Index

_

__all__ (in module nemo_rl.data.hf_datasets) __call__() (nemo_rl.algorithms.interfaces.LossFunction method) (nemo_rl.algorithms.loss_functions.ClippedPGLossFn method) (nemo_rl.algorithms.loss_functions.DPOLossFn method) (nemo_rl.algorithms.loss_functions.NLLLoss 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.models.generation.vllm.VllmGeneration method) (nemo_rl.models.policy.hf_policy.HfPolicy method) (nemo_rl.utils.logger.Logger method) __description__ (in module nemo_rl.package_info) __download_url__ (in module nemo_rl.package_info) __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.AllTaskProcessedDataset method) __homepage__ (in module nemo_rl.package_info) __keywords__ (in module nemo_rl.package_info) __len__() (nemo_rl.data.datasets.AllTaskProcessedDataset 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.models.generation.vllm.VllmGenerationWorker method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker 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.fsdp1_policy_worker.FSDP1PolicyWorker 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.utils.logger.RayGpuMonitorLogger method) _compute_distributed_log_softmax() (in module nemo_rl.distributed.model_utils) _create_visualization_grid() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) _create_workers_from_bundle_indices() (nemo_rl.distributed.worker_groups.RayWorkerGroup method) _default_dpo_save_state() (in module nemo_rl.algorithms.dpo) _default_grpo_save_state() (in module nemo_rl.algorithms.grpo) _default_sft_save_state() (in module nemo_rl.algorithms.sft) _fetch_and_parse_metrics() (nemo_rl.utils.logger.RayGpuMonitorLogger method) _get_node_ip_and_free_port() (in module nemo_rl.distributed.virtual_cluster) _get_tied_worker_bundle_indices() (nemo_rl.models.generation.vllm.VllmGeneration method) (nemo_rl.models.policy.hf_policy.HfPolicy method) _get_worker_cells() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) _init_placement_groups() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) _load_checkpoint_history() (in module nemo_rl.utils.checkpoint) _mute_output() (in module nemo_rl.environments.math_environment) _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_qwen() (in module nemo_rl.models.dtensor.parallelize) _parse_gpu_metric() (nemo_rl.utils.logger.RayGpuMonitorLogger method) _parse_gpu_sku() (nemo_rl.utils.logger.RayGpuMonitorLogger method) _prepare_input_fn() (nemo_rl.models.dtensor.parallelize.RotaryEmbedParallel static method) _prepare_output_fn() (nemo_rl.models.dtensor.parallelize.RotaryEmbedParallel static method) _print_legend() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) _print_visualization() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) _REDUCTION_FUNCTIONS (nemo_rl.utils.timer.Timer attribute) _rich_logging_configured (in module nemo_rl.utils.logger) _validate_tensor_consistency() (in module nemo_rl.data.llm_message_utils)

A

activation_checkpointing (nemo_rl.models.policy.DTensorConfig attribute) activation_checkpointing_enabled (nemo_rl.models.policy.PolicyConfig attribute) add_bos (nemo_rl.data.DataConfig attribute) add_eos (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.hf_datasets.prompt_response_dataset.PromptResponseDataset method) add_ref_logprobs_to_data() (in module nemo_rl.algorithms.dpo) advantages (nemo_rl.algorithms.loss_functions.ClippedPGLossDataDict attribute) all_gather() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) AllTaskProcessedDataset (class in nemo_rl.data.datasets)

B

backend (nemo_rl.models.generation.interfaces.GenerationConfig attribute) backward() (nemo_rl.distributed.model_utils.DistributedLogprob static method)

BASE (nemo_rl.distributed.virtual_cluster.PY_EXECUTABLES attribute) batched_message_log_to_flat_message() (in module nemo_rl.data.llm_message_utils) BatchedDataDict (class in nemo_rl.distributed.batched_data_dict)

C

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) chat_template (nemo_rl.models.policy.TokenizerConfig attribute) checkpoint_dir (nemo_rl.algorithms.grpo.GRPOConfig attribute) (nemo_rl.utils.checkpoint.CheckpointingConfig attribute) checkpointing (nemo_rl.algorithms.dpo.MasterConfig attribute) (nemo_rl.algorithms.grpo.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) 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.dpo.MasterConfig attribute) (nemo_rl.algorithms.grpo.MasterConfig attribute) (nemo_rl.algorithms.sft.MasterConfig attribute) (nemo_rl.evals.eval.MasterConfig attribute) ClusterConfig (class in nemo_rl.distributed.virtual_cluster) collection_interval (nemo_rl.utils.logger.GPUMonitoringConfig attribute) COMMON_CHAT_TEMPLATES (class in nemo_rl.data.hf_datasets.chat_templates) configure_generation_config() (in module nemo_rl.models.generation.interfaces) configure_rich_logging() (in module nemo_rl.utils.logger) configure_worker() (nemo_rl.models.generation.vllm.VllmGenerationWorker static method) consumed_samples (nemo_rl.algorithms.dpo.DPOSaveState attribute) (nemo_rl.algorithms.grpo.GRPOSaveState attribute) (nemo_rl.algorithms.sft.SFTSaveState attribute) convert_dcp_to_hf() (in module nemo_rl.utils.native_checkpoint) copy_defaults() (nemo_rl.data.interfaces.TaskDataSpec method) cpu_offload (nemo_rl.models.policy.DTensorConfig attribute) create_local_venv() (in module nemo_rl.utils.venvs)

D

data (nemo_rl.algorithms.dpo.MasterConfig attribute) (nemo_rl.algorithms.grpo.MasterConfig attribute) (nemo_rl.algorithms.sft.MasterConfig attribute) (nemo_rl.evals.eval.MasterConfig 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.hf_datasets.deepscaler) DEFAULT_PY_EXECUTABLE (nemo_rl.environments.math_environment.HFVerifyWorker attribute) (nemo_rl.environments.math_environment.MathEnvironment attribute) (nemo_rl.models.generation.vllm.VllmGenerationWorker attribute) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker attribute) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker attribute) DEFAULT_VENV_DIR (in module nemo_rl.utils.venvs) define_metric() (nemo_rl.utils.logger.WandbLogger method) 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)

DistributedLogprob (class in nemo_rl.distributed.model_utils) download_and_process_oasst() (in module nemo_rl.data.hf_datasets.oasst) dpo (nemo_rl.algorithms.dpo.MasterConfig attribute) dpo_collate_fn() (in module nemo_rl.data.datasets) dpo_train() (in module nemo_rl.algorithms.dpo) DPOConfig (class in nemo_rl.algorithms.dpo) DPODataset (class in nemo_rl.data.hf_datasets.dpo) 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) dtensor_cfg (nemo_rl.models.policy.PolicyConfig attribute) DTensorConfig (class in nemo_rl.models.policy) DTensorPolicyWorker (class in nemo_rl.models.policy.dtensor_policy_worker) dynamic_batching (nemo_rl.models.policy.PolicyConfig attribute) DynamicBatchingCfg (class in nemo_rl.distributed.batched_data_dict) DynamicBatchingConfig (class in nemo_rl.models.policy)

E

enabled (nemo_rl.models.policy.DTensorConfig attribute) (nemo_rl.models.policy.DynamicBatchingConfig attribute) (nemo_rl.utils.checkpoint.CheckpointingConfig attribute) encode_single() (nemo_rl.data.datasets.AllTaskProcessedDataset method) env (nemo_rl.evals.eval.MasterConfig attribute) env_configs (nemo_rl.algorithms.grpo.MasterConfig 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.sft.SFTSaveState attribute) eval (nemo_rl.evals.eval.MasterConfig attribute) eval_collate_fn() (in module nemo_rl.data.datasets) EvalConfig (class in nemo_rl.evals.eval) extra_env_info (nemo_rl.data.interfaces.DatumSpec attribute)

F

finalize_checkpoint() (nemo_rl.utils.checkpoint.CheckpointManager method) finish_generation() (nemo_rl.models.generation.interfaces.GenerationInterface method) (nemo_rl.models.generation.vllm.VllmGeneration method) (nemo_rl.models.policy.hf_policy.HfPolicy method) finish_training() (nemo_rl.models.interfaces.PolicyInterface method) (nemo_rl.models.policy.hf_policy.HfPolicy method) FlatMessagesType (in module nemo_rl.data.interfaces) flatten_dict() (in module nemo_rl.utils.logger) flush() (nemo_rl.utils.logger.RayGpuMonitorLogger method) flush_interval (nemo_rl.utils.logger.GPUMonitoringConfig attribute)

format_helpsteer3() (in module nemo_rl.data.hf_datasets.helpsteer3) format_math() (in module nemo_rl.data.hf_datasets.deepscaler) (in module nemo_rl.data.hf_datasets.openmathinstruct2) format_squad() (in module nemo_rl.data.hf_datasets.squad) forward() (nemo_rl.distributed.model_utils.DistributedLogprob static method) from_batches() (nemo_rl.distributed.batched_data_dict.BatchedDataDict class method) from_parallel_logits_to_logprobs() (in module nemo_rl.distributed.model_utils) FSDP1PolicyWorker (class in nemo_rl.models.policy.fsdp1_policy_worker) fsdp_offload_enabled (nemo_rl.models.policy.PolicyConfig attribute) futures (nemo_rl.distributed.worker_groups.MultiWorkerFuture attribute)

G

gather_jagged_object_lists() (in module nemo_rl.distributed.collectives) generate (nemo_rl.evals.eval.MasterConfig attribute) generate() (nemo_rl.models.generation.interfaces.GenerationInterface method) (nemo_rl.models.generation.vllm.VllmGeneration method) (nemo_rl.models.generation.vllm.VllmGenerationWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) (nemo_rl.models.policy.hf_policy.HfPolicy method) generate_responses() (in module nemo_rl.experience.rollouts) generate_text() (nemo_rl.models.generation.vllm.VllmGeneration method) (nemo_rl.models.generation.vllm.VllmGenerationWorker method) generation (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) get_all_worker_results() (nemo_rl.distributed.worker_groups.RayWorkerGroup 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.hf_datasets.oasst) get_device_uuid() (in module nemo_rl.utils.nvml) get_dict() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) get_elapsed() (nemo_rl.utils.timer.Timer method) get_first_index_that_differs() (in module nemo_rl.data.llm_message_utils) get_formatted_message_log() (in module nemo_rl.data.llm_message_utils) get_gpu_info() (in module nemo_rl.models.policy.utils) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) get_grad_norm() (in module nemo_rl.models.dtensor.parallelize) get_keys_from_message_log() (in module nemo_rl.data.llm_message_utils) get_latest_checkpoint_path() (nemo_rl.utils.checkpoint.CheckpointManager method)

get_latest_elapsed() (nemo_rl.utils.timer.Timer method) get_logprobs() (nemo_rl.models.interfaces.PolicyInterface method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) (nemo_rl.models.policy.hf_policy.HfPolicy method) get_logprobs_from_vocab_parallel_logits() (in module nemo_rl.models.dtensor.parallelize) get_master_address_and_port() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) get_next_experiment_dir() (in module nemo_rl.utils.logger) get_placement_groups() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) get_reference_policy_logprobs() (nemo_rl.models.interfaces.PolicyInterface method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) (nemo_rl.models.policy.hf_policy.HfPolicy method) get_results() (nemo_rl.distributed.worker_groups.MultiWorkerFuture method) get_timing_metrics() (nemo_rl.utils.timer.Timer method) get_tokenizer() (in module nemo_rl.algorithms.utils) get_weights_ipc_handles() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) (nemo_rl.models.policy.hf_policy.HfPolicy method) git_root (in module nemo_rl.distributed.virtual_cluster) (in module nemo_rl.utils.venvs) global_post_process_and_metrics() (nemo_rl.environments.interfaces.EnvironmentInterface method) (nemo_rl.environments.math_environment.MathEnvironment method) gpu_memory_utilization (nemo_rl.models.generation.vllm.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) ground_truth (nemo_rl.environments.math_environment.MathEnvironmentMetadata attribute) group_count (nemo_rl.distributed.worker_groups.RayWorkerGroup property) grpo (nemo_rl.algorithms.grpo.MasterConfig attribute) grpo_train() (in module nemo_rl.algorithms.grpo) GRPOConfig (class in nemo_rl.algorithms.grpo) GRPOSaveState (class in nemo_rl.algorithms.grpo)

H

HelpSteer3Dataset (class in nemo_rl.data.hf_datasets.helpsteer3) HfPolicy (class in nemo_rl.models.policy.hf_policy)

HFVerifyWorker (class in nemo_rl.environments.math_environment) higher_is_better (nemo_rl.utils.checkpoint.CheckpointingConfig attribute)

I

idx (nemo_rl.data.interfaces.DatumSpec attribute) import_class_from_path() (in module nemo_rl.models.policy.utils) init_ray() (in module nemo_rl.distributed.virtual_cluster) init_tmp_checkpoint() (nemo_rl.utils.checkpoint.CheckpointManager method) input_ids (nemo_rl.algorithms.loss_functions.ClippedPGLossDataDict attribute) (nemo_rl.algorithms.loss_functions.DPOLossDataDict attribute) (nemo_rl.models.generation.interfaces.GenerationDatumSpec attribute)

input_key (nemo_rl.data.DataConfig attribute) (nemo_rl.distributed.batched_data_dict.DynamicBatchingCfg attribute) input_lengths (nemo_rl.models.generation.interfaces.GenerationDatumSpec attribute) input_lengths_key (nemo_rl.distributed.batched_data_dict.DynamicBatchingCfg attribute) is_alive() (nemo_rl.models.generation.vllm.VllmGenerationWorker method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) is_gemma3_model() (in module nemo_rl.models.huggingface.common)

K

keep_top_k (nemo_rl.utils.checkpoint.CheckpointingConfig attribute)

L

learning_rate (nemo_rl.models.policy.PolicyConfig attribute) length (nemo_rl.data.interfaces.DatumSpec attribute) llm() (nemo_rl.models.generation.vllm.VllmGenerationWorker method) LLMMessageLogType (in module nemo_rl.data.interfaces) load_checkpoint() (in module nemo_rl.utils.native_checkpoint) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) load_config() (in module nemo_rl.utils.config) load_config_with_inheritance() (in module nemo_rl.utils.config) 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) 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.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.TensorboardLogger method) (nemo_rl.utils.logger.WandbLogger 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.dpo.MasterConfig attribute) (nemo_rl.algorithms.grpo.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.models.policy.PolicyConfig attribute) logprob_mb_tokens (nemo_rl.models.policy.DynamicBatchingConfig attribute) logprobs (nemo_rl.models.generation.interfaces.GenerationOutputSpec attribute) loss_fn (nemo_rl.algorithms.grpo.MasterConfig attribute) loss_multiplier (nemo_rl.data.interfaces.DatumSpec attribute) loss_type (nemo_rl.algorithms.loss_functions.NLLLoss attribute) LossFunction (class in nemo_rl.algorithms.interfaces) LossType (class in nemo_rl.algorithms.loss_functions)

M

MAJOR (in module nemo_rl.package_info) make_microbatch_iterator() (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) manual_load_to_gpu() (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) manual_offload_to_cpu() (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) masked_mean() (in module nemo_rl.algorithms.utils) MasterConfig (class in nemo_rl.algorithms.dpo) (class in nemo_rl.algorithms.grpo) (class in nemo_rl.algorithms.sft) (class in nemo_rl.evals.eval) matches() (nemo_rl.models.huggingface.common.ModelFlag method) MathDataConfig (class in nemo_rl.data) 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.models.policy.PolicyConfig attribute) max_input_seq_length (nemo_rl.data.DataConfig attribute) max_model_len (nemo_rl.models.generation.vllm.VllmSpecificArgs attribute) max_new_tokens (nemo_rl.models.generation.interfaces.GenerationConfig attribute) max_num_epochs (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.sft.SFTConfig attribute) max_num_steps (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.grpo.GRPOConfig attribute) (nemo_rl.algorithms.sft.SFTConfig attribute) max_tokens_per_microbatch (nemo_rl.distributed.batched_data_dict.DynamicBatchingCfg attribute) message_log (nemo_rl.data.interfaces.DatumSpec 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) metric (nemo_rl.evals.eval.EvalConfig attribute) metric_name (nemo_rl.utils.checkpoint.CheckpointingConfig attribute) metrics (nemo_rl.utils.logger.GpuMetricSnapshot attribute) MINOR (in module nemo_rl.package_info) model_name (nemo_rl.models.generation.interfaces.GenerationConfig attribute) (nemo_rl.models.policy.PolicyConfig attribute) ModelFlag (class in nemo_rl.models.huggingface.common) ModelState (class in nemo_rl.utils.native_checkpoint) module nemo_rl nemo_rl.algorithms nemo_rl.algorithms.dpo nemo_rl.algorithms.grpo nemo_rl.algorithms.interfaces nemo_rl.algorithms.loss_functions nemo_rl.algorithms.sft nemo_rl.algorithms.utils nemo_rl.converters nemo_rl.converters.huggingface nemo_rl.converters.huggingface.vllm_export nemo_rl.converters.megatron nemo_rl.converters.megatron.vllm_export nemo_rl.data nemo_rl.data.datasets nemo_rl.data.hf_datasets nemo_rl.data.hf_datasets.chat_templates nemo_rl.data.hf_datasets.deepscaler nemo_rl.data.hf_datasets.dpo nemo_rl.data.hf_datasets.helpsteer3 nemo_rl.data.hf_datasets.oasst nemo_rl.data.hf_datasets.openmathinstruct2 nemo_rl.data.hf_datasets.prompt_response_dataset nemo_rl.data.hf_datasets.squad nemo_rl.data.interfaces nemo_rl.data.llm_message_utils nemo_rl.distributed nemo_rl.distributed.batched_data_dict nemo_rl.distributed.collectives nemo_rl.distributed.model_utils nemo_rl.distributed.virtual_cluster nemo_rl.distributed.worker_groups nemo_rl.environments nemo_rl.environments.interfaces nemo_rl.environments.math_environment nemo_rl.environments.metrics nemo_rl.environments.utils nemo_rl.evals nemo_rl.evals.eval nemo_rl.experience nemo_rl.experience.rollouts nemo_rl.metrics nemo_rl.metrics.metrics_utils nemo_rl.models nemo_rl.models.dtensor nemo_rl.models.dtensor.parallelize nemo_rl.models.generation nemo_rl.models.generation.interfaces nemo_rl.models.generation.vllm nemo_rl.models.generation.vllm_backend nemo_rl.models.huggingface nemo_rl.models.huggingface.common nemo_rl.models.interfaces nemo_rl.models.megatron nemo_rl.models.megatron.common nemo_rl.models.policy nemo_rl.models.policy.dtensor_policy_worker nemo_rl.models.policy.fsdp1_policy_worker nemo_rl.models.policy.hf_policy nemo_rl.models.policy.utils nemo_rl.package_info nemo_rl.utils nemo_rl.utils.checkpoint nemo_rl.utils.config nemo_rl.utils.logger nemo_rl.utils.native_checkpoint nemo_rl.utils.nvml nemo_rl.utils.timer nemo_rl.utils.venvs

monitor_gpus (nemo_rl.utils.logger.LoggerConfig attribute) move_buffer_to_device() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) move_to_cpu() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) move_to_cuda() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) move_to_device() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) MultiWorkerFuture (class in nemo_rl.distributed.worker_groups)

N

name (nemo_rl.models.policy.TokenizerConfig attribute) (nemo_rl.utils.logger.WandbConfig attribute) nemo_rl module nemo_rl.algorithms 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.sft module nemo_rl.algorithms.utils module nemo_rl.converters module nemo_rl.converters.huggingface module nemo_rl.converters.huggingface.vllm_export module nemo_rl.converters.megatron module nemo_rl.converters.megatron.vllm_export module nemo_rl.data module nemo_rl.data.datasets module nemo_rl.data.hf_datasets module nemo_rl.data.hf_datasets.chat_templates module nemo_rl.data.hf_datasets.deepscaler module nemo_rl.data.hf_datasets.dpo module nemo_rl.data.hf_datasets.helpsteer3 module nemo_rl.data.hf_datasets.oasst module nemo_rl.data.hf_datasets.openmathinstruct2 module nemo_rl.data.hf_datasets.prompt_response_dataset module nemo_rl.data.hf_datasets.squad module nemo_rl.data.interfaces module nemo_rl.data.llm_message_utils 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.virtual_cluster module nemo_rl.distributed.worker_groups module nemo_rl.environments module nemo_rl.environments.interfaces module nemo_rl.environments.math_environment module nemo_rl.environments.metrics module nemo_rl.environments.utils module

nemo_rl.evals module nemo_rl.evals.eval module nemo_rl.experience module nemo_rl.experience.rollouts module nemo_rl.metrics module nemo_rl.metrics.metrics_utils 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.interfaces module nemo_rl.models.generation.vllm module nemo_rl.models.generation.vllm_backend module nemo_rl.models.huggingface module nemo_rl.models.huggingface.common module nemo_rl.models.interfaces module nemo_rl.models.megatron module nemo_rl.models.megatron.common module nemo_rl.models.policy module nemo_rl.models.policy.dtensor_policy_worker module nemo_rl.models.policy.fsdp1_policy_worker module nemo_rl.models.policy.hf_policy module nemo_rl.models.policy.utils module nemo_rl.package_info module nemo_rl.utils module nemo_rl.utils.checkpoint module nemo_rl.utils.config module nemo_rl.utils.logger module nemo_rl.utils.native_checkpoint module nemo_rl.utils.nvml module nemo_rl.utils.timer module nemo_rl.utils.venvs module 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_rewards (nemo_rl.algorithms.grpo.GRPOConfig attribute) num_generations_per_prompt (nemo_rl.algorithms.grpo.GRPOConfig attribute) num_nodes (nemo_rl.distributed.virtual_cluster.ClusterConfig attribute) num_prompts_per_step (nemo_rl.algorithms.grpo.GRPOConfig attribute) num_tests_per_prompt (nemo_rl.evals.eval.EvalConfig attribute) num_workers (nemo_rl.environments.math_environment.MathEnvConfig attribute) nvml_context() (in module nemo_rl.utils.nvml)

O

OasstDataset (class in nemo_rl.data.hf_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.fsdp1_policy_worker.FSDP1PolicyWorker method) (nemo_rl.models.policy.hf_policy.HfPolicy method) offload_before_refit() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) (nemo_rl.models.policy.hf_policy.HfPolicy method)

OpenMathInstruct2Dataset (class in nemo_rl.data.hf_datasets.openmathinstruct2) OptimizerState (class in nemo_rl.utils.native_checkpoint) output_ids (nemo_rl.models.generation.interfaces.GenerationOutputSpec attribute) output_key (nemo_rl.data.DataConfig attribute) OverridesError

P

pad_token_id (nemo_rl.models.generation.interfaces.GenerationConfig attribute) PARALLIZE_FUNCTIONS (in module nemo_rl.models.dtensor.parallelize) parse_conversations() (in module nemo_rl.data.hf_datasets.oasst) parse_hydra_overrides() (in module nemo_rl.utils.config) passthrough_prompt_response (nemo_rl.data.hf_datasets.chat_templates.COMMON_CHAT_TEMPLATES attribute) PATCH (in module nemo_rl.package_info) policy (nemo_rl.algorithms.dpo.MasterConfig attribute) (nemo_rl.algorithms.grpo.MasterConfig attribute) (nemo_rl.algorithms.sft.MasterConfig attribute) PolicyConfig (class in nemo_rl.models.policy) PolicyInterface (class in nemo_rl.models.interfaces) PRE_RELEASE (in module nemo_rl.package_info) precision (nemo_rl.models.policy.PolicyConfig attribute) preference_average_log_probs (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.loss_functions.DPOLossConfig attribute) preference_loss() (nemo_rl.algorithms.loss_functions.DPOLossFn method) preference_loss_weight (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.loss_functions.DPOLossConfig attribute) prepare_deepscaler_dataset() (in module nemo_rl.data.hf_datasets.deepscaler) prepare_for_generation() (nemo_rl.models.generation.interfaces.GenerationInterface method) (nemo_rl.models.generation.vllm.VllmGeneration method) (nemo_rl.models.policy.hf_policy.HfPolicy method)

prepare_for_lp_inference() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) (nemo_rl.models.policy.hf_policy.HfPolicy method) prepare_for_training() (nemo_rl.models.interfaces.PolicyInterface method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) (nemo_rl.models.policy.hf_policy.HfPolicy method) prepare_openinstructmath2_dataset() (in module nemo_rl.data.hf_datasets.openmathinstruct2) prepare_weights_for_ipc() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) (nemo_rl.models.policy.hf_policy.HfPolicy method) prev_logprobs (nemo_rl.algorithms.loss_functions.ClippedPGLossDataDict attribute) print_all_worker_groups() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) print_cluster_grid() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) print_message_log_samples() (in module nemo_rl.utils.logger) print_worker_layout() (nemo_rl.distributed.worker_groups.RayWorkerGroup method) problem_key (nemo_rl.data.MathDataConfig attribute) project (nemo_rl.utils.logger.WandbConfig attribute) prompt_file (nemo_rl.data.DataConfig attribute) (nemo_rl.data.interfaces.TaskDataSpec attribute) PromptResponseDataset (class in nemo_rl.data.hf_datasets.prompt_response_dataset) PY_EXECUTABLES (class in nemo_rl.distributed.virtual_cluster)

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) RayWorkerGroup (class in nemo_rl.distributed.worker_groups) rebalance_nd_tensor() (in module nemo_rl.distributed.collectives) reduce() (nemo_rl.utils.timer.Timer method) 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) 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_old_checkpoints() (nemo_rl.utils.checkpoint.CheckpointManager method)

reorder_data() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) repeat_interleave() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) report_device_id() (nemo_rl.models.generation.vllm.VllmGenerationWorker method) (nemo_rl.models.generation.vllm_backend.VllmInternalWorkerExtension method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) reset() (nemo_rl.utils.timer.Timer method) reset_peak_memory_stats() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) resolve_path() (in module nemo_rl.utils.config) ResourceInsufficientError respect_tied_workers (nemo_rl.distributed.worker_groups.MultiWorkerFuture attribute) rewards (nemo_rl.environments.interfaces.EnvironmentReturn attribute) rl_collate_fn() (in module nemo_rl.data.datasets) RotaryEmbedParallel (class in nemo_rl.models.dtensor.parallelize) run_all_workers_multiple_data() (nemo_rl.distributed.worker_groups.RayWorkerGroup method) run_all_workers_single_data() (nemo_rl.distributed.worker_groups.RayWorkerGroup method) run_env_eval() (in module nemo_rl.evals.eval) run_multi_turn_rollout() (in module nemo_rl.experience.rollouts)

S

sample_mask (nemo_rl.algorithms.loss_functions.ClippedPGLossDataDict attribute) (nemo_rl.algorithms.loss_functions.DPOLossDataDict attribute) save_checkpoint() (in module nemo_rl.utils.native_checkpoint) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) (nemo_rl.models.policy.hf_policy.HfPolicy method) seed (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.sft.SFTConfig attribute) (nemo_rl.evals.eval.EvalConfig attribute) select_indices() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) sequence_length_round (nemo_rl.distributed.batched_data_dict.DynamicBatchingCfg attribute) (nemo_rl.models.policy.DynamicBatchingConfig attribute) SEQUENCE_LEVEL (nemo_rl.algorithms.loss_functions.LossType attribute) sequence_parallel (nemo_rl.models.policy.DTensorConfig attribute) set_seed() (in module nemo_rl.algorithms.utils) setup() (in module nemo_rl.algorithms.dpo) (in module nemo_rl.algorithms.grpo) (in module nemo_rl.algorithms.sft) (in module nemo_rl.evals.eval) 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_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) shard_by_batch_size() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) shutdown() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method) (nemo_rl.distributed.worker_groups.RayWorkerGroup method) (nemo_rl.environments.math_environment.MathEnvironment method) (nemo_rl.models.generation.vllm.VllmGeneration method) (nemo_rl.models.generation.vllm.VllmGenerationWorker method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) (nemo_rl.models.policy.hf_policy.HfPolicy method)

simple_role_header (nemo_rl.data.hf_datasets.chat_templates.COMMON_CHAT_TEMPLATES attribute) size (nemo_rl.distributed.batched_data_dict.BatchedDataDict property) SKIP_DTENSOR_TIED_WEIGHTS_CHECK (nemo_rl.models.huggingface.common.ModelFlag attribute) skip_tokenizer_init (nemo_rl.models.generation.vllm.VllmSpecificArgs attribute) sleep() (nemo_rl.models.generation.vllm.VllmGenerationWorker method) slice() (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) split_output_tensor() (nemo_rl.algorithms.loss_functions.DPOLossFn method) SquadDataset (class in nemo_rl.data.hf_datasets.squad) start() (nemo_rl.utils.logger.RayGpuMonitorLogger method) (nemo_rl.utils.timer.Timer method) 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.grpo.GRPOSaveState attribute) (nemo_rl.algorithms.sft.SFTSaveState attribute) (nemo_rl.utils.logger.GpuMetricSnapshot attribute) step() (nemo_rl.environments.interfaces.EnvironmentInterface method) (nemo_rl.environments.math_environment.MathEnvironment method) stop() (nemo_rl.utils.logger.RayGpuMonitorLogger method) (nemo_rl.utils.timer.Timer method) stop_strings (nemo_rl.data.interfaces.DatumSpec attribute) (nemo_rl.environments.math_environment.MathEnvConfig attribute) (nemo_rl.models.generation.interfaces.GenerationDatumSpec attribute) stop_token_ids (nemo_rl.models.generation.interfaces.GenerationConfig attribute) surpress_user_warnings() (in module nemo_rl.algorithms.utils) SYSTEM (nemo_rl.distributed.virtual_cluster.PY_EXECUTABLES attribute) SYSTEM_PROMPT (in module nemo_rl.data.hf_datasets.oasst) system_prompt_file (nemo_rl.data.DataConfig attribute) (nemo_rl.data.interfaces.TaskDataSpec attribute)

T

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) temperature (nemo_rl.models.generation.interfaces.GenerationConfig attribute) tensor_parallel_size (nemo_rl.models.generation.vllm.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) terminateds (nemo_rl.environments.interfaces.EnvironmentReturn attribute) time() (nemo_rl.utils.timer.Timer method) Timer (class in nemo_rl.utils.timer) to() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method) to_local_if_dtensor() (in module nemo_rl.models.dtensor.parallelize) TOKEN_LEVEL (nemo_rl.algorithms.loss_functions.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.DPOLossDataDict attribute) tokenizer (nemo_rl.models.policy.PolicyConfig attribute) TokenizerConfig (class in nemo_rl.models.policy) top_k (nemo_rl.models.generation.interfaces.GenerationConfig attribute) top_p (nemo_rl.models.generation.interfaces.GenerationConfig attribute) total_steps (nemo_rl.algorithms.dpo.DPOSaveState attribute) (nemo_rl.algorithms.sft.SFTSaveState attribute) train() (nemo_rl.models.interfaces.PolicyInterface method) (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) (nemo_rl.models.policy.hf_policy.HfPolicy method) train_global_batch_size (nemo_rl.models.policy.PolicyConfig attribute) train_mb_tokens (nemo_rl.models.policy.DynamicBatchingConfig attribute) train_micro_batch_size (nemo_rl.models.policy.PolicyConfig attribute) truncate_tensors() (nemo_rl.distributed.batched_data_dict.BatchedDataDict method)

U

unpadded_sequence_lengths (nemo_rl.models.generation.interfaces.GenerationOutputSpec attribute) unshard_fsdp2_model() (in module nemo_rl.models.policy.dtensor_policy_worker) update_weights() (nemo_rl.models.generation.vllm.VllmGeneration method) update_weights_from_ipc_handles() (nemo_rl.models.generation.vllm.VllmGenerationWorker method) (nemo_rl.models.generation.vllm_backend.VllmInternalWorkerExtension method)

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_reference_model() (nemo_rl.models.policy.dtensor_policy_worker.DTensorPolicyWorker method) (nemo_rl.models.policy.fsdp1_policy_worker.FSDP1PolicyWorker method) used_workers (nemo_rl.distributed.worker_groups.MultiWorkerFuture attribute)

V

val_at_start (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.grpo.GRPOConfig attribute) (nemo_rl.algorithms.sft.SFTConfig attribute) val_batch_size (nemo_rl.algorithms.grpo.GRPOConfig attribute) val_batches (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.sft.SFTConfig attribute) val_dataset_name (nemo_rl.data.DataConfig attribute) val_global_batch_size (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.sft.SFTConfig attribute) val_loss (nemo_rl.algorithms.dpo.DPOSaveState attribute) (nemo_rl.algorithms.sft.SFTSaveState attribute) val_micro_batch_size (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.sft.SFTConfig attribute) val_period (nemo_rl.algorithms.dpo.DPOConfig attribute) (nemo_rl.algorithms.grpo.GRPOConfig attribute) (nemo_rl.algorithms.sft.SFTConfig attribute)

val_reward (nemo_rl.algorithms.grpo.GRPOSaveState attribute) validate() (in module nemo_rl.algorithms.dpo) (in module nemo_rl.algorithms.grpo) (in module nemo_rl.algorithms.sft) verify() (nemo_rl.environments.math_environment.HFVerifyWorker 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.VllmConfig attribute) VLLM_LOAD_FORMAT_AUTO (nemo_rl.models.huggingface.common.ModelFlag attribute) VllmConfig (class in nemo_rl.models.generation.vllm) VllmGeneration (class in nemo_rl.models.generation.vllm) VllmGenerationWorker (class in nemo_rl.models.generation.vllm) VllmInternalWorkerExtension (class in nemo_rl.models.generation.vllm_backend) VllmSpecificArgs (class in nemo_rl.models.generation.vllm)

W

wake_up() (nemo_rl.models.generation.vllm.VllmGenerationWorker 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) worker_metadata (nemo_rl.distributed.worker_groups.RayWorkerGroup property) workers (nemo_rl.distributed.worker_groups.RayWorkerGroup property) world_size() (nemo_rl.distributed.virtual_cluster.RayVirtualCluster method)