl.models.generation.interfaces.GenerationOutputSpec attribute) __getitem__() (nemo_rl.data.datasets.processed_dataset.AllTaskProcessedDataset method) (nemo_rl.data.datasets.response_datasets.oai_format_dataset.PreservingDataset method) __homepage__ (in module nemo_rl.package_info) __iter__() (nemo_rl.data.datasets.response_datasets.oai_format_dataset.PreservingDataset method) __keywords__ (in module nemo_rl.package_info) __len__() (nemo_rl.data.datasets.processed_dataset.AllTaskProcessedDataset method) (nemo_rl.data.datasets.response_datasets.oai_format_dataset.PreservingDataset method) (nemo_rl.data.multimodal_utils.PackedTensor method) __license__ (in module nemo_rl.package_info) __package_name__ (in module nemo_rl.package_info) __post_init__() (nemo_rl.data.interfaces.TaskDataSpec method) __repository_url__ (in module nemo_rl.package_info) __repr__() (nemo_rl.distributed.named_sharding.NamedSharding method) (nemo_rl.models.generation.vllm.vllm_worker.BaseVllmGenerationWorker method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) __shortversion__ (in module nemo_rl.package_info) __version__ (in module nemo_rl.package_info) _add_noise_to_weights() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) _adjust_bin_count() (nemo_rl.data.packing.algorithms.SequencePacker method) _apply_temperature_scaling() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) _async_generate_base() (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) _build_sampling_params() (nemo_rl.models.generation.vllm.vllm_worker.BaseVllmGenerationWorker method) _calculate_refit_param_info() (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) _calculate_target_weights() (nemo_rl.algorithms.async_utils.AsyncTrajectoryCollector method) _classify_items() (nemo_rl.data.packing.algorithms.ModifiedFirstFitDecreasingPacker method) _cleanup_finished_threads() (nemo_rl.algorithms.async_utils.AsyncTrajectoryCollector method) _collect() (nemo_rl.utils.logger.RayGpuMonitorLogger method) _collect_gpu_sku() (nemo_rl.utils.logger.RayGpuMonitorLogger method) _collect_metrics() (nemo_rl.utils.logger.RayGpuMonitorLogger method) _collection_loop() (nemo_rl.algorithms.async_utils.AsyncTrajectoryCollector method) (nemo_rl.utils.logger.RayGpuMonitorLogger method) _compute_distributed_log_softmax() (in module nemo_rl.distributed.model_utils) _construct_multichoice_prompt() (in module nemo_rl.data.processors) _create_engine() (nemo_rl.models.generation.vllm.vllm_worker.VllmGenerationWorker method) (nemo_rl.models.generation.vllm.vllm_worker_async.VllmAsyncGenerationWorker method) _create_indexed_lengths() (nemo_rl.data.packing.algorithms.SequencePacker method) _create_placement_groups_internal() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) _create_workers_from_bundle_indices() (nemo_rl.distributed.worker_groups.RayWorkerGroup method) _default_distillation_save_state() (in module nemo_rl.algorithms.distillation) _default_dpo_save_state() (in module nemo_rl.algorithms.dpo) _default_grpo_save_state() (in module nemo_rl.algorithms.grpo) _default_rm_save_state() (in module nemo_rl.algorithms.rm)

_default_sft_save_state() (in module nemo_rl.algorithms.sft) _dpo_loss() (nemo_rl.algorithms.loss_functions.DPOLossFn method) _env_builder() (in module nemo_rl.utils.venvs) _estimate_bins_needed() (nemo_rl.data.packing.algorithms.SequencePacker method) _fetch_and_parse_metrics() (nemo_rl.utils.logger.RayGpuMonitorLogger method) _generate_texts() (in module nemo_rl.evals.eval) _get_free_port_local() (in module nemo_rl.distributed.virtual_cluster) _get_module_from_param_name() (in module nemo_rl.models.generation.fp8) _get_next_target_for_generation() (nemo_rl.algorithms.async_utils.AsyncTrajectoryCollector method) _get_node_ip_and_free_port() (in module nemo_rl.distributed.virtual_cluster) _get_node_ip_local() (in module nemo_rl.distributed.virtual_cluster) _get_params_in_layers() (in module nemo_rl.models.generation.fp8) _get_sorted_bundle_indices() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) _get_tied_worker_bundle_indices() (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) _get_tokens_on_this_cp_rank() (in module nemo_rl.distributed.model_utils) _hybrid_model_flops() (in module nemo_rl.utils.flops_formulas) _infer_checkpoint_root() (in module nemo_rl.utils.automodel_checkpoint) _init_placement_groups() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) _is_fp8_weight() (in module nemo_rl.models.generation.fp8) _load_checkpoint_history() (in module nemo_rl.utils.checkpoint) _log_code() (nemo_rl.utils.logger.WandbLogger method) _log_diffs() (nemo_rl.utils.logger.WandbLogger method) _mamba_layer_flops() (in module nemo_rl.utils.flops_formulas) _merge_stop_strings() (nemo_rl.models.generation.vllm.vllm_worker.BaseVllmGenerationWorker method) _mlp_layer_flops() (in module nemo_rl.utils.flops_formulas) _mute_output() (in module nemo_rl.environments.math_environment) (in module nemo_rl.environments.vlm_environment) _non_mla_attn_layer_flops() (in module nemo_rl.utils.flops_formulas) _pack_implementation() (nemo_rl.data.packing.algorithms.ConcatenativePacker method) (nemo_rl.data.packing.algorithms.FirstFitPacker method) (nemo_rl.data.packing.algorithms.ModifiedFirstFitDecreasingPacker method) (nemo_rl.data.packing.algorithms.SequencePacker method) _pack_sequences_for_megatron() (in module nemo_rl.models.megatron.common) _pad_tensor() (in module nemo_rl.data.llm_message_utils) _parallelize_gemma3() (in module nemo_rl.models.dtensor.parallelize) _parallelize_llama() (in module nemo_rl.models.dtensor.parallelize) _parallelize_model() (in module nemo_rl.models.dtensor.parallelize) _parallelize_nm5_h() (in module nemo_rl.models.dtensor.parallelize) _parallelize_qwen() (in module nemo_rl.models.dtensor.parallelize) _parse_gpu_sku() (nemo_rl.utils.logger.RayGpuMonitorLogger method) _parse_metric() (nemo_rl.utils.logger.RayGpuMonitorLogger method) _patch_nsight_file() (in module nemo_rl) _per_token_group_quant_fp8() (in module nemo_rl.models.generation.fp8) _per_token_group_quant_fp8_colmajor() (in module nemo_rl.models.generation.fp8) _post_init() (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) _postprocess_penguin_to_nemo_rl_result() (nemo_rl.environments.penguin.Penguin method) _preference_loss() (nemo_rl.algorithms.loss_functions.PreferenceLoss method) _prepare_input_fn() (nemo_rl.models.dtensor.parallelize.RotaryEmbedParallel static method) _prepare_output_fn() (nemo_rl.models.dtensor.parallelize.RotaryEmbedParallel static method) _prepare_sequences() (nemo_rl.data.packing.algorithms.FirstFitDecreasingPacker method) (nemo_rl.data.packing.algorithms.FirstFitPacker method) (nemo_rl.data.packing.algorithms.FirstFitShufflePacker method) _print_results() (in module nemo_rl.evals.eval) _process_batch() (nemo_rl.algorithms.async_utils.AsyncTrajectoryCollector method) _REDUCTION_FUNCTIONS (nemo_rl.utils.timer.Timer attribute) _rekey() (nemo_rl.data.datasets.eval_datasets.aime.AIMEDataset method) (nemo_rl.data.datasets.eval_datasets.gpqa.GPQADataset method) (nemo_rl.data.datasets.eval_datasets.local_math_dataset.LocalMathDataset method) (nemo_rl.data.datasets.eval_datasets.math.MathDataset method) (nemo_rl.data.datasets.eval_datasets.mmlu.MMLUDataset method) (nemo_rl.data.datasets.eval_datasets.mmlu_pro.MMLUProDataset method) _replace_prefix_tokens() (in module nemo_rl.models.generation.vllm.vllm_worker_async) _report_device_id() (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) _report_dp_openai_server_base_urls() (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) _rich_logging_configured (in module nemo_rl.utils.logger) _run_env_eval_impl() (in module nemo_rl.evals.eval) _run_prompt_group_worker() (nemo_rl.algorithms.async_utils.AsyncTrajectoryCollector method) _save_evaluation_data_to_json() (in module nemo_rl.evals.eval) _setup_vllm_openai_api_server() (nemo_rl.models.generation.vllm.vllm_worker_async.VllmAsyncGenerationWorker method) _setup_vllm_server() (nemo_rl.models.generation.vllm.vllm_worker_async.VllmAsyncGenerationWorker method) _should_pause_for_generation_limits() (nemo_rl.algorithms.async_utils.AsyncTrajectoryCollector method) _should_use_async_rollouts() (in module nemo_rl.algorithms.grpo) _unpack_sequences_from_megatron() (in module nemo_rl.models.megatron.common) _v2 (nemo_rl.models.policy.DTensorConfig attribute) _validate_sequence_lengths() (nemo_rl.data.packing.algorithms.SequencePacker method) _validate_tensor_consistency() (in module nemo_rl.data.llm_message_utils)

A

accuracy (nemo_rl.algorithms.dpo.DPOValMetrics attribute) (nemo_rl.algorithms.rm.RMValMetrics attribute) ACK (nemo_rl.models.policy.utils.IPCProtocol attribute) activation_checkpointing (nemo_rl.models.policy.DTensorConfig attribute) (nemo_rl.models.policy.MegatronConfig attribute) ACTOR_ENVIRONMENT_REGISTRY (in module nemo_rl.distributed.ray_actor_environment_registry) adam_beta1 (nemo_rl.models.policy.MegatronOptimizerConfig attribute) adam_beta2 (nemo_rl.models.policy.MegatronOptimizerConfig attribute) adam_eps (nemo_rl.models.policy.MegatronOptimizerConfig attribute) add_bos (nemo_rl.data.DataConfig attribute) add_eos (nemo_rl.data.DataConfig attribute) add_generation_prompt (nemo_rl.data.DataConfig attribute) add_loss_mask_to_message_log() (in module nemo_rl.data.llm_message_utils) add_messages_key() (nemo_rl.data.datasets.response_datasets.oai_format_dataset.OpenAIFormatDataset method) (nemo_rl.data.datasets.response_datasets.response_dataset.ResponseDataset method) add_ref_logprobs_to_data() (in module nemo_rl.algorithms.dpo) add_system_prompt (nemo_rl.data.DataConfig attribute) ADDITIONAL_OPTIONAL_KEY_TENSORS (nemo_rl.distributed.batched_data_dict.BatchedDataDict attribute) advantages (nemo_rl.algorithms.loss_functions.ClippedPGLossDataDict attribute) AIMEDataset (class in nemo_rl.data.datasets.eval_datasets.aime) algorithm (nemo_rl.distributed.batched_data_dict.SequencePackingArgs attribute) (nemo_rl.models.policy.SequencePackingConfig attribute)

all_gather() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) allgather_cp_sharded_tensor() (in module nemo_rl.distributed.model_utils) AllGatherCPTensor (class in nemo_rl.distributed.model_utils) AllTaskProcessedDataset (class in nemo_rl.data.datasets.processed_dataset) answers (nemo_rl.environments.interfaces.EnvironmentReturn attribute) apply_fp8_patches() (in module nemo_rl.models.generation.fp8) apply_reward_shaping() (in module nemo_rl.algorithms.reward_functions) apply_rope_fusion (nemo_rl.models.policy.MegatronConfig attribute) as_tensor() (nemo_rl.data.multimodal_utils.PackedTensor method) assert_no_double_bos() (in module nemo_rl.data.datasets.utils) async_engine (nemo_rl.models.generation.vllm.config.VllmSpecificArgs attribute) async_generate_response_for_sample_turn() (in module nemo_rl.experience.rollouts) async_grpo (nemo_rl.algorithms.grpo.GRPOConfig attribute) async_grpo_train() (in module nemo_rl.algorithms.grpo) AsyncGRPOConfig (class in nemo_rl.algorithms.grpo) AsyncTrajectoryCollector (class in nemo_rl.algorithms.async_utils) attention_heads (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) AUTOMODEL (nemo_rl.distributed.virtual_cluster.PY_EXECUTABLES attribute) AUTOMODEL_FACTORY (in module nemo_rl.models.policy.utils) average_in_collective (nemo_rl.models.policy.MegatronDDPConfig attribute)

B

backend (nemo_rl.models.generation.interfaces.GenerationConfig attribute) backward() (nemo_rl.distributed.model_utils.AllGatherCPTensor method) (nemo_rl.distributed.model_utils.ChunkedDistributedEntropy static method) (nemo_rl.distributed.model_utils.ChunkedDistributedGatherLogprob static method) (nemo_rl.distributed.model_utils.ChunkedDistributedLogprob static method) (nemo_rl.distributed.model_utils.DistributedLogprob static method) BASE (nemo_rl.distributed.virtual_cluster.PY_EXECUTABLES attribute) base_urls (nemo_rl.environments.penguin.PenguinConfig attribute) BaseVllmGenerationWorker (class in nemo_rl.models.generation.vllm.vllm_worker) batch_multiplier (nemo_rl.algorithms.grpo.GRPOConfig attribute) batch_size (nemo_rl.environments.reward_model_environment.RewardModelEnvironmentConfig attribute) batched_message_log_to_flat_message() (in module nemo_rl.data.llm_message_utils) BatchedDataDict (class in nemo_rl.distributed.batched_data_dict)

bbox_giou_reward() (in module nemo_rl.environments.rewards) bert() (in module nemo_rl.utils.flops_formulas) bf16 (nemo_rl.models.policy.MegatronOptimizerConfig attribute) bias_activation_fusion (nemo_rl.models.policy.MegatronConfig attribute) BinaryPreferenceDataset (class in nemo_rl.data.datasets.preference_datasets.binary_preference_dataset) boxed (in module nemo_rl.environments.rewards) broadcast_object_across_pp_ranks() (in module nemo_rl.models.policy.megatron_policy_worker) broadcast_tensor() (in module nemo_rl.models.megatron.common) broadcast_weights_for_collective() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.interfaces.ColocatablePolicyInterface method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method)

C

calculate_aligned_size() (in module nemo_rl.models.policy.utils) calculate_baseline_and_std_per_prompt() (in module nemo_rl.algorithms.utils) calculate_kl_penalty_joschu2020() (in module nemo_rl.algorithms.utils) calculate_pass_rate_per_prompt() (in module nemo_rl.environments.metrics) calculate_rewards() (in module nemo_rl.experience.rollouts) calculate_stats_only() (nemo_rl.data.packing.metrics.PackingMetrics method) called_workers (nemo_rl.distributed.worker_groups.MultiWorkerFuture attribute) cast_tensor_to_fp8_blockwise() (in module nemo_rl.models.generation.fp8) causal_self_attn (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) chat_template (nemo_rl.models.policy.TokenizerConfig attribute) chat_template_kwargs (nemo_rl.models.policy.TokenizerConfig attribute) chdir() (nemo_rl.environments.code_environment.CodeExecutionWorker method) check_model_allow_flash_attn_args() (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) check_save() (nemo_rl.utils.timer.TimeoutChecker method) check_vocab_equality() (in module nemo_rl.algorithms.distillation) checkpoint_dir (nemo_rl.utils.checkpoint.CheckpointingConfig attribute) checkpoint_must_save_by (nemo_rl.utils.checkpoint.CheckpointingConfig attribute) checkpoint_path (nemo_rl.environments.reward_model_environment.RewardModelEnvironmentConfig attribute) checkpointing (nemo_rl.algorithms.distillation.MasterConfig attribute) (nemo_rl.algorithms.dpo.MasterConfig attribute) (nemo_rl.algorithms.grpo.MasterConfig attribute) (nemo_rl.algorithms.rm.MasterConfig attribute) (nemo_rl.algorithms.sft.MasterConfig attribute) CheckpointingConfig (class in nemo_rl.utils.checkpoint) CheckpointManager (class in nemo_rl.utils.checkpoint) chunk() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) chunk_list_to_workers() (in module nemo_rl.environments.utils) ChunkedDistributedEntropy (class in nemo_rl.distributed.model_utils) ChunkedDistributedGatherLogprob (class in nemo_rl.distributed.model_utils) ChunkedDistributedLogprob (class in nemo_rl.distributed.model_utils) class_token_len (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) cleanup() (nemo_rl.models.generation.vllm.vllm_backend.VllmInternalWorkerExtension method) clear() (nemo_rl.algorithms.async_utils.ReplayBuffer method) clear_cache_every_n_steps (nemo_rl.models.policy.DTensorConfig attribute) CLEVRCoGenTDataset (class in nemo_rl.data.datasets.response_datasets.clevr) clip_grad (nemo_rl.models.policy.MegatronOptimizerConfig attribute) clip_grad_by_total_norm_() (in module nemo_rl.models.dtensor.parallelize) ClippedPGLossConfig (class in nemo_rl.algorithms.loss_functions) ClippedPGLossDataDict (class in nemo_rl.algorithms.loss_functions) ClippedPGLossFn (class in nemo_rl.algorithms.loss_functions) cluster (nemo_rl.algorithms.distillation.MasterConfig attribute) (nemo_rl.algorithms.dpo.MasterConfig attribute) (nemo_rl.algorithms.grpo.MasterConfig attribute) (nemo_rl.algorithms.rm.MasterConfig attribute) (nemo_rl.algorithms.sft.MasterConfig attribute) (nemo_rl.evals.eval.MasterConfig attribute)

ClusterConfig (class in nemo_rl.distributed.virtual_cluster) CodeEnvConfig (class in nemo_rl.environments.code_environment) CodeEnvironment (class in nemo_rl.environments.code_environment) CodeEnvMetadata (class in nemo_rl.environments.code_environment) CodeExecutionWorker (class in nemo_rl.environments.code_environment) collection_interval (nemo_rl.utils.logger.GPUMonitoringConfig attribute) ColocatablePolicyInterface (class in nemo_rl.models.policy.interfaces) colocated (nemo_rl.models.generation.interfaces.GenerationConfig attribute) ColocationConfig (class in nemo_rl.models.generation.interfaces) combine_reward_functions() (in module nemo_rl.environments.rewards) COMMON_CHAT_TEMPLATES (class in nemo_rl.data.chat_templates) COMPLETE (nemo_rl.models.policy.utils.IPCProtocol attribute) compute_metrics() (nemo_rl.data.packing.algorithms.SequencePacker method) compute_score() (in module nemo_rl.environments.dapo_math_verifier) concat() (nemo_rl.data.multimodal_utils.PackedTensor class method) CONCATENATIVE (nemo_rl.data.packing.algorithms.PackingAlgorithm attribute) ConcatenativePacker (class in nemo_rl.data.packing.algorithms) configure_dynamo_cache() (in module nemo_rl.models.policy.utils) configure_generation_config() (in module nemo_rl.models.generation) configure_rich_logging() (in module nemo_rl.utils.logger) configure_worker() (nemo_rl.models.generation.vllm.vllm_worker.BaseVllmGenerationWorker static method) consumed_samples (nemo_rl.algorithms.distillation.DistillationSaveState attribute) (nemo_rl.algorithms.dpo.DPOSaveState attribute) (nemo_rl.algorithms.grpo.GRPOSaveState attribute) (nemo_rl.algorithms.rm.RMSaveState attribute) (nemo_rl.algorithms.sft.SFTSaveState attribute) context (nemo_rl.environments.code_environment.CodeEnvMetadata attribute) context_parallel_size (nemo_rl.models.policy.DTensorConfig attribute) (nemo_rl.models.policy.MegatronConfig attribute) convert_config_to_flops_config() (in module nemo_rl.utils.flops_tracker) convert_dcp_to_hf() (in module nemo_rl.utils.native_checkpoint) convert_to_seconds() (in module nemo_rl.utils.timer) converter_type (nemo_rl.models.policy.MegatronConfig attribute) copy_defaults() (nemo_rl.data.interfaces.TaskDataSpec method) cpu_offload (nemo_rl.models.policy.DTensorConfig attribute) create_context_parallel_ctx() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker static method) create_local_venv() (in module nemo_rl.utils.venvs) create_local_venv_on_each_node() (in module nemo_rl.utils.venvs) create_worker() (nemo_rl.distributed.worker_groups.RayWorkerBuilder.IsolatedWorkerInitializer method) create_worker_async() (nemo_rl.distributed.worker_groups.RayWorkerBuilder method) cu_seqlens_k (nemo_rl.models.huggingface.common.FlashAttentionKwargs attribute) cu_seqlens_q (nemo_rl.models.huggingface.common.FlashAttentionKwargs attribute) current_epoch (nemo_rl.algorithms.distillation.DistillationSaveState attribute) (nemo_rl.algorithms.grpo.GRPOSaveState attribute) current_step (nemo_rl.algorithms.distillation.DistillationSaveState attribute) (nemo_rl.algorithms.grpo.GRPOSaveState attribute) custom_parallel_plan (nemo_rl.models.policy.DTensorConfig attribute)

D

DAPOMath17KDataset (class in nemo_rl.data.datasets.response_datasets.dapo_math) data (nemo_rl.algorithms.distillation.MasterConfig attribute) (nemo_rl.algorithms.dpo.MasterConfig attribute) (nemo_rl.algorithms.grpo.MasterConfig attribute) (nemo_rl.algorithms.rm.MasterConfig attribute) (nemo_rl.algorithms.sft.MasterConfig attribute) (nemo_rl.evals.eval.MasterConfig attribute) data_parallel_sharding_strategy (nemo_rl.models.policy.MegatronDDPConfig attribute) DataConfig (class in nemo_rl.data) dataset_name (nemo_rl.data.DataConfig attribute) DatumSpec (class in nemo_rl.data.interfaces) DeepScalerDataset (class in nemo_rl.data.datasets.response_datasets.deepscaler) deepseekv3() (in module nemo_rl.utils.flops_formulas) DEFAULT_PY_EXECUTABLE (nemo_rl.environments.reward_model_environment.RewardModelEnvironment attribute) DEFAULT_VENV_DIR (in module nemo_rl.utils.venvs) defer_fp32_logits (nemo_rl.models.policy.MegatronConfig attribute) define_metric() (nemo_rl.utils.logger.SwanlabLogger method) (nemo_rl.utils.logger.WandbLogger method) destroy_parallel_state() (in module nemo_rl.models.policy.megatron_policy_worker) detect_checkpoint_format() (in module nemo_rl.utils.automodel_checkpoint) device_id_to_physical_device_id() (in module nemo_rl.utils.nvml) DictT (in module nemo_rl.distributed.batched_data_dict) dir_path (in module nemo_rl.distributed.virtual_cluster) (in module nemo_rl.utils.venvs) disable_forward_pre_hook() (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) distillation (nemo_rl.algorithms.distillation.MasterConfig attribute) distillation_train() (in module nemo_rl.algorithms.distillation) DistillationConfig (class in nemo_rl.algorithms.distillation) DistillationLossConfig (class in nemo_rl.algorithms.loss_functions) DistillationLossDataDict (class in nemo_rl.algorithms.loss_functions)

DistillationLossFn (class in nemo_rl.algorithms.loss_functions) DistillationSaveState (class in nemo_rl.algorithms.distillation) distributed_data_parallel_config (nemo_rl.models.policy.MegatronConfig attribute) distributed_vocab_topk() (in module nemo_rl.distributed.model_utils) DistributedLogprob (class in nemo_rl.distributed.model_utils) download_and_process_oasst() (in module nemo_rl.data.datasets.response_datasets.oasst) download_and_unzip() (in module nemo_rl.data.datasets.response_datasets.refcoco) download_dir (nemo_rl.data.DataConfig attribute) dp_size (nemo_rl.distributed.worker_groups.RayWorkerGroup property) dpo (nemo_rl.algorithms.dpo.MasterConfig attribute) dpo_train() (in module nemo_rl.algorithms.dpo) DPOConfig (class in nemo_rl.algorithms.dpo) DPODatumSpec (class in nemo_rl.data.interfaces) DPOLossConfig (class in nemo_rl.algorithms.loss_functions) DPOLossDataDict (class in nemo_rl.algorithms.loss_functions) DPOLossFn (class in nemo_rl.algorithms.loss_functions) DPOSaveState (class in nemo_rl.algorithms.dpo) DPOValMetrics (class in nemo_rl.algorithms.dpo) dtensor_cfg (nemo_rl.environments.reward_model_environment.RewardModelEnvironmentConfig attribute) (nemo_rl.models.policy.PolicyConfig attribute) dtensor_from_parallel_logits_to_logprobs() (in module nemo_rl.distributed.model_utils) DTensorConfig (class in nemo_rl.models.policy) DTensorPolicyWorker (class in nemo_rl.models.policy.dtensor_policy_worker) DTensorPolicyWorkerV2 (class in nemo_rl.models.policy.dtensor_policy_worker_v2) dynamic_batching (nemo_rl.environments.reward_model_environment.RewardModelEnvironmentConfig attribute) (nemo_rl.models.policy.PolicyConfig attribute) dynamic_sampling() (in module nemo_rl.algorithms.grpo) dynamic_sampling_max_gen_batches (nemo_rl.algorithms.grpo.GRPOConfig attribute) DynamicBatchingArgs (class in nemo_rl.distributed.batched_data_dict) DynamicBatchingConfig (class in nemo_rl.models.policy)

E

empty_like() (nemo_rl.data.multimodal_utils.PackedTensor class method) empty_unused_memory_level (nemo_rl.models.policy.MegatronConfig attribute) enable_forward_pre_hook() (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) enabled (nemo_rl.algorithms.grpo.AsyncGRPOConfig attribute) (nemo_rl.algorithms.grpo.RewardScalingConfig attribute) (nemo_rl.algorithms.reward_functions.RewardShapingConfig attribute) (nemo_rl.environments.reward_model_environment.RewardModelEnvironmentConfig attribute) (nemo_rl.models.generation.interfaces.ColocationConfig attribute) (nemo_rl.models.policy.DTensorConfig attribute) (nemo_rl.models.policy.DynamicBatchingConfig attribute) (nemo_rl.models.policy.MegatronConfig attribute) (nemo_rl.models.policy.RewardModelConfig attribute) (nemo_rl.models.policy.SequencePackingConfig attribute) (nemo_rl.utils.checkpoint.CheckpointingConfig attribute) enc_seq_len (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) encode_single() (nemo_rl.data.datasets.processed_dataset.AllTaskProcessedDataset method) end_weight_decay (nemo_rl.models.policy.MegatronSchedulerConfig attribute) enforce_eager (nemo_rl.models.generation.vllm.config.VllmSpecificArgs attribute) EnglishMultichoiceVerifyWorker (class in nemo_rl.environments.math_environment) env (nemo_rl.algorithms.distillation.MasterConfig attribute) (nemo_rl.algorithms.grpo.MasterConfig attribute) (nemo_rl.evals.eval.MasterConfig attribute)

env_vars (nemo_rl.models.policy.DTensorConfig attribute) (nemo_rl.models.policy.MegatronConfig attribute) EnvironmentInterface (class in nemo_rl.environments.interfaces) EnvironmentReturn (class in nemo_rl.environments.interfaces) epoch (nemo_rl.algorithms.dpo.DPOSaveState attribute) (nemo_rl.algorithms.rm.RMSaveState attribute) (nemo_rl.algorithms.sft.SFTSaveState attribute) eval (nemo_rl.evals.eval.MasterConfig attribute) eval_collate_fn() (in module nemo_rl.data.collate_fn) eval_cons_k() (in module nemo_rl.evals.eval) eval_pass_k() (in module nemo_rl.evals.eval) EvalConfig (class in nemo_rl.evals.eval) exact_answer_alphanumeric_reward() (in module nemo_rl.environments.rewards) execute() (nemo_rl.environments.code_environment.CodeExecutionWorker method) experiment_name (nemo_rl.utils.logger.MLflowConfig attribute) expert_model_parallel_size (nemo_rl.models.policy.MegatronConfig attribute) expert_parallel_size (nemo_rl.models.generation.vllm.config.VllmSpecificArgs attribute) expert_tensor_parallel_size (nemo_rl.models.policy.MegatronConfig attribute) export_model_from_megatron() (in module nemo_rl.models.megatron.community_import) expose_http_server (nemo_rl.models.generation.vllm.config.VllmSpecificArgs attribute) extra_env_info (nemo_rl.data.interfaces.DatumSpec attribute) extracted_answer (nemo_rl.environments.math_environment.MathEnvironmentMetadata attribute)

F

ffn_hs (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) finalize_checkpoint() (nemo_rl.utils.checkpoint.CheckpointManager method) finish_generation() (nemo_rl.models.generation.interfaces.GenerationInterface method) (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) (nemo_rl.models.policy.lm_policy.Policy method) finish_training() (nemo_rl.models.policy.interfaces.PolicyInterface method) (nemo_rl.models.policy.lm_policy.Policy method) FIRST_FIT_DECREASING (nemo_rl.data.packing.algorithms.PackingAlgorithm attribute) FIRST_FIT_SHUFFLE (nemo_rl.data.packing.algorithms.PackingAlgorithm attribute) FirstFitDecreasingPacker (class in nemo_rl.data.packing.algorithms) FirstFitPacker (class in nemo_rl.data.packing.algorithms) FirstFitShufflePacker (class in nemo_rl.data.packing.algorithms) FlashAttentionKwargs (class in nemo_rl.models.huggingface.common) FlatMessagesType (in module nemo_rl.data.interfaces) flatten_dict() (in module nemo_rl.utils.logger) flattened_concat() (nemo_rl.data.multimodal_utils.PackedTensor class method) FLOPSConfig (class in nemo_rl.utils.flops_formulas) FLOPTracker (class in nemo_rl.utils.flops_tracker) flush() (nemo_rl.utils.logger.RayGpuMonitorLogger method) flush_interval (nemo_rl.utils.logger.GPUMonitoringConfig attribute) flux() (in module nemo_rl.utils.flops_formulas) format_answer_fromtags() (in module nemo_rl.data.datasets.response_datasets.clevr) format_clevr_cogent_dataset() (in module nemo_rl.data.datasets.response_datasets.clevr) format_dapo_math_17k() (in module nemo_rl.data.datasets.response_datasets.dapo_math) format_geometry3k_dataset() (in module nemo_rl.data.datasets.response_datasets.geometry3k)

format_math() (in module nemo_rl.data.datasets.response_datasets.deepscaler) (in module nemo_rl.data.datasets.response_datasets.openmathinstruct2) format_prompt_for_vllm_generation() (in module nemo_rl.models.generation.vllm.utils) format_refcoco_dataset() (in module nemo_rl.data.datasets.response_datasets.refcoco) format_result() (nemo_rl.environments.code_environment.CodeExecutionWorker method) format_reward() (in module nemo_rl.environments.rewards) format_squad() (in module nemo_rl.data.datasets.response_datasets.squad) forward() (nemo_rl.distributed.model_utils.AllGatherCPTensor method) (nemo_rl.distributed.model_utils.ChunkedDistributedEntropy static method) (nemo_rl.distributed.model_utils.ChunkedDistributedGatherLogprob static method) (nemo_rl.distributed.model_utils.ChunkedDistributedLogprob static method) (nemo_rl.distributed.model_utils.DistributedLogprob static method) forward_step_arbitrary_loss() (in module nemo_rl.models.megatron.common) fp16 (nemo_rl.models.policy.MegatronOptimizerConfig attribute) FP8_BLOCK_QUANT_KWARGS (in module nemo_rl.models.generation.fp8) fp8_param_names (nemo_rl.models.generation.fp8.FP8State attribute) fp8_patches_applied (in module nemo_rl.models.generation.fp8) fp8_state (in module nemo_rl.models.generation.fp8) FP8Config (class in nemo_rl.models.generation.fp8) FP8State (class in nemo_rl.models.generation.fp8) freeze_moe_router (nemo_rl.models.policy.MegatronConfig attribute) from_batches() (nemo_rl.distributed.batched_data_dict.BatchedDataDict class method) from_config() (nemo_rl.utils.flops_tracker.FLOPTracker class method) from_parallel_logits_to_logprobs() (in module nemo_rl.distributed.model_utils) from_parallel_logits_to_logprobs_packed_sequences() (in module nemo_rl.distributed.model_utils) futures (nemo_rl.distributed.worker_groups.MultiWorkerFuture attribute)

G

gather_jagged_object_lists() (in module nemo_rl.distributed.collectives) gather_logits_at_global_indices() (in module nemo_rl.distributed.model_utils) gbs (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) generate() (nemo_rl.models.generation.interfaces.GenerationInterface method) (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) (nemo_rl.models.generation.vllm.vllm_worker.VllmGenerationWorker method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) generate_async() (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) (nemo_rl.models.generation.vllm.vllm_worker_async.VllmAsyncGenerationWorker method) generate_responses() (in module nemo_rl.experience.rollouts) generate_responses_async() (in module nemo_rl.experience.rollouts) generate_text() (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) (nemo_rl.models.generation.vllm.vllm_worker.VllmGenerationWorker method) generate_text_async() (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) (nemo_rl.models.generation.vllm.vllm_worker_async.VllmAsyncGenerationWorker method) generation (nemo_rl.environments.reward_model_environment.RewardModelEnvironmentConfig attribute) (nemo_rl.evals.eval.MasterConfig attribute) (nemo_rl.models.policy.PolicyConfig attribute) generation_batch_size (nemo_rl.models.policy.PolicyConfig attribute) generation_lengths (nemo_rl.models.generation.interfaces.GenerationOutputSpec attribute) generation_logprobs (nemo_rl.algorithms.loss_functions.ClippedPGLossDataDict attribute) GenerationConfig (class in nemo_rl.models.generation.interfaces) GenerationDatumSpec (class in nemo_rl.models.generation.interfaces) GenerationInterface (class in nemo_rl.models.generation.interfaces) GenerationOutputSpec (class in nemo_rl.models.generation.interfaces) Geometry3KDataset (class in nemo_rl.data.datasets.response_datasets.geometry3k) get_actor_python_env() (in module nemo_rl.distributed.ray_actor_environment_registry) get_aggregated_metrics() (nemo_rl.data.packing.algorithms.SequencePacker method) get_aggregated_stats() (nemo_rl.data.packing.metrics.PackingMetrics method) get_all_worker_results() (nemo_rl.distributed.worker_groups.RayWorkerGroup method) get_available_address_and_port() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) get_axis_index() (nemo_rl.distributed.named_sharding.NamedSharding method) get_axis_size() (nemo_rl.distributed.named_sharding.NamedSharding method) get_batch() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) get_best_checkpoint_path() (nemo_rl.utils.checkpoint.CheckpointManager method) get_cpu_state_dict() (in module nemo_rl.models.policy.dtensor_policy_worker) get_data_records() (in module nemo_rl.data.datasets.response_datasets.oasst) get_dataloader_state() (nemo_rl.algorithms.async_utils.AsyncTrajectoryCollector method) get_debug_info() (nemo_rl.algorithms.async_utils.ReplayBuffer method) get_default_hf_config() (in module nemo_rl.utils.flops_tracker) get_device_uuid() (in module nemo_rl.utils.nvml) get_dict() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) get_dim_to_pack_along() (in module nemo_rl.data.multimodal_utils) get_dp_leader_worker_idx() (nemo_rl.distributed.worker_groups.RayWorkerGroup method) get_elapsed() (nemo_rl.utils.timer.Timer method) get_existing_target_weights() (nemo_rl.algorithms.async_utils.ReplayBuffer method) get_extra_kwargs() (in module nemo_rl.data.datasets.utils) get_first_index_that_differs() (in module nemo_rl.data.llm_message_utils) get_flash_attention_kwargs() (in module nemo_rl.models.huggingface.common) get_formatted_message_log() (in module nemo_rl.data.llm_message_utils) get_free_memory_bytes() (in module nemo_rl.utils.nvml) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) get_gpu_id() (nemo_rl.distributed.virtual_cluster.GetGPUIDActor method) get_gpu_info() (in module nemo_rl.models.policy.utils) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) get_grad_norm() (in module nemo_rl.models.dtensor.parallelize) get_handle_from_tensor() (in module nemo_rl.models.policy.utils) get_hf_tp_plan() (in module nemo_rl.models.dtensor.parallelize) get_images_from_message() (in module nemo_rl.data.llm_message_utils) get_keys_from_message_log() (in module nemo_rl.data.llm_message_utils) get_last_target_weight_already_generated() (nemo_rl.algorithms.async_utils.ReplayBuffer method) get_latest_checkpoint_path() (nemo_rl.utils.checkpoint.CheckpointManager method) get_latest_elapsed() (nemo_rl.utils.timer.Timer method)

get_logprobs() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.interfaces.PolicyInterface method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) get_logprobs_from_vocab_parallel_logits() (in module nemo_rl.distributed.model_utils) get_master_address_and_port() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) get_megatron_checkpoint_dir() (in module nemo_rl.models.policy.utils) get_microbatch_iterator_dynamic_shapes_len() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) get_microbatch_iterator_for_packable_sequences_len() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) get_multimodal_dict() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) get_multimodal_keys_from_processor() (in module nemo_rl.data.multimodal_utils) get_next_experiment_dir() (in module nemo_rl.utils.logger) get_nsight_config_if_pattern_matches() (in module nemo_rl.distributed.worker_group_utils) get_num_buffers() (in module nemo_rl.utils.packed_tensor) get_packer() (in module nemo_rl.data.packing.algorithms) get_placement_groups() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) get_ranks() (nemo_rl.distributed.named_sharding.NamedSharding method) get_ranks_by_coord() (nemo_rl.distributed.named_sharding.NamedSharding method) get_reference_policy_logprobs() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.interfaces.PolicyInterface method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) get_results() (nemo_rl.distributed.worker_groups.MultiWorkerFuture method) get_runtime_env_for_policy_worker() (in module nemo_rl.models.policy.utils) get_target_packed_tensor_size() (in module nemo_rl.utils.packed_tensor) get_theoretical_tflops() (in module nemo_rl.utils.flops_tracker) get_timing_metrics() (nemo_rl.utils.timer.Timer method) get_tokenizer() (in module nemo_rl.algorithms.utils) get_topk_logits() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.interfaces.PolicyInterface method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) get_weight_version() (nemo_rl.algorithms.async_utils.AsyncTrajectoryCollector method) get_worker_coords() (nemo_rl.distributed.named_sharding.NamedSharding method) get_zmq_address() (nemo_rl.models.generation.vllm.vllm_backend.VllmInternalWorkerExtension method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) GetGPUIDActor (class in nemo_rl.distributed.virtual_cluster) git_root (in module nemo_rl.distributed.virtual_cluster) (in module nemo_rl.utils.venvs) global_fp8_config (in module nemo_rl.models.generation.fp8) global_post_process_and_metrics() (nemo_rl.environments.code_environment.CodeEnvironment method) (nemo_rl.environments.interfaces.EnvironmentInterface method) (nemo_rl.environments.math_environment.MathEnvironment method) (nemo_rl.environments.penguin.Penguin method) (nemo_rl.environments.reward_model_environment.RewardModelEnvironment method) (nemo_rl.environments.vlm_environment.VLMEnvironment method) global_valid_seqs (nemo_rl.algorithms.dpo.DPOValMetrics attribute) global_valid_toks (nemo_rl.algorithms.dpo.DPOValMetrics attribute) GPQADataset (class in nemo_rl.data.datasets.eval_datasets.gpqa) gpt3() (in module nemo_rl.utils.flops_formulas) gpu_memory_utilization (nemo_rl.models.generation.vllm.config.VllmSpecificArgs attribute) gpu_monitoring (nemo_rl.utils.logger.LoggerConfig attribute) GpuMetricSnapshot (class in nemo_rl.utils.logger) GPUMonitoringConfig (class in nemo_rl.utils.logger) gpus_per_node (nemo_rl.distributed.virtual_cluster.ClusterConfig attribute) (nemo_rl.models.generation.interfaces.ResourcesConfig attribute) grad_reduce_in_fp32 (nemo_rl.models.policy.MegatronDDPConfig attribute) ground_truth (nemo_rl.environments.math_environment.MathEnvironmentMetadata attribute) (nemo_rl.environments.vlm_environment.VLMEnvironmentMetadata attribute) group_and_cat_tensors() (in module nemo_rl.models.huggingface.common) grpo (nemo_rl.algorithms.grpo.MasterConfig attribute) grpo_train() (in module nemo_rl.algorithms.grpo) GRPOConfig (class in nemo_rl.algorithms.grpo) GRPOLoggerConfig (class in nemo_rl.algorithms.grpo) GRPOSaveState (class in nemo_rl.algorithms.grpo)

H

health_check() (nemo_rl.environments.penguin.Penguin method) HelpSteer3Dataset (class in nemo_rl.data.datasets.preference_datasets.helpsteer3) hf_config_overrides (nemo_rl.models.policy.PolicyConfig attribute) HFVerifyWorker (class in nemo_rl.environments.math_environment)

higher_is_better (nemo_rl.utils.checkpoint.CheckpointingConfig attribute) hs (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) http_server_serving_chat_kwargs (nemo_rl.models.generation.vllm.config.VllmSpecificArgs attribute) hybrid_override_pattern (nemo_rl.utils.flops_formulas.FLOPSConfig attribute)

I

idx (nemo_rl.data.interfaces.DatumSpec attribute) (nemo_rl.data.interfaces.DPODatumSpec attribute) img_h (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) img_seq_len (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) img_w (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) import_class_from_path() (in module nemo_rl.models.policy.utils) import_model_from_hf_name() (in module nemo_rl.models.megatron.community_import) in_channels (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) init_collective() (nemo_rl.models.generation.interfaces.GenerationInterface method) (nemo_rl.models.generation.vllm.vllm_backend.VllmInternalWorkerExtension method) (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) (nemo_rl.models.generation.vllm.vllm_worker.VllmGenerationWorker method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.interfaces.ColocatablePolicyInterface method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) init_collective_async() (nemo_rl.models.generation.vllm.vllm_worker_async.VllmAsyncGenerationWorker method) init_fp8() (in module nemo_rl.models.generation.fp8) init_ray() (in module nemo_rl.distributed.virtual_cluster) init_tmp_checkpoint() (nemo_rl.utils.checkpoint.CheckpointManager method) initial_global_config_dict (nemo_rl.environments.penguin.PenguinConfig attribute) inp_s (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) input_ids (nemo_rl.algorithms.loss_functions.ClippedPGLossDataDict attribute) (nemo_rl.algorithms.loss_functions.DistillationLossDataDict attribute) (nemo_rl.algorithms.loss_functions.DPOLossDataDict attribute) (nemo_rl.algorithms.loss_functions.PreferenceLossDataDict attribute) (nemo_rl.models.generation.interfaces.GenerationDatumSpec attribute)

input_key (nemo_rl.data.DataConfig attribute) (nemo_rl.distributed.batched_data_dict.DynamicBatchingArgs attribute) (nemo_rl.distributed.batched_data_dict.SequencePackingArgs attribute) input_lengths (nemo_rl.algorithms.loss_functions.DistillationLossDataDict attribute) (nemo_rl.models.generation.interfaces.GenerationDatumSpec attribute) input_lengths_key (nemo_rl.distributed.batched_data_dict.DynamicBatchingArgs attribute) (nemo_rl.distributed.batched_data_dict.SequencePackingArgs attribute) IPCProtocol (class in nemo_rl.models.policy.utils) is_alive() (nemo_rl.models.generation.vllm.vllm_worker.BaseVllmGenerationWorker method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) is_correct_minerva() (in module nemo_rl.environments.dapo_math_verifier) is_correct_strict_box() (in module nemo_rl.environments.dapo_math_verifier) is_fp8_model() (in module nemo_rl.models.generation.fp8) is_gemma_model() (in module nemo_rl.models.huggingface.common) is_hybrid_model (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) is_peft (nemo_rl.utils.checkpoint.CheckpointingConfig attribute) is_using_tf32() (in module nemo_rl.utils.flops_tracker) is_vllm_v1_engine_enabled() (in module nemo_rl.models.policy.utils)

K

k_value (nemo_rl.evals.eval.EvalConfig attribute) keep_top_k (nemo_rl.utils.checkpoint.CheckpointingConfig attribute) kl_type (nemo_rl.algorithms.loss_functions.DistillationLossConfig attribute)

kv_lora_rank (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) kwargs (nemo_rl.models.policy.PytorchOptimizerConfig attribute) (nemo_rl.models.policy.SinglePytorchSchedulerConfig attribute)

L

last_boxed_only_string() (in module nemo_rl.environments.dapo_math_verifier) layers (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) layout (nemo_rl.distributed.named_sharding.NamedSharding property) length (nemo_rl.data.interfaces.DatumSpec attribute) length_chosen (nemo_rl.data.interfaces.DPODatumSpec attribute) length_rejected (nemo_rl.data.interfaces.DPODatumSpec attribute) llama() (in module nemo_rl.utils.flops_formulas) llm() (nemo_rl.models.generation.vllm.vllm_worker.BaseVllmGenerationWorker method) LLMMessageLogType (in module nemo_rl.data.interfaces) load_checkpoint() (in module nemo_rl.utils.automodel_checkpoint) (in module nemo_rl.utils.native_checkpoint) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) load_config() (in module nemo_rl.utils.config) load_config_with_inheritance() (in module nemo_rl.utils.config) load_dataset_from_path() (in module nemo_rl.data.datasets.utils) load_eval_dataset() (in module nemo_rl.data.datasets.eval_datasets) load_format (nemo_rl.models.generation.vllm.config.VllmSpecificArgs attribute) load_preference_dataset() (in module nemo_rl.data.datasets.preference_datasets) load_response_dataset() (in module nemo_rl.data.datasets.response_datasets) load_state_dict() (nemo_rl.utils.native_checkpoint.ModelState method) (nemo_rl.utils.native_checkpoint.OptimizerState method) load_training_info() (nemo_rl.utils.checkpoint.CheckpointManager method) load_weights() (in module nemo_rl.models.generation.fp8) LocalMathDataset (class in nemo_rl.data.datasets.eval_datasets.local_math_dataset) log_batched_dict_as_jsonl() (nemo_rl.utils.logger.Logger method) log_dir (nemo_rl.utils.logger.LoggerConfig attribute) (nemo_rl.utils.logger.TensorboardConfig attribute) log_hyperparams() (nemo_rl.utils.logger.Logger method) (nemo_rl.utils.logger.LoggerInterface method) (nemo_rl.utils.logger.MLflowLogger method) (nemo_rl.utils.logger.SwanlabLogger method) (nemo_rl.utils.logger.TensorboardLogger method) (nemo_rl.utils.logger.WandbLogger method) log_metrics() (nemo_rl.utils.logger.Logger method) (nemo_rl.utils.logger.LoggerInterface method) (nemo_rl.utils.logger.MLflowLogger method) (nemo_rl.utils.logger.SwanlabLogger method) (nemo_rl.utils.logger.TensorboardLogger method) (nemo_rl.utils.logger.WandbLogger method)

log_plot() (nemo_rl.utils.logger.MLflowLogger method) (nemo_rl.utils.logger.SwanlabLogger method) (nemo_rl.utils.logger.TensorboardLogger method) (nemo_rl.utils.logger.WandbLogger method) log_plot_token_mult_prob_error() (nemo_rl.utils.logger.Logger method) Logger (class in nemo_rl.utils.logger) logger (in module nemo_rl.distributed.virtual_cluster) (in module nemo_rl.utils.venvs) (nemo_rl.algorithms.distillation.MasterConfig attribute) (nemo_rl.algorithms.dpo.MasterConfig attribute) (nemo_rl.algorithms.grpo.MasterConfig attribute) (nemo_rl.algorithms.rm.MasterConfig attribute) (nemo_rl.algorithms.sft.MasterConfig attribute) LoggerConfig (class in nemo_rl.utils.logger) LoggerInterface (class in nemo_rl.utils.logger) logprob_batch_size (nemo_rl.environments.reward_model_environment.RewardModelEnvironmentConfig attribute) (nemo_rl.models.policy.PolicyConfig attribute) logprob_chunk_size (nemo_rl.models.policy.PolicyConfig attribute) logprob_mb_tokens (nemo_rl.models.policy.DynamicBatchingConfig attribute) (nemo_rl.models.policy.SequencePackingConfig attribute) LogprobOutputSpec (class in nemo_rl.models.policy.interfaces) logprobs (nemo_rl.models.generation.interfaces.GenerationOutputSpec attribute) (nemo_rl.models.policy.interfaces.LogprobOutputSpec attribute) loss (nemo_rl.algorithms.dpo.DPOValMetrics attribute) (nemo_rl.algorithms.rm.RMValMetrics attribute) loss_fn (nemo_rl.algorithms.distillation.MasterConfig attribute) (nemo_rl.algorithms.grpo.MasterConfig attribute) loss_multiplier (nemo_rl.data.interfaces.DatumSpec attribute) (nemo_rl.data.interfaces.DPODatumSpec attribute) loss_type (nemo_rl.algorithms.interfaces.LossFunction attribute) (nemo_rl.algorithms.loss_functions.NLLLoss attribute) LossFunction (class in nemo_rl.algorithms.interfaces) LossType (class in nemo_rl.algorithms.interfaces) lr (nemo_rl.models.policy.MegatronOptimizerConfig attribute) lr_decay_iters (nemo_rl.models.policy.MegatronSchedulerConfig attribute) lr_decay_style (nemo_rl.models.policy.MegatronSchedulerConfig attribute) lr_warmup_init (nemo_rl.models.policy.MegatronSchedulerConfig attribute) lr_warmup_iters (nemo_rl.models.policy.MegatronSchedulerConfig attribute)

M

MAJOR (in module nemo_rl.package_info) make_microbatch_iterator() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) make_microbatch_iterator_for_packable_sequences() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) make_microbatch_iterator_with_dynamic_shapes() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) make_sequence_length_divisible_by (nemo_rl.models.policy.PolicyConfig attribute) mamba_head_dim (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) mamba_num_groups (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) mamba_num_heads (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) mamba_state_dim (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) map() (nemo_rl.data.datasets.response_datasets.oai_format_dataset.PreservingDataset method) mark_iteration() (nemo_rl.utils.timer.TimeoutChecker method) masked_mean() (in module nemo_rl.algorithms.utils) MasterConfig (class in nemo_rl.algorithms.distillation) (class in nemo_rl.algorithms.dpo) (class in nemo_rl.algorithms.grpo) (class in nemo_rl.algorithms.rm) (class in nemo_rl.algorithms.sft) (class in nemo_rl.evals.eval) matches() (nemo_rl.models.huggingface.common.ModelFlag method) math_data_processor() (in module nemo_rl.data.processors) math_expression_reward() (in module nemo_rl.environments.rewards) math_hf_data_processor() (in module nemo_rl.data.processors) math_verify_func (in module nemo_rl.environments.rewards) MathDataConfig (class in nemo_rl.data) MathDataset (class in nemo_rl.data.datasets.eval_datasets.math) MathEnvConfig (class in nemo_rl.environments.math_environment) MathEnvironment (class in nemo_rl.environments.math_environment) MathEnvironmentMetadata (class in nemo_rl.environments.math_environment) max_grad_norm (nemo_rl.environments.reward_model_environment.RewardModelEnvironmentConfig attribute) (nemo_rl.models.policy.PolicyConfig attribute) max_input_seq_length (nemo_rl.data.DataConfig attribute) max_model_len (nemo_rl.models.generation.vllm.config.VllmSpecificArgs attribute) max_new_tokens (nemo_rl.models.generation.interfaces.GenerationConfig attribute) max_num_epochs (nemo_rl.algorithms.distillation.DistillationConfig attribute) (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.grpo.GRPOConfig attribute) (nemo_rl.algorithms.rm.RMConfig attribute) (nemo_rl.algorithms.sft.SFTConfig attribute) max_num_steps (nemo_rl.algorithms.distillation.DistillationConfig attribute) (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.grpo.GRPOConfig attribute) (nemo_rl.algorithms.rm.RMConfig attribute) (nemo_rl.algorithms.sft.SFTConfig attribute) max_response_length (nemo_rl.algorithms.reward_functions.RewardShapingConfig attribute) max_rollout_turns (nemo_rl.algorithms.distillation.DistillationConfig attribute) (nemo_rl.algorithms.grpo.GRPOConfig attribute) max_seqlen_k (nemo_rl.models.huggingface.common.FlashAttentionKwargs attribute) max_seqlen_q (nemo_rl.models.huggingface.common.FlashAttentionKwargs attribute) max_sequences_per_bin (nemo_rl.data.packing.algorithms.ConcatenativePacker attribute) max_tokens_per_microbatch (nemo_rl.distributed.batched_data_dict.DynamicBatchingArgs attribute) (nemo_rl.distributed.batched_data_dict.SequencePackingArgs attribute) max_total_sequence_length (nemo_rl.models.policy.PolicyConfig attribute) max_trajectory_age_steps (nemo_rl.algorithms.grpo.AsyncGRPOConfig attribute) max_val_samples (nemo_rl.algorithms.distillation.DistillationConfig attribute) (nemo_rl.algorithms.grpo.GRPOConfig attribute) maybe_gpu_profile_step() (in module nemo_rl.utils.nsys) maybe_init_zmq() (nemo_rl.models.generation.vllm.vllm_backend.VllmInternalWorkerExtension method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) maybe_pad_last_batch() (in module nemo_rl.algorithms.utils) MCORE (nemo_rl.distributed.virtual_cluster.PY_EXECUTABLES attribute) MCORE_EXECUTABLE (in module nemo_rl.distributed.ray_actor_environment_registry) megatron_cfg (nemo_rl.models.policy.PolicyConfig attribute) megatron_path (in module nemo_rl) MegatronConfig (class in nemo_rl.models.policy) MegatronDDPConfig (class in nemo_rl.models.policy) MegatronOptimizerConfig (class in nemo_rl.models.policy) MegatronPolicyWorker (class in nemo_rl.models.policy.megatron_policy_worker) MegatronSchedulerConfig (class in nemo_rl.models.policy) message_log (nemo_rl.data.interfaces.DatumSpec attribute) message_log_chosen (nemo_rl.data.interfaces.DPODatumSpec attribute) message_log_rejected (nemo_rl.data.interfaces.DPODatumSpec attribute) message_log_shape() (in module nemo_rl.data.llm_message_utils) message_log_to_flat_messages() (in module nemo_rl.data.llm_message_utils) metadata (nemo_rl.environments.interfaces.EnvironmentReturn attribute) MetadataT (in module nemo_rl.environments.interfaces) metric (nemo_rl.evals.eval.EvalConfig attribute) metric_name (nemo_rl.utils.checkpoint.CheckpointingConfig attribute) metrics (nemo_rl.utils.logger.GpuMetricSnapshot attribute) milestones (nemo_rl.models.policy.SinglePytorchSchedulerConfig attribute) min_lr (nemo_rl.models.policy.MegatronOptimizerConfig attribute) MINOR (in module nemo_rl.package_info) mixed_kl_weight (nemo_rl.algorithms.loss_functions.DistillationLossConfig attribute) mixtral() (in module nemo_rl.utils.flops_formulas) mlflow (nemo_rl.utils.logger.LoggerConfig attribute) mlflow_enabled (nemo_rl.utils.logger.LoggerConfig attribute) MLflowConfig (class in nemo_rl.utils.logger) MLflowLogger (class in nemo_rl.utils.logger) MMLUDataset (class in nemo_rl.data.datasets.eval_datasets.mmlu) MMLUProDataset (class in nemo_rl.data.datasets.eval_datasets.mmlu_pro) model_cache_dir (nemo_rl.utils.checkpoint.CheckpointingConfig attribute) model_channels (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) model_name (nemo_rl.environments.penguin.PenguinConfig attribute) (nemo_rl.environments.reward_model_environment.RewardModelEnvironmentConfig attribute) (nemo_rl.models.generation.interfaces.GenerationConfig attribute) (nemo_rl.models.policy.PolicyConfig attribute) model_parallel_size (nemo_rl.models.generation.fp8.FP8Config attribute) model_pattern (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) model_repo_id (nemo_rl.utils.checkpoint.CheckpointingConfig attribute) model_save_format (nemo_rl.utils.checkpoint.CheckpointingConfig attribute) ModelFlag (class in nemo_rl.models.huggingface.common) ModelState (class in nemo_rl.utils.native_checkpoint) MODIFIED_FIRST_FIT_DECREASING (nemo_rl.data.packing.algorithms.PackingAlgorithm attribute) ModifiedFirstFitDecreasingPacker (class in nemo_rl.data.packing.algorithms) module nemo_rl nemo_rl.algorithms nemo_rl.algorithms.async_utils nemo_rl.algorithms.distillation nemo_rl.algorithms.dpo nemo_rl.algorithms.grpo nemo_rl.algorithms.interfaces nemo_rl.algorithms.loss_functions nemo_rl.algorithms.reward_functions nemo_rl.algorithms.rm nemo_rl.algorithms.sft nemo_rl.algorithms.utils nemo_rl.data nemo_rl.data.chat_templates nemo_rl.data.collate_fn nemo_rl.data.datasets nemo_rl.data.datasets.eval_datasets nemo_rl.data.datasets.eval_datasets.aime nemo_rl.data.datasets.eval_datasets.gpqa nemo_rl.data.datasets.eval_datasets.local_math_dataset nemo_rl.data.datasets.eval_datasets.math nemo_rl.data.datasets.eval_datasets.mmlu nemo_rl.data.datasets.eval_datasets.mmlu_pro nemo_rl.data.datasets.preference_datasets nemo_rl.data.datasets.preference_datasets.binary_preference_dataset nemo_rl.data.datasets.preference_datasets.helpsteer3 nemo_rl.data.datasets.preference_datasets.preference_dataset nemo_rl.data.datasets.preference_datasets.tulu3 nemo_rl.data.datasets.processed_dataset nemo_rl.data.datasets.response_datasets nemo_rl.data.datasets.response_datasets.clevr nemo_rl.data.datasets.response_datasets.dapo_math nemo_rl.data.datasets.response_datasets.deepscaler nemo_rl.data.datasets.response_datasets.geometry3k nemo_rl.data.datasets.response_datasets.oai_format_dataset nemo_rl.data.datasets.response_datasets.oasst nemo_rl.data.datasets.response_datasets.openmathinstruct2 nemo_rl.data.datasets.response_datasets.refcoco nemo_rl.data.datasets.response_datasets.response_dataset nemo_rl.data.datasets.response_datasets.squad nemo_rl.data.datasets.utils nemo_rl.data.interfaces nemo_rl.data.llm_message_utils nemo_rl.data.multimodal_utils nemo_rl.data.packing nemo_rl.data.packing.algorithms nemo_rl.data.packing.metrics nemo_rl.data.processors nemo_rl.distributed nemo_rl.distributed.batched_data_dict nemo_rl.distributed.collectives nemo_rl.distributed.model_utils nemo_rl.distributed.named_sharding nemo_rl.distributed.ray_actor_environment_registry nemo_rl.distributed.virtual_cluster nemo_rl.distributed.worker_group_utils nemo_rl.distributed.worker_groups nemo_rl.environments nemo_rl.environments.code_environment nemo_rl.environments.dapo_math_verifier nemo_rl.environments.interfaces nemo_rl.environments.math_environment nemo_rl.environments.metrics nemo_rl.environments.penguin nemo_rl.environments.reward_model_environment nemo_rl.environments.rewards nemo_rl.environments.utils nemo_rl.environments.vlm_environment nemo_rl.evals nemo_rl.evals.answer_parsing nemo_rl.evals.eval nemo_rl.experience nemo_rl.experience.rollouts nemo_rl.models nemo_rl.models.dtensor nemo_rl.models.dtensor.parallelize nemo_rl.models.generation nemo_rl.models.generation.fp8 nemo_rl.models.generation.interfaces nemo_rl.models.generation.vllm nemo_rl.models.generation.vllm.config nemo_rl.models.generation.vllm.utils nemo_rl.models.generation.vllm.vllm_backend nemo_rl.models.generation.vllm.vllm_generation nemo_rl.models.generation.vllm.vllm_worker nemo_rl.models.generation.vllm.vllm_worker_async nemo_rl.models.huggingface nemo_rl.models.huggingface.common nemo_rl.models.megatron nemo_rl.models.megatron.common nemo_rl.models.megatron.community_import nemo_rl.models.policy nemo_rl.models.policy.dtensor_policy_worker nemo_rl.models.policy.dtensor_policy_worker_v2 nemo_rl.models.policy.interfaces nemo_rl.models.policy.lm_policy nemo_rl.models.policy.megatron_policy_worker nemo_rl.models.policy.utils nemo_rl.package_info nemo_rl.utils nemo_rl.utils.automodel_checkpoint nemo_rl.utils.checkpoint nemo_rl.utils.config nemo_rl.utils.flops_formulas nemo_rl.utils.flops_tracker nemo_rl.utils.logger nemo_rl.utils.native_checkpoint nemo_rl.utils.nsys nemo_rl.utils.nvml nemo_rl.utils.packed_tensor nemo_rl.utils.prefetch_venvs nemo_rl.utils.timer nemo_rl.utils.venvs

moe_ffn_hidden_size (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) moe_layer_freq (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) moe_router_topk (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) moe_shared_expert_intermediate_size (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) monitor_gpus (nemo_rl.utils.logger.LoggerConfig attribute) monkey_patch_vllm_ray_executor() (in module nemo_rl.models.generation.fp8) move_buffer_to_device() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) move_model() (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) move_to_cpu() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) move_to_cuda() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) move_to_device() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) mtp_num_layers (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) multichoice_qa_processor() (in module nemo_rl.data.processors) MULTILINGUAL_ANSWER_PATTERN_TEMPLATE (in module nemo_rl.evals.answer_parsing) MULTILINGUAL_ANSWER_REGEXES (in module nemo_rl.evals.answer_parsing) MultilingualMultichoiceVerifyWorker (class in nemo_rl.environments.math_environment) MultiWorkerFuture (class in nemo_rl.distributed.worker_groups) my_init() (in module nemo_rl.models.generation.fp8) my_run_engine_core() (in module nemo_rl.models.generation.fp8)

N

name (nemo_rl.models.policy.PytorchOptimizerConfig attribute) (nemo_rl.models.policy.SinglePytorchSchedulerConfig attribute) (nemo_rl.models.policy.TokenizerConfig attribute) (nemo_rl.utils.logger.SwanlabConfig attribute) (nemo_rl.utils.logger.WandbConfig attribute) NamedSharding (class in nemo_rl.distributed.named_sharding) names (nemo_rl.distributed.named_sharding.NamedSharding property) ndim (nemo_rl.distributed.named_sharding.NamedSharding property) nemo_rl module nemo_rl.algorithms module nemo_rl.algorithms.async_utils module nemo_rl.algorithms.distillation module nemo_rl.algorithms.dpo module nemo_rl.algorithms.grpo module nemo_rl.algorithms.interfaces module nemo_rl.algorithms.loss_functions module nemo_rl.algorithms.reward_functions module nemo_rl.algorithms.rm module nemo_rl.algorithms.sft module nemo_rl.algorithms.utils module nemo_rl.data module nemo_rl.data.chat_templates module nemo_rl.data.collate_fn module nemo_rl.data.datasets module nemo_rl.data.datasets.eval_datasets module nemo_rl.data.datasets.eval_datasets.aime module nemo_rl.data.datasets.eval_datasets.gpqa module nemo_rl.data.datasets.eval_datasets.local_math_dataset module nemo_rl.data.datasets.eval_datasets.math module nemo_rl.data.datasets.eval_datasets.mmlu module nemo_rl.data.datasets.eval_datasets.mmlu_pro module nemo_rl.data.datasets.preference_datasets module nemo_rl.data.datasets.preference_datasets.binary_preference_dataset module nemo_rl.data.datasets.preference_datasets.helpsteer3 module nemo_rl.data.datasets.preference_datasets.preference_dataset module nemo_rl.data.datasets.preference_datasets.tulu3 module nemo_rl.data.datasets.processed_dataset module nemo_rl.data.datasets.response_datasets module nemo_rl.data.datasets.response_datasets.clevr module nemo_rl.data.datasets.response_datasets.dapo_math module nemo_rl.data.datasets.response_datasets.deepscaler module nemo_rl.data.datasets.response_datasets.geometry3k module nemo_rl.data.datasets.response_datasets.oai_format_dataset module nemo_rl.data.datasets.response_datasets.oasst module nemo_rl.data.datasets.response_datasets.openmathinstruct2 module nemo_rl.data.datasets.response_datasets.refcoco module nemo_rl.data.datasets.response_datasets.response_dataset module nemo_rl.data.datasets.response_datasets.squad module nemo_rl.data.datasets.utils module nemo_rl.data.interfaces module nemo_rl.data.llm_message_utils module nemo_rl.data.multimodal_utils module nemo_rl.data.packing module nemo_rl.data.packing.algorithms module nemo_rl.data.packing.metrics module nemo_rl.data.processors module nemo_rl.distributed module nemo_rl.distributed.batched_data_dict module nemo_rl.distributed.collectives module nemo_rl.distributed.model_utils module nemo_rl.distributed.named_sharding module nemo_rl.distributed.ray_actor_environment_registry module nemo_rl.distributed.virtual_cluster module nemo_rl.distributed.worker_group_utils module nemo_rl.distributed.worker_groups module nemo_rl.environments module nemo_rl.environments.code_environment module nemo_rl.environments.dapo_math_verifier module nemo_rl.environments.interfaces module nemo_rl.environments.math_environment module

nemo_rl.environments.metrics module nemo_rl.environments.penguin module nemo_rl.environments.reward_model_environment module nemo_rl.environments.rewards module nemo_rl.environments.utils module nemo_rl.environments.vlm_environment module nemo_rl.evals module nemo_rl.evals.answer_parsing module nemo_rl.evals.eval module nemo_rl.experience module nemo_rl.experience.rollouts module nemo_rl.models module nemo_rl.models.dtensor module nemo_rl.models.dtensor.parallelize module nemo_rl.models.generation module nemo_rl.models.generation.fp8 module nemo_rl.models.generation.interfaces module nemo_rl.models.generation.vllm module nemo_rl.models.generation.vllm.config module nemo_rl.models.generation.vllm.utils module nemo_rl.models.generation.vllm.vllm_backend module nemo_rl.models.generation.vllm.vllm_generation module nemo_rl.models.generation.vllm.vllm_worker module nemo_rl.models.generation.vllm.vllm_worker_async module nemo_rl.models.huggingface module nemo_rl.models.huggingface.common module nemo_rl.models.megatron module nemo_rl.models.megatron.common module nemo_rl.models.megatron.community_import module nemo_rl.models.policy module nemo_rl.models.policy.dtensor_policy_worker module nemo_rl.models.policy.dtensor_policy_worker_v2 module nemo_rl.models.policy.interfaces module nemo_rl.models.policy.lm_policy module nemo_rl.models.policy.megatron_policy_worker module nemo_rl.models.policy.utils module nemo_rl.package_info module nemo_rl.utils module nemo_rl.utils.automodel_checkpoint module nemo_rl.utils.checkpoint module nemo_rl.utils.config module nemo_rl.utils.flops_formulas module nemo_rl.utils.flops_tracker module nemo_rl.utils.logger module nemo_rl.utils.native_checkpoint module nemo_rl.utils.nsys module nemo_rl.utils.nvml module nemo_rl.utils.packed_tensor module nemo_rl.utils.prefetch_venvs module nemo_rl.utils.timer module nemo_rl.utils.venvs module nemotron() (in module nemo_rl.utils.flops_formulas) nemotronh() (in module nemo_rl.utils.flops_formulas) next_stop_strings (nemo_rl.environments.interfaces.EnvironmentReturn attribute) NLLLoss (class in nemo_rl.algorithms.loss_functions) node_count() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) normalize_extracted_answer() (in module nemo_rl.evals.answer_parsing) normalize_final_answer() (in module nemo_rl.environments.dapo_math_verifier) normalize_response() (in module nemo_rl.evals.answer_parsing) normalize_rewards (nemo_rl.algorithms.grpo.GRPOConfig attribute) NRL_NSYS_PROFILE_STEP_RANGE (in module nemo_rl.utils.nsys) NRL_NSYS_WORKER_PATTERNS (in module nemo_rl.utils.nsys) num_first_layers_in_bf16 (nemo_rl.models.generation.fp8.FP8Config attribute) num_generations_per_prompt (nemo_rl.algorithms.distillation.DistillationConfig attribute) (nemo_rl.algorithms.grpo.GRPOConfig attribute) num_last_layers_in_bf16 (nemo_rl.models.generation.fp8.FP8Config attribute) num_layers_in_first_pipeline_stage (nemo_rl.models.policy.MegatronConfig attribute) num_layers_in_last_pipeline_stage (nemo_rl.models.policy.MegatronConfig attribute) num_nodes (nemo_rl.distributed.virtual_cluster.ClusterConfig attribute) (nemo_rl.models.generation.interfaces.ResourcesConfig attribute) num_prompts_per_step (nemo_rl.algorithms.distillation.DistillationConfig attribute) (nemo_rl.algorithms.grpo.GRPOConfig attribute) num_tests_per_prompt (nemo_rl.evals.eval.EvalConfig attribute) num_val_samples_to_print (nemo_rl.algorithms.grpo.GRPOLoggerConfig attribute) (nemo_rl.utils.logger.LoggerConfig attribute) num_valid_samples (nemo_rl.algorithms.dpo.DPOValMetrics attribute) (nemo_rl.algorithms.rm.RMValMetrics attribute) num_workers (nemo_rl.data.DataConfig attribute) (nemo_rl.environments.code_environment.CodeEnvConfig attribute) (nemo_rl.environments.math_environment.MathEnvConfig attribute) (nemo_rl.environments.vlm_environment.VLMEnvConfig attribute) nvml_context() (in module nemo_rl.utils.nvml)

O

OasstDataset (class in nemo_rl.data.datasets.response_datasets.oasst) observations (nemo_rl.environments.interfaces.EnvironmentReturn attribute) offload_after_refit() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.interfaces.ColocatablePolicyInterface method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) offload_before_refit() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.interfaces.ColocatablePolicyInterface method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) OpenAIFormatDataset (class in nemo_rl.data.datasets.response_datasets.oai_format_dataset) OpenMathInstruct2Dataset (class in nemo_rl.data.datasets.response_datasets.openmathinstruct2) optimizer (nemo_rl.models.policy.MegatronConfig attribute) (nemo_rl.models.policy.MegatronOptimizerConfig attribute) (nemo_rl.models.policy.PolicyConfig attribute)

optimizer_cpu_offload (nemo_rl.models.policy.MegatronOptimizerConfig attribute) optimizer_offload_fraction (nemo_rl.models.policy.MegatronOptimizerConfig attribute) OptimizerState (class in nemo_rl.utils.native_checkpoint) original_init (in module nemo_rl.models.generation.fp8) original_run_engine_core (in module nemo_rl.models.generation.fp8) output_ids (nemo_rl.models.generation.interfaces.GenerationOutputSpec attribute) output_key (nemo_rl.data.DataConfig attribute) overlap_grad_reduce (nemo_rl.models.policy.MegatronDDPConfig attribute) overlap_param_gather (nemo_rl.models.policy.MegatronDDPConfig attribute) overlong_buffer_length (nemo_rl.algorithms.reward_functions.RewardShapingConfig attribute) overlong_buffer_penalty (nemo_rl.algorithms.reward_functions.RewardShapingConfig attribute) overlong_filtering (nemo_rl.algorithms.grpo.GRPOConfig attribute) OverridesError

P

pack() (nemo_rl.data.packing.algorithms.SequencePacker method) pack_sequences() (in module nemo_rl.models.huggingface.common) packed_broadcast_consumer() (in module nemo_rl.utils.packed_tensor) packed_broadcast_producer() (in module nemo_rl.utils.packed_tensor) PackedTensor (class in nemo_rl.data.multimodal_utils) PackingAlgorithm (class in nemo_rl.data.packing.algorithms) PackingMetrics (class in nemo_rl.data.packing.metrics) pad_token_id (nemo_rl.models.generation.interfaces.GenerationConfig attribute) PARALLIZE_FUNCTIONS (in module nemo_rl.models.dtensor.parallelize) params_dtype (nemo_rl.models.policy.MegatronOptimizerConfig attribute) parse_conversations() (in module nemo_rl.data.datasets.response_datasets.oasst) parse_hydra_overrides() (in module nemo_rl.utils.config) passthrough_prompt_response (nemo_rl.data.chat_templates.COMMON_CHAT_TEMPLATES attribute) PATCH (in module nemo_rl.package_info) patch_dim (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) patch_transformers_module_dir() (in module nemo_rl.utils.venvs) PathLike (in module nemo_rl.data.interfaces) (in module nemo_rl.models.policy.lm_policy) (in module nemo_rl.utils.checkpoint) pause() (nemo_rl.algorithms.async_utils.AsyncTrajectoryCollector method) peft_config (nemo_rl.utils.checkpoint.CheckpointingConfig attribute) Penguin (class in nemo_rl.environments.penguin) PENGUIN (nemo_rl.distributed.virtual_cluster.PY_EXECUTABLES attribute) penguin_example_to_nemo_rl_datum_spec() (in module nemo_rl.environments.penguin) PenguinConfig (class in nemo_rl.environments.penguin) per_token_group_quant_fp8() (in module nemo_rl.models.generation.fp8) pil_to_base64() (in module nemo_rl.data.datasets.utils) pipeline_dtype (nemo_rl.models.policy.MegatronConfig attribute) pipeline_model_parallel_size (nemo_rl.models.policy.MegatronConfig attribute) pipeline_parallel_size (nemo_rl.models.generation.vllm.config.VllmSpecificArgs attribute) Policy (class in nemo_rl.models.policy.lm_policy) policy (nemo_rl.algorithms.distillation.MasterConfig attribute) (nemo_rl.algorithms.dpo.MasterConfig attribute) (nemo_rl.algorithms.grpo.MasterConfig attribute) (nemo_rl.algorithms.rm.MasterConfig attribute) (nemo_rl.algorithms.sft.MasterConfig attribute) PolicyConfig (class in nemo_rl.models.policy) PolicyInterface (class in nemo_rl.models.policy.interfaces) post_init() (nemo_rl.models.generation.vllm.vllm_worker.VllmGenerationWorker method) post_init_async() (nemo_rl.models.generation.vllm.vllm_worker_async.VllmAsyncGenerationWorker method) PRE_RELEASE (in module nemo_rl.package_info) precision (nemo_rl.environments.reward_model_environment.RewardModelEnvironmentConfig attribute) (nemo_rl.models.generation.vllm.config.VllmSpecificArgs attribute) (nemo_rl.models.policy.PolicyConfig attribute) preference_average_log_probs (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.loss_functions.DPOLossConfig attribute) preference_collate_fn() (in module nemo_rl.data.collate_fn) preference_loss (nemo_rl.algorithms.dpo.DPOValMetrics attribute) preference_loss_weight (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.loss_functions.DPOLossConfig attribute) PreferenceDataset (class in nemo_rl.data.datasets.preference_datasets.preference_dataset)

PreferenceLoss (class in nemo_rl.algorithms.loss_functions) PreferenceLossDataDict (class in nemo_rl.algorithms.loss_functions) prefetch_venvs() (in module nemo_rl.utils.prefetch_venvs) prepare_clevr_cogent_dataset() (in module nemo_rl.data.datasets.response_datasets.clevr) prepare_dapo_math_17k_dataset() (in module nemo_rl.data.datasets.response_datasets.dapo_math) prepare_deepscaler_dataset() (in module nemo_rl.data.datasets.response_datasets.deepscaler) prepare_for_generation() (nemo_rl.models.generation.interfaces.GenerationInterface method) (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) (nemo_rl.models.policy.lm_policy.Policy method) prepare_for_lp_inference() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) prepare_for_refit() (nemo_rl.algorithms.async_utils.AsyncTrajectoryCollector method) prepare_for_training() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.interfaces.PolicyInterface method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) prepare_geometry3k_dataset() (in module nemo_rl.data.datasets.response_datasets.geometry3k) prepare_openinstructmath2_dataset() (in module nemo_rl.data.datasets.response_datasets.openmathinstruct2) prepare_refcoco_dataset() (in module nemo_rl.data.datasets.response_datasets.refcoco) prepare_refit_info() (nemo_rl.models.generation.interfaces.GenerationInterface method) (nemo_rl.models.generation.vllm.vllm_backend.VllmInternalWorkerExtension method) (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) (nemo_rl.models.generation.vllm.vllm_worker.VllmGenerationWorker method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.interfaces.ColocatablePolicyInterface method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) prepare_refit_info_async() (nemo_rl.models.generation.vllm.vllm_worker_async.VllmAsyncGenerationWorker method) preprocess_data() (nemo_rl.environments.reward_model_environment.RewardModelEnvironment method) PreservingDataset (class in nemo_rl.data.datasets.response_datasets.oai_format_dataset) prev_logprobs (nemo_rl.algorithms.loss_functions.ClippedPGLossDataDict attribute) print_aggregated_stats() (nemo_rl.data.packing.metrics.PackingMetrics method) print_message_log_samples() (in module nemo_rl.utils.logger) print_metrics() (nemo_rl.data.packing.algorithms.SequencePacker method) print_node_ip_and_gpu_id() (nemo_rl.models.policy.lm_policy.Policy method) print_performance_metrics() (in module nemo_rl.algorithms.utils) problem_key (nemo_rl.data.MathDataConfig attribute) process_weights_after_loading() (in module nemo_rl.models.generation.fp8) ProfilablePolicy (class in nemo_rl.utils.nsys) project (nemo_rl.utils.logger.SwanlabConfig attribute) (nemo_rl.utils.logger.WandbConfig attribute) projector_type (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) prompt_file (nemo_rl.data.DataConfig attribute) (nemo_rl.data.interfaces.TaskDataSpec attribute) push_with_wait_signal() (nemo_rl.algorithms.async_utils.ReplayBuffer method) PY_EXECUTABLES (class in nemo_rl.distributed.virtual_cluster) PytorchOptimizerConfig (class in nemo_rl.models.policy)

Q

q_lora_rank (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) qk_head_dim (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) qk_pos_emb_head_dim (nemo_rl.utils.flops_formulas.FLOPSConfig attribute)

query_groups (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) qwen2() (in module nemo_rl.utils.flops_formulas) qwen3() (in module nemo_rl.utils.flops_formulas)

R

ratio_clip_c (nemo_rl.algorithms.loss_functions.ClippedPGLossConfig attribute) ratio_clip_max (nemo_rl.algorithms.loss_functions.ClippedPGLossConfig attribute) ratio_clip_min (nemo_rl.algorithms.loss_functions.ClippedPGLossConfig attribute) RayGpuMonitorLogger (class in nemo_rl.utils.logger) RayVirtualCluster (class in nemo_rl.distributed.virtual_cluster) RayWorkerBuilder (class in nemo_rl.distributed.worker_groups) RayWorkerBuilder.IsolatedWorkerInitializer (class in nemo_rl.distributed.worker_groups) RayWorkerGroup (class in nemo_rl.distributed.worker_groups) rebalance_nd_tensor() (in module nemo_rl.distributed.collectives) rebuild_cuda_tensor_from_ipc() (in module nemo_rl.models.policy.utils) recursive_merge_options() (in module nemo_rl.distributed.worker_group_utils) reduce() (nemo_rl.utils.timer.Timer method) RefCOCODataset (class in nemo_rl.data.datasets.response_datasets.refcoco) reference_logprobs (nemo_rl.models.policy.interfaces.ReferenceLogprobOutputSpec attribute) reference_policy_kl_penalty (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.loss_functions.ClippedPGLossConfig attribute) (nemo_rl.algorithms.loss_functions.DPOLossConfig attribute) reference_policy_logprobs (nemo_rl.algorithms.loss_functions.ClippedPGLossDataDict attribute) (nemo_rl.algorithms.loss_functions.DPOLossDataDict attribute) ReferenceLogprobOutputSpec (class in nemo_rl.models.policy.interfaces) refit_buffer_size_gb (nemo_rl.models.policy.PolicyConfig attribute) refit_policy_generation() (in module nemo_rl.algorithms.grpo) remap_dataset_keys() (in module nemo_rl.data.llm_message_utils) remove_boxed() (in module nemo_rl.environments.dapo_math_verifier) remove_old_checkpoints() (nemo_rl.utils.checkpoint.CheckpointManager method) REMOVED_EXPRESSIONS (in module nemo_rl.environments.dapo_math_verifier) reorder_data() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) repeat_interleave() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) ReplayBuffer (class in nemo_rl.algorithms.async_utils) report_device_id() (nemo_rl.models.generation.vllm.vllm_backend.VllmInternalWorkerExtension method) (nemo_rl.models.generation.vllm.vllm_worker.VllmGenerationWorker method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) report_device_id_async() (nemo_rl.models.generation.vllm.vllm_worker_async.VllmAsyncGenerationWorker method) report_dp_openai_server_base_url() (nemo_rl.models.generation.vllm.vllm_worker_async.VllmAsyncGenerationWorker method) report_node_ip_and_gpu_id() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) reset() (nemo_rl.data.packing.metrics.PackingMetrics method) (nemo_rl.utils.flops_tracker.FLOPTracker method) (nemo_rl.utils.timer.Timer method) reset_metrics() (nemo_rl.data.packing.algorithms.SequencePacker method) reset_peak_memory_stats() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) reset_prefix_cache() (nemo_rl.models.generation.vllm.vllm_worker.VllmGenerationWorker method)

reset_prefix_cache_async() (nemo_rl.models.generation.vllm.vllm_worker_async.VllmAsyncGenerationWorker method) resolve_model_class() (in module nemo_rl.models.policy.utils) resolve_path() (in module nemo_rl.utils.config) ResourceInsufficientError resources (nemo_rl.environments.reward_model_environment.RewardModelEnvironmentConfig attribute) (nemo_rl.models.generation.interfaces.ColocationConfig attribute) ResourcesConfig (class in nemo_rl.models.generation.interfaces) ResponseDataset (class in nemo_rl.data.datasets.response_datasets.response_dataset) resume() (nemo_rl.algorithms.async_utils.AsyncTrajectoryCollector method) resume_after_refit() (nemo_rl.algorithms.async_utils.AsyncTrajectoryCollector method) return_from_workers (nemo_rl.distributed.worker_groups.MultiWorkerFuture attribute) return_model_config() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) return_state_dict() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) reward_functions (nemo_rl.environments.vlm_environment.VLMEnvConfig attribute) reward_model_cfg (nemo_rl.models.policy.PolicyConfig attribute) reward_model_type (nemo_rl.models.policy.RewardModelConfig attribute) reward_scaling (nemo_rl.algorithms.grpo.GRPOConfig attribute) reward_shaping (nemo_rl.algorithms.grpo.GRPOConfig attribute) RewardModelConfig (class in nemo_rl.models.policy) RewardModelEnvironment (class in nemo_rl.environments.reward_model_environment) RewardModelEnvironmentConfig (class in nemo_rl.environments.reward_model_environment) rewards (nemo_rl.environments.interfaces.EnvironmentReturn attribute) rewards_chosen_mean (nemo_rl.algorithms.dpo.DPOValMetrics attribute) (nemo_rl.algorithms.rm.RMValMetrics attribute) rewards_rejected_mean (nemo_rl.algorithms.dpo.DPOValMetrics attribute) (nemo_rl.algorithms.rm.RMValMetrics attribute) RewardScalingConfig (class in nemo_rl.algorithms.grpo) RewardShapingConfig (class in nemo_rl.algorithms.reward_functions) rl_collate_fn() (in module nemo_rl.data.collate_fn) rm (nemo_rl.algorithms.rm.MasterConfig attribute) rm_train() (in module nemo_rl.algorithms.rm) RMConfig (class in nemo_rl.algorithms.rm) RMSaveState (class in nemo_rl.algorithms.rm) RMValMetrics (class in nemo_rl.algorithms.rm) RotaryEmbedParallel (class in nemo_rl.models.dtensor.parallelize) run_all_workers_multiple_data() (nemo_rl.distributed.worker_groups.RayWorkerGroup method) run_all_workers_sharded_data() (nemo_rl.distributed.worker_groups.RayWorkerGroup method) run_all_workers_single_data() (nemo_rl.distributed.worker_groups.RayWorkerGroup method) run_async_multi_turn_rollout() (in module nemo_rl.experience.rollouts) run_env_eval() (in module nemo_rl.evals.eval) run_multi_turn_rollout() (in module nemo_rl.experience.rollouts) run_name (nemo_rl.utils.logger.MLflowConfig attribute) run_rollouts() (nemo_rl.environments.penguin.Penguin method) run_sample_multi_turn_rollout() (in module nemo_rl.experience.rollouts) run_single_worker_single_data() (nemo_rl.distributed.worker_groups.RayWorkerGroup method)

S

safe_import() (nemo_rl.environments.code_environment.CodeExecutionWorker method) safe_open() (nemo_rl.environments.code_environment.CodeExecutionWorker method) sample() (nemo_rl.algorithms.async_utils.ReplayBuffer method) sample_mask (nemo_rl.algorithms.loss_functions.ClippedPGLossDataDict attribute) (nemo_rl.algorithms.loss_functions.DistillationLossDataDict attribute) (nemo_rl.algorithms.loss_functions.DPOLossDataDict attribute) (nemo_rl.algorithms.loss_functions.PreferenceLossDataDict attribute) sanitize() (nemo_rl.environments.code_environment.CodeExecutionWorker method) save_checkpoint() (in module nemo_rl.utils.automodel_checkpoint) (in module nemo_rl.utils.native_checkpoint) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.interfaces.PolicyInterface method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) save_consolidated (nemo_rl.utils.checkpoint.CheckpointingConfig attribute) save_path (nemo_rl.evals.eval.EvalConfig attribute) save_period (nemo_rl.utils.checkpoint.CheckpointingConfig attribute) scale_rewards() (in module nemo_rl.algorithms.grpo) scheduler (nemo_rl.models.policy.MegatronConfig attribute) (nemo_rl.models.policy.PolicyConfig attribute) SchedulerMilestones (in module nemo_rl.models.policy) score() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.interfaces.PolicyInterface method) (nemo_rl.models.policy.lm_policy.Policy method) ScoreOutputSpec (class in nemo_rl.models.policy.interfaces) scores (nemo_rl.models.policy.interfaces.ScoreOutputSpec attribute) seed (nemo_rl.algorithms.distillation.DistillationConfig attribute) (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.grpo.GRPOConfig attribute) (nemo_rl.algorithms.rm.RMConfig attribute) (nemo_rl.algorithms.sft.SFTConfig attribute) (nemo_rl.data.DataConfig attribute) (nemo_rl.evals.eval.EvalConfig attribute) seen_params (nemo_rl.models.generation.fp8.FP8State attribute) select_indices() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) sequence_length_pad_multiple (nemo_rl.distributed.batched_data_dict.SequencePackingArgs attribute) sequence_length_round (nemo_rl.distributed.batched_data_dict.DynamicBatchingArgs attribute) (nemo_rl.models.policy.DynamicBatchingConfig attribute) SEQUENCE_LEVEL (nemo_rl.algorithms.interfaces.LossType attribute) sequence_level_importance_ratios (nemo_rl.algorithms.loss_functions.ClippedPGLossConfig attribute) sequence_packing (nemo_rl.environments.reward_model_environment.RewardModelEnvironmentConfig attribute) (nemo_rl.models.policy.PolicyConfig attribute) sequence_parallel (nemo_rl.models.policy.DTensorConfig attribute) (nemo_rl.models.policy.MegatronConfig attribute) SequencePacker (class in nemo_rl.data.packing.algorithms) SequencePackingArgs (class in nemo_rl.distributed.batched_data_dict) SequencePackingConfig (class in nemo_rl.models.policy) SequencePackingLossWrapper (class in nemo_rl.algorithms.loss_functions) set_seed() (in module nemo_rl.algorithms.utils) set_weight_version() (nemo_rl.algorithms.async_utils.AsyncTrajectoryCollector method) setup() (in module nemo_rl.algorithms.distillation) (in module nemo_rl.algorithms.dpo) (in module nemo_rl.algorithms.grpo) (in module nemo_rl.algorithms.rm) (in module nemo_rl.algorithms.sft) (in module nemo_rl.evals.eval) setup_megatron_model() (in module nemo_rl.models.policy.megatron_policy_worker) setup_penguin_config() (in module nemo_rl.environments.penguin) sft (nemo_rl.algorithms.sft.MasterConfig attribute) sft_average_log_probs (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.loss_functions.DPOLossConfig attribute) sft_loss (nemo_rl.algorithms.dpo.DPOValMetrics attribute) sft_loss_weight (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.loss_functions.DPOLossConfig attribute) sft_train() (in module nemo_rl.algorithms.sft) SFTConfig (class in nemo_rl.algorithms.sft) SFTSaveState (class in nemo_rl.algorithms.sft) sgd_momentum (nemo_rl.models.policy.MegatronOptimizerConfig attribute) shape (nemo_rl.distributed.named_sharding.NamedSharding property) shard_by_batch_size() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) shuffle (nemo_rl.data.DataConfig attribute) shutdown() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) (nemo_rl.distributed.worker_groups.RayWorkerGroup method) (nemo_rl.environments.code_environment.CodeEnvironment method) (nemo_rl.environments.math_environment.MathEnvironment method) (nemo_rl.environments.penguin.Penguin method) (nemo_rl.environments.reward_model_environment.RewardModelEnvironment method) (nemo_rl.environments.vlm_environment.VLMEnvironment method) (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) (nemo_rl.models.generation.vllm.vllm_worker.VllmGenerationWorker method) (nemo_rl.models.generation.vllm.vllm_worker_async.VllmAsyncGenerationWorker method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.interfaces.PolicyInterface method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method)

simple_role_header (nemo_rl.data.chat_templates.COMMON_CHAT_TEMPLATES attribute) SinglePytorchSchedulerConfig (class in nemo_rl.models.policy) size (nemo_rl.distributed.batched_data_dict.BatchedDataDict property) (nemo_rl.distributed.named_sharding.NamedSharding property) size() (nemo_rl.algorithms.async_utils.ReplayBuffer method) skip_tokenizer_init (nemo_rl.models.generation.vllm.config.VllmSpecificArgs attribute) sleep() (nemo_rl.models.generation.vllm.vllm_worker.VllmGenerationWorker method) sleep_async() (nemo_rl.models.generation.vllm.vllm_worker_async.VllmAsyncGenerationWorker method) slice() (nemo_rl.data.multimodal_utils.PackedTensor method) (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) SlicedDataDict (class in nemo_rl.distributed.batched_data_dict) sliding_window_overwrite() (in module nemo_rl.models.policy.utils) solution_key (nemo_rl.data.MathDataConfig attribute) source_max (nemo_rl.algorithms.grpo.RewardScalingConfig attribute) source_min (nemo_rl.algorithms.grpo.RewardScalingConfig attribute) split (nemo_rl.data.DataConfig attribute) split_output_tensor() (nemo_rl.algorithms.loss_functions.PreferenceLoss method) SquadDataset (class in nemo_rl.data.datasets.response_datasets.squad) start() (nemo_rl.utils.logger.RayGpuMonitorLogger method) (nemo_rl.utils.timer.Timer method) start_collection() (nemo_rl.algorithms.async_utils.AsyncTrajectoryCollector method) start_gpu_profiling() (nemo_rl.models.generation.vllm.vllm_backend.VllmInternalWorkerExtension method) (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) (nemo_rl.models.generation.vllm.vllm_worker.BaseVllmGenerationWorker method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) (nemo_rl.utils.nsys.ProfilablePolicy method) start_iterations() (nemo_rl.utils.timer.TimeoutChecker method) start_weight_decay (nemo_rl.models.policy.MegatronSchedulerConfig attribute) state_dict() (nemo_rl.utils.native_checkpoint.ModelState method) (nemo_rl.utils.native_checkpoint.OptimizerState method) step (nemo_rl.algorithms.dpo.DPOSaveState attribute) (nemo_rl.algorithms.rm.RMSaveState attribute) (nemo_rl.algorithms.sft.SFTSaveState attribute) (nemo_rl.utils.logger.GpuMetricSnapshot attribute) step() (nemo_rl.environments.code_environment.CodeEnvironment method) (nemo_rl.environments.interfaces.EnvironmentInterface method) (nemo_rl.environments.math_environment.MathEnvironment method) (nemo_rl.environments.penguin.Penguin method) (nemo_rl.environments.reward_model_environment.RewardModelEnvironment method) (nemo_rl.environments.vlm_environment.VLMEnvironment method) stop() (nemo_rl.utils.logger.RayGpuMonitorLogger method) (nemo_rl.utils.timer.Timer method) stop_gpu_profiling() (nemo_rl.models.generation.vllm.vllm_backend.VllmInternalWorkerExtension method) (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) (nemo_rl.models.generation.vllm.vllm_worker.BaseVllmGenerationWorker method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) (nemo_rl.utils.nsys.ProfilablePolicy method) stop_strings (nemo_rl.data.interfaces.DatumSpec attribute) (nemo_rl.environments.math_environment.MathEnvConfig attribute) (nemo_rl.environments.vlm_environment.VLMEnvConfig attribute) (nemo_rl.models.generation.interfaces.GenerationConfig attribute) (nemo_rl.models.generation.interfaces.GenerationDatumSpec attribute) stop_token_ids (nemo_rl.models.generation.interfaces.GenerationConfig attribute) stream_weights_via_ipc_zmq() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.interfaces.ColocatablePolicyInterface method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) stream_weights_via_ipc_zmq_impl() (in module nemo_rl.models.policy.utils) SUBSTITUTIONS (in module nemo_rl.environments.dapo_math_verifier) surpress_user_warnings() (in module nemo_rl.algorithms.utils) swanlab (nemo_rl.utils.logger.LoggerConfig attribute) swanlab_enabled (nemo_rl.utils.logger.LoggerConfig attribute) SwanlabConfig (class in nemo_rl.utils.logger) SwanlabLogger (class in nemo_rl.utils.logger) SYSTEM (nemo_rl.distributed.virtual_cluster.PY_EXECUTABLES attribute) SYSTEM_PROMPT (in module nemo_rl.data.datasets.response_datasets.oasst) system_prompt_file (nemo_rl.data.DataConfig attribute) (nemo_rl.data.interfaces.TaskDataSpec attribute)

T

T (in module nemo_rl.distributed.collectives) target_max (nemo_rl.algorithms.grpo.RewardScalingConfig attribute) target_min (nemo_rl.algorithms.grpo.RewardScalingConfig attribute) task_name (nemo_rl.data.interfaces.DatumSpec attribute) (nemo_rl.data.interfaces.TaskDataSpec attribute) TaskDataProcessFnCallable (class in nemo_rl.data.interfaces) TaskDataSpec (class in nemo_rl.data.interfaces) teacher (nemo_rl.algorithms.distillation.MasterConfig attribute) teacher_topk_indices (nemo_rl.algorithms.loss_functions.DistillationLossDataDict attribute) teacher_topk_logits (nemo_rl.algorithms.loss_functions.DistillationLossDataDict attribute) temperature (nemo_rl.models.generation.interfaces.GenerationConfig attribute) Tensor (in module nemo_rl.algorithms.loss_functions) (in module nemo_rl.algorithms.reward_functions) (in module nemo_rl.data.llm_message_utils) (in module nemo_rl.models.huggingface.common) tensor_model_parallel_size (nemo_rl.models.policy.MegatronConfig attribute) tensor_parallel_size (nemo_rl.models.generation.vllm.config.VllmSpecificArgs attribute) (nemo_rl.models.policy.DTensorConfig attribute) tensorboard (nemo_rl.utils.logger.LoggerConfig attribute) tensorboard_enabled (nemo_rl.utils.logger.LoggerConfig attribute) TensorboardConfig (class in nemo_rl.utils.logger) TensorboardLogger (class in nemo_rl.utils.logger) terminate_on_evaluation (nemo_rl.environments.code_environment.CodeEnvConfig attribute) terminateds (nemo_rl.environments.interfaces.EnvironmentReturn attribute) THEORETICAL_TFLOPS (in module nemo_rl.utils.flops_tracker) time() (nemo_rl.utils.timer.Timer method) TimeoutChecker (class in nemo_rl.utils.timer) Timer (class in nemo_rl.utils.timer) to() (nemo_rl.data.multimodal_utils.PackedTensor method) (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) to_local_if_dtensor() (in module nemo_rl.models.dtensor.parallelize) to_preference_data_format() (in module nemo_rl.data.datasets.preference_datasets.binary_preference_dataset) (in module nemo_rl.data.datasets.preference_datasets.helpsteer3) (in module nemo_rl.data.datasets.preference_datasets.tulu3) TOKEN_LEVEL (nemo_rl.algorithms.interfaces.LossType attribute) token_level_loss (nemo_rl.algorithms.loss_functions.ClippedPGLossConfig attribute) token_mask (nemo_rl.algorithms.loss_functions.ClippedPGLossDataDict attribute) (nemo_rl.algorithms.loss_functions.DistillationLossDataDict attribute) (nemo_rl.algorithms.loss_functions.DPOLossDataDict attribute) (nemo_rl.algorithms.loss_functions.PreferenceLossDataDict attribute) tokenizer (nemo_rl.evals.eval.MasterConfig attribute) (nemo_rl.models.policy.PolicyConfig attribute) TokenizerConfig (class in nemo_rl.models.policy) TokenizerType (in module nemo_rl.algorithms.async_utils) (in module nemo_rl.algorithms.distillation) (in module nemo_rl.algorithms.grpo) (in module nemo_rl.data.collate_fn) (in module nemo_rl.data.datasets.processed_dataset) (in module nemo_rl.data.datasets.utils) (in module nemo_rl.data.interfaces) (in module nemo_rl.data.llm_message_utils) (in module nemo_rl.data.processors) (in module nemo_rl.experience.rollouts) (in module nemo_rl.models.generation) (in module nemo_rl.models.policy.megatron_policy_worker)

top_k (nemo_rl.models.generation.interfaces.GenerationConfig attribute) TOP_K_THRESHOLD (in module nemo_rl.models.generation.vllm.vllm_generation) top_p (nemo_rl.models.generation.interfaces.GenerationConfig attribute) TOP_P_THRESHOLD (in module nemo_rl.models.generation.vllm.vllm_generation) topk_indices (nemo_rl.models.policy.interfaces.TopkLogitsOutputSpec attribute) topk_logits (nemo_rl.models.policy.interfaces.TopkLogitsOutputSpec attribute) topk_logits_k (nemo_rl.algorithms.distillation.DistillationConfig attribute) TopkLogitsOutputSpec (class in nemo_rl.models.policy.interfaces) total_steps (nemo_rl.algorithms.distillation.DistillationSaveState attribute) (nemo_rl.algorithms.dpo.DPOSaveState attribute) (nemo_rl.algorithms.grpo.GRPOSaveState attribute) (nemo_rl.algorithms.rm.RMSaveState attribute) (nemo_rl.algorithms.sft.SFTSaveState attribute) total_valid_tokens (nemo_rl.algorithms.distillation.DistillationSaveState attribute) (nemo_rl.algorithms.dpo.DPOSaveState attribute) (nemo_rl.algorithms.grpo.GRPOSaveState attribute) (nemo_rl.algorithms.rm.RMSaveState attribute) (nemo_rl.algorithms.sft.SFTSaveState attribute) track() (nemo_rl.utils.flops_tracker.FLOPTracker method) track_batch() (nemo_rl.utils.flops_tracker.FLOPTracker method) tracking_uri (nemo_rl.utils.logger.MLflowConfig attribute) train() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.interfaces.PolicyInterface method) (nemo_rl.models.policy.lm_policy.Policy method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) train_context() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker static method) train_data_path (nemo_rl.data.DataConfig attribute) train_global_batch_size (nemo_rl.models.policy.PolicyConfig attribute) train_mb_tokens (nemo_rl.models.policy.DynamicBatchingConfig attribute) (nemo_rl.models.policy.SequencePackingConfig attribute) train_micro_batch_size (nemo_rl.models.policy.PolicyConfig attribute) transformer() (in module nemo_rl.utils.flops_formulas) translate_parallel_style() (in module nemo_rl.models.dtensor.parallelize) truncate_tensors() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) truncated_importance_sampling_ratio (nemo_rl.algorithms.loss_functions.ClippedPGLossConfig attribute) Tulu3PreferenceDataset (class in nemo_rl.data.datasets.preference_datasets.tulu3)

U

unpack_tensor() (in module nemo_rl.models.huggingface.common) unpadded_sequence_lengths (nemo_rl.models.generation.interfaces.GenerationOutputSpec attribute) unshard_fsdp2_model() (in module nemo_rl.models.policy.dtensor_policy_worker) update() (nemo_rl.data.packing.metrics.PackingMetrics method) update_weights_from_collective() (nemo_rl.models.generation.interfaces.GenerationInterface method) (nemo_rl.models.generation.vllm.vllm_backend.VllmInternalWorkerExtension method) (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) (nemo_rl.models.generation.vllm.vllm_worker.VllmGenerationWorker method) update_weights_from_collective_async() (nemo_rl.models.generation.vllm.vllm_worker_async.VllmAsyncGenerationWorker method) update_weights_via_ipc_zmq() (nemo_rl.models.generation.interfaces.GenerationInterface method) (nemo_rl.models.generation.vllm.vllm_backend.VllmInternalWorkerExtension method) (nemo_rl.models.generation.vllm.vllm_generation.VllmGeneration method) (nemo_rl.models.generation.vllm.vllm_worker.VllmGenerationWorker method)

update_weights_via_ipc_zmq_async() (nemo_rl.models.generation.vllm.vllm_worker_async.VllmAsyncGenerationWorker method) use_activation_pow2_scale (nemo_rl.models.generation.fp8.FP8Config attribute) use_custom_fsdp (nemo_rl.models.policy.MegatronDDPConfig attribute) use_distributed_optimizer (nemo_rl.models.policy.MegatronOptimizerConfig attribute) use_dynamic_sampling (nemo_rl.algorithms.grpo.GRPOConfig attribute) use_importance_sampling_correction (nemo_rl.algorithms.loss_functions.ClippedPGLossConfig attribute) use_leave_one_out_baseline (nemo_rl.algorithms.grpo.GRPOConfig attribute) use_on_policy_kl_approximation (nemo_rl.algorithms.loss_functions.ClippedPGLossConfig attribute) use_precision_aware_optimizer (nemo_rl.models.policy.MegatronOptimizerConfig attribute) use_reference_model() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2 method) (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method) USE_SYSTEM_EXECUTABLE (in module nemo_rl.distributed.ray_actor_environment_registry) use_weight_pow2_scale (nemo_rl.models.generation.fp8.FP8Config attribute)

V

v_head_dim (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) val_at_start (nemo_rl.algorithms.distillation.DistillationConfig attribute) (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.grpo.GRPOConfig attribute) (nemo_rl.algorithms.rm.RMConfig attribute) (nemo_rl.algorithms.sft.SFTConfig attribute) val_batch_size (nemo_rl.algorithms.distillation.DistillationConfig attribute) (nemo_rl.algorithms.grpo.GRPOConfig attribute) val_batches (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.rm.RMConfig attribute) (nemo_rl.algorithms.sft.SFTConfig attribute) val_data_paths (nemo_rl.data.DataConfig attribute) val_dataset_name (nemo_rl.data.DataConfig attribute) val_global_batch_size (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.rm.RMConfig attribute) (nemo_rl.algorithms.sft.SFTConfig attribute) val_loss (nemo_rl.algorithms.sft.SFTSaveState attribute) val_micro_batch_size (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.rm.RMConfig attribute) (nemo_rl.algorithms.sft.SFTConfig attribute) val_period (nemo_rl.algorithms.distillation.DistillationConfig attribute) (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.grpo.GRPOConfig attribute) (nemo_rl.algorithms.rm.RMConfig attribute) (nemo_rl.algorithms.sft.SFTConfig attribute) val_reward (nemo_rl.algorithms.distillation.DistillationSaveState attribute) (nemo_rl.algorithms.grpo.GRPOSaveState attribute) validate() (in module nemo_rl.algorithms.distillation) (in module nemo_rl.algorithms.dpo) (in module nemo_rl.algorithms.grpo) (in module nemo_rl.algorithms.rm) (in module nemo_rl.algorithms.sft)

validate_one_dataset() (in module nemo_rl.algorithms.dpo) (in module nemo_rl.algorithms.rm) vec_in_dim (nemo_rl.utils.flops_formulas.FLOPSConfig attribute) verifier_type (nemo_rl.environments.math_environment.MathEnvConfig attribute) verify() (in module nemo_rl.environments.dapo_math_verifier) (nemo_rl.environments.math_environment.EnglishMultichoiceVerifyWorker method) (nemo_rl.environments.math_environment.HFVerifyWorker method) (nemo_rl.environments.math_environment.MultilingualMultichoiceVerifyWorker method) (nemo_rl.environments.vlm_environment.VLMVerifyWorker method) verify_right_padding() (in module nemo_rl.models.generation.interfaces) VERSION (in module nemo_rl.package_info) VLLM (nemo_rl.distributed.virtual_cluster.PY_EXECUTABLES attribute) vllm_cfg (nemo_rl.models.generation.vllm.config.VllmConfig attribute) VLLM_EXECUTABLE (in module nemo_rl.distributed.ray_actor_environment_registry) vllm_kwargs (nemo_rl.models.generation.vllm.config.VllmConfig attribute) VLLM_LOAD_FORMAT_AUTO (nemo_rl.models.huggingface.common.ModelFlag attribute) vllm_patches (nemo_rl.models.generation.fp8.FP8State attribute) VllmAsyncGenerationWorker (class in nemo_rl.models.generation.vllm.vllm_worker_async) VllmConfig (class in nemo_rl.models.generation.vllm.config) VllmGeneration (class in nemo_rl.models.generation.vllm.vllm_generation) VllmGenerationWorker (class in nemo_rl.models.generation.vllm.vllm_worker) VllmInternalWorkerExtension (class in nemo_rl.models.generation.vllm.vllm_backend) VllmSpecificArgs (class in nemo_rl.models.generation.vllm.config) VLMEnvConfig (class in nemo_rl.environments.vlm_environment) VLMEnvironment (class in nemo_rl.environments.vlm_environment) VLMEnvironmentMetadata (class in nemo_rl.environments.vlm_environment) VLMVerifyWorker (class in nemo_rl.environments.vlm_environment) vocab_size (nemo_rl.utils.flops_formulas.FLOPSConfig attribute)

W

wait_for_pending_generations() (nemo_rl.algorithms.async_utils.AsyncTrajectoryCollector method) wake_up() (nemo_rl.models.generation.vllm.vllm_worker.VllmGenerationWorker method) wake_up_async() (nemo_rl.models.generation.vllm.vllm_worker_async.VllmAsyncGenerationWorker method) wandb (nemo_rl.utils.logger.LoggerConfig attribute) wandb_enabled (nemo_rl.utils.logger.LoggerConfig attribute) WandbConfig (class in nemo_rl.utils.logger) WandbLogger (class in nemo_rl.utils.logger)

weight_decay (nemo_rl.models.policy.MegatronOptimizerConfig attribute) weight_decay_incr_style (nemo_rl.models.policy.MegatronSchedulerConfig attribute) worker_metadata (nemo_rl.distributed.worker_groups.RayWorkerGroup property) workers (nemo_rl.distributed.worker_groups.RayWorkerGroup property) working_dir (nemo_rl.environments.code_environment.CodeEnvMetadata attribute) world_size() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) wrap_with_nvtx_name() (in module nemo_rl.utils.nsys)

Z

zero_out_weights() (nemo_rl.models.policy.megatron_policy_worker.MegatronPolicyWorker method)

zero_outside_topk (nemo_rl.algorithms.loss_functions.DistillationLossConfig attribute)