MCPcopy Create free account

hub / github.com/Gen-Verse/OpenClaw-RL / functions

Functions10,561 in github.com/Gen-Verse/OpenClaw-RL

↓ 1 callersFunctionsample_to_datum_combined
Convert one sample into a Tinker Datum with combined OPD+RL advantages. combined_adv_i = w_opd * (teacher_lp_i - student_lp_i) + w_rl * reward
openclaw-tinker/data_formatter.py:149
↓ 1 callersFunctionsave
(rollout_id)
slime/train.py:49
↓ 1 callersFunctionsave
Persist a training checkpoint safely with forward hooks disabled. Args: iteration (int): Current global iteration number. model (
slime/slime/backends/megatron_utils/model.py:708
↓ 1 callersMethodsave
Save the state of the data source
slime/slime/rollout/data_source.py:31
↓ 1 callersMethodsaveImage
(self, imgarray, text)
gui-rl/desktop_env/server/pyxcursor.py:137
↓ 1 callersMethodsave_block
Save block of codes to disk. Args: config (RetroPreprocessingConfig): Retro preprocessing config. block (dict): Range
Megatron-LM/megatron/core/datasets/retro/index/indexes/faiss_par_add.py:95
↓ 1 callersFunctionsave_block_db
Save block of chunked tokens to disk. These blocks are later used for training and adding to the vector index. Args: block (dict): Ra
Megatron-LM/megatron/core/datasets/retro/db/build.py:265
↓ 1 callersFunctionsave_checkpoint
Required top-level function that creates the saver and calls its .save().
Megatron-LM/tools/checkpoint/saver_core.py:63
↓ 1 callersMethodsave_checkpoint
Initialize params, forward pass data, and save checkpoint.
Megatron-LM/tests/functional_tests/test_cases/common/ckpt_converter/__main__.py:404
↓ 1 callersMethodsave_common
Save common part of the state dict.
Megatron-LM/megatron/core/dist_checkpointing/strategies/base.py:182
↓ 1 callersFunctionsave_config
Save copy of config within retro project dir.
Megatron-LM/tools/retro/preprocess_data.py:216
↓ 1 callersFunctionsave_config
Save given config to checkpoint directory. Args: config: checkpoint config checkpoint_dir: checkpoint directory Returns:
Megatron-LM/megatron/core/dist_checkpointing/core.py:76
↓ 1 callersFunctionsave_distributed_checkpoint
Save model checkpoint using Megatron-Core distributed checkpointing. Args: checkpoint_path: Directory path to save checkpoint.
Megatron-LM/examples/run_simple_mcore_train_loop.py:154
↓ 1 callersFunctionsave_hf_model
Save Megatron model in HuggingFace format. Args: model (Sequence[DDP]): Sequence of DDP-wrapped model chunks. rollout_id (int): R
slime/slime/backends/megatron_utils/model.py:736
↓ 1 callersFunctionsave_indexed_dataset_infos
Save dataset order & meta-info. Args: project_dir (str): Path to Retro project dir. indexed_dataset_infos (List[Dict]): List of m
Megatron-LM/megatron/core/datasets/retro/db/utils.py:84
↓ 1 callersMethodsave_local_models_to_checkpoint
Save local models in self.models to a megatron checkpoint.
Megatron-LM/tools/checkpoint/saver_base.py:329
↓ 1 callersFunctionsave_lora_checkpoint
Save only the LoRA adapter weights + config to *path*. Only rank 0 writes to disk. All ranks participate in the state-dict gathering (handle
slime/slime/backends/fsdp_utils/lora_utils.py:76
↓ 1 callersMethodsave_loss_to_tracker
Save the indexer loss for logging. Args: loss: The loss tensor. layer_number: Layer index of the loss, 1-indexed.
Megatron-LM/megatron/core/transformer/experimental_attention_variant/dsa.py:55
↓ 1 callersMethodsave_parameter_state
Save the distributed parameter states of all optimizers to a file. Args: filename (str): path to save parameter state to.
Megatron-LM/megatron/core/optimizer/optimizer.py:1340
↓ 1 callersMethodsave_shard
Save the block data that was created this in this process
Megatron-LM/megatron/legacy/data/realm_index.py:75
↓ 1 callersMethodsave_sharded_objects
Save sharded objects from the state dict.
Megatron-LM/megatron/core/dist_checkpointing/strategies/base.py:186
↓ 1 callersMethodsave_state
Method to save the state of the VM.
gui-rl/desktop_env/providers/base.py:26
↓ 1 callersMethodsave_state
(self)
Megatron-LM/examples/multimodal/dataloader_provider.py:164
↓ 1 callersFunctionsave_state_dict_async_finalize
Finalization of save_state_dict_async_plan. The input arguments are the same as the save_state_dict_async_plan output, the `write_result
Megatron-LM/megatron/core/dist_checkpointing/strategies/state_dict_saver.py:213
↓ 1 callersFunctionsave_state_dict_async_plan
First stage of saving a state dict to storage. This is an async adjustment of torch.distributed.checkpoint.state_dict_saver. In order to
Megatron-LM/megatron/core/dist_checkpointing/strategies/state_dict_saver.py:41
↓ 1 callersMethodsave_state_dict_to_file
Save the parameter state of the optimizer. For torch format only. Args: filename: The filename to save the parameter state.
Megatron-LM/megatron/core/optimizer/layer_wise_optimizer.py:305
↓ 1 callersMethodsave_state_dict_to_hf_checkpoint
(self)
Megatron-LM/tools/checkpoint/saver_hf_llava.py:352
↓ 1 callersFunctionsave_tensors
(args, model_name, state_dict, output_dir, chunk_size, vocab_size=None)
slime/tools/convert_torch_dist_to_hf.py:106
↓ 1 callersMethodscale_gradients
Scale all gradients inside the buffers by `scaling_factor`.
Megatron-LM/megatron/core/distributed/data_parallel_base.py:45
↓ 1 callersFunctionscaled_init_method_normal
Init method based on N(0, sigma/sqrt(2*num_layers).
Megatron-LM/megatron/core/utils.py:743
↓ 1 callersMethodschedule_async_call
Schedule `async_req` with some process forking or reusing persistent worker This method must be called on all ranks. Args
Megatron-LM/megatron/core/dist_checkpointing/strategies/async_utils.py:120
↓ 1 callersFunctionschedule_async_save
Schedule the async save request. Args: async_request (AsyncRequest): the async save request.
Megatron-LM/megatron/training/async_utils.py:28
↓ 1 callersMethodschedule_chunked_prefill
This function schedules chunked prefill requests. Invariant: - There are at most one chunked prefill request in the waiti
Megatron-LM/megatron/core/inference/engines/dynamic_engine.py:933
↓ 1 callersMethodschedule_non_chunked_prefill
Perform the same original scheduling logic for non-chunked runs
Megatron-LM/megatron/core/inference/engines/dynamic_engine.py:913
↓ 1 callersMethodschedule_requests
Drains the ZMQ socket for a batch of requests and adds them to the engine. This method is a collective and synchronous operation that must be
Megatron-LM/megatron/core/inference/engines/dynamic_engine.py:1279
↓ 1 callersFunctionsearch
Perform search using either local search engine or Google search. The search backend is determined by SEARCH_R1_CONFIGS["search_backend"].
slime/examples/search-r1/generate_with_search.py:59
↓ 1 callersFunctionsearch_for_fit
Finds the index of largest number that fits into the knapsack with the given capacity.
Megatron-LM/examples/multimodal/dataset_helpers.py:95
↓ 1 callersMethodselect_model
(self)
swe-rl/mini-swe-agent/src/minisweagent/models/extra/roulette.py:33
↓ 1 callersFunctionselect_rollout_data
Select the most recent groups when there are too many samples. Groups all samples by instance_id, sorts groups by timestamp. Args:
slime/slime_plugins/rollout_buffer/rollout_buffer_example.py:22
↓ 1 callersMethodselect_worker
(self, task_key: str)
terminal-rl/router_server.py:66
↓ 1 callersMethodsend_backward
Send backward gradient tensor. Note: Gradient senders are activation 'RECEIVERS' Args: grad_tensor: The gradient tensor
Megatron-LM/megatron/core/pipeline_parallel/bridge_communicator.py:416
↓ 1 callersMethodsend_backward_recv_forward
Batched send and recv with previous rank in pipeline.
Megatron-LM/megatron/core/pipeline_parallel/p2p_communication.py:539
↓ 1 callersMethodsend_backward_recv_forward
Combined operation: send backward gradient and receive forward activation. Args: grad_tensor: The gradient tensor to send backwar
Megatron-LM/megatron/core/pipeline_parallel/bridge_communicator.py:663
↓ 1 callersMethodsend_data_to_buffer
(self, data)
slime/slime_plugins/rollout_buffer/generator/base_generator.py:177
↓ 1 callersMethodsend_forward_recv_backward
Batched send and recv with next rank in pipeline.
Megatron-LM/megatron/core/pipeline_parallel/p2p_communication.py:506
↓ 1 callersMethodsend_forward_recv_backward
Combined operation: send forward activation and receive backward gradient. Args: input_tensor: The tensor to send forward
Megatron-LM/megatron/core/pipeline_parallel/bridge_communicator.py:542
↓ 1 callersMethodsend_model_over_queue
Creates model schema and sends the model over the queue
Megatron-LM/tools/checkpoint/loader_base.py:492
↓ 1 callersMethodsend_slack_notification
Send Slack notification via webhook.
Megatron-LM/tests/test_utils/python_scripts/auto_reminder_github.py:278
↓ 1 callersFunctionsend_to_openclaw
Send a message to OpenClaw, maintaining session via the user field.
openclaw-test/teacher_chat.py:149
↓ 1 callersFunctionsend_to_openclaw
Send a message to OpenClaw, maintaining session via the user field.
openclaw-test/TA_chat.py:144
↓ 1 callersFunctionsend_to_openclaw
Send a message to OpenClaw, maintaining session via the user field.
openclaw-test/student_chat.py:114
↓ 1 callersMethodsend_vision_backbone_over_queue
Using self.all_models, extract model parameters and send them over the queue.
Megatron-LM/tools/checkpoint/loader_llava.py:161
↓ 1 callersMethodsend_vision_projection_over_queue
(self)
Megatron-LM/tools/checkpoint/loader_llava.py:288
↓ 1 callersMethodserialize
Serialize error. Args: error (ContextOverflowError): Error. Returns: (dict) Serialized error data.
Megatron-LM/megatron/core/inference/contexts/dynamic_context.py:140
↓ 1 callersFunctionserialize_tensor
Serialize tensor to bytes. Args: tensor (Tensor): Tensor. Returns: (bytes) Byte representation of tensor.
Megatron-LM/megatron/core/inference/inference_request.py:17
↓ 1 callersFunctionset_attn_state
Set self-attention params.
Megatron-LM/tools/checkpoint/loader_mixtral_hf.py:77
↓ 1 callersFunctionset_attn_state
Set self-attention params.
Megatron-LM/tools/checkpoint/loader_llama_mistral.py:328
↓ 1 callersFunctionset_build_dependency
(workload_manifests: List[dotdict])
Megatron-LM/tests/test_utils/python_scripts/recipe_parser.py:75
↓ 1 callersFunctionset_checkpoint_version
(value)
Megatron-LM/megatron/training/checkpointing.py:70
↓ 1 callersFunctionset_deepep_num_sms
Sets the number of SMs to use for DeepEP
Megatron-LM/megatron/core/transformer/moe/fused_a2a.py:258
↓ 1 callersMethodset_for_recompute_input_layernorm
Set the attention layer for recompute input_layernorm. Only needed for fp8.
Megatron-LM/megatron/core/transformer/attention.py:1038
↓ 1 callersMethodset_for_recompute_pre_mlp_layernorm
Set the MoE layer for recompute pre_mlp_layernorm. Only needed for fp8/fp4.
Megatron-LM/megatron/core/transformer/moe/moe_layer.py:335
↓ 1 callersFunctionset_ideal_affinity_for_current_gpu
Set CPU affinity for the current GPU to optimize host-device transfers.
Megatron-LM/megatron/core/pipeline_parallel/utils.py:83
↓ 1 callersMethodset_is_first_microbatch
Update the is_first_microbatch flag for weight caching. Args: is_first_microbatch (bool): Whether this is the first microbatch in
Megatron-LM/megatron/core/transformer/cuda_graphs.py:1136
↓ 1 callersFunctionset_jit_fusion_options
Set PyTorch JIT layer fusion options.
Megatron-LM/megatron/training/initialize.py:439
↓ 1 callersMethodset_last_layer
Mark whether the current forward chunk is processing the last layer.
Megatron-LM/megatron/core/pipeline_parallel/fine_grained_activation_offload.py:152
↓ 1 callersFunctionset_layer_state
Set transformer layer params.
Megatron-LM/tools/checkpoint/loader_mixtral_hf.py:118
↓ 1 callersFunctionset_layer_state
Set transformer layer params.
Megatron-LM/tools/checkpoint/loader_llama_mistral.py:372
↓ 1 callersMethodset_max_parse_errors
(self, max_parse_errors: int)
terminal-rl/agent/camel_agent.py:360
↓ 1 callersFunctionset_mlp_state
Set MLP params.
Megatron-LM/tools/checkpoint/loader_mixtral_hf.py:100
↓ 1 callersFunctionset_mlp_state
Set MLP params.
Megatron-LM/tools/checkpoint/loader_llama_mistral.py:359
↓ 1 callersFunctionset_postprocess_state
Set output layer & norm params.
Megatron-LM/tools/checkpoint/loader_mixtral_hf.py:72
↓ 1 callersFunctionset_postprocess_state
Set output layer & norm params.
Megatron-LM/tools/checkpoint/loader_llama_mistral.py:321
↓ 1 callersFunctionset_preprocess_state
Set embedding params.
Megatron-LM/tools/checkpoint/loader_mixtral_hf.py:67
↓ 1 callersFunctionset_preprocess_state
Set embedding params.
Megatron-LM/tools/checkpoint/loader_llama_mistral.py:315
↓ 1 callersMethodset_rollout_manager
(self, rollout_manager)
slime/slime/ray/train_actor.py:131
↓ 1 callersFunctionset_rounder
Utility function to set the DynamicInferenceContext rounder.
Megatron-LM/tests/unit_tests/inference/contexts/test_dynamic_context.py:23
↓ 1 callersFunctionset_routing_replay
(replay)
slime/slime/utils/routing_replay.py:8
↓ 1 callersMethodset_special_tokens
Add a list of additional tokens to the encoder. The additional tokens are indexed starting from the last index of the current
Megatron-LM/megatron/training/tokenizer/gpt2_tokenization.py:184
↓ 1 callersMethodset_up_planner
( self, state_dict, metadata, is_coordinator: bool = False )
Megatron-LM/tools/checkpoint/checkpoint_inspector.py:319
↓ 1 callersFunctionsetup
Initialize fault tolerance Args: args (argparse.Namespace): parsed Megatron-LM command line arguments Raises: ValueError: if
Megatron-LM/megatron/training/ft_integration.py:75
↓ 1 callersMethodsetup
(self)
openclaw-tinker/trainer.py:53
↓ 1 callersFunctionsetup_logging
Sets the default logging level based on cmdline args and env vars. Precedence: 1. Command line argument `--logging-level` 2. Env var `MEG
Megatron-LM/megatron/training/initialize.py:526
↓ 1 callersMethodsetup_manual_hooks
Set CUDA Graph manual hooks for the submodules that contain direct parameters and are covered by cudagraphs.
Megatron-LM/megatron/core/transformer/module.py:216
↓ 1 callersMethodsetup_metadata
(self, routing_map: torch.Tensor, probs: torch.Tensor)
Megatron-LM/megatron/core/transformer/moe/token_dispatcher.py:1158
↓ 1 callersMethodsetup_method
setup env
Megatron-LM/tests/unit_tests/models/test_mimo_audio_submodules.py:118
↓ 1 callersMethodsetup_method
(self, method, use_cpu_initialization=False, swiglu=True)
Megatron-LM/tests/unit_tests/transformer/moe/test_grouped_mlp.py:228
↓ 1 callersMethodsetup_method
(self, method)
Megatron-LM/tests/unit_tests/transformer/moe/test_sequential_mlp.py:22
↓ 1 callersMethodsetup_model
(self, tensor_parallel_size, pipeline_parallel_size)
Megatron-LM/tests/unit_tests/inference/model_inference_wrappers/t5/test_t5_inference_wrapper.py:29
↓ 1 callersFunctionsetup_model_and_optimizer
Build model(s), wrap with DDP, and construct optimizer and scheduler. Args: args (Namespace): Training/runtime arguments (argparse namesp
slime/slime/backends/megatron_utils/model.py:84
↓ 1 callersMethodsetup_process_groups_for_ddp
Helper method to set up process groups for DDP with proper validation and fallbacks. Args: pg_collection: Optional proce
Megatron-LM/megatron/core/process_groups_config.py:444
↓ 1 callersMethodsetup_process_groups_for_optimizer
Helper method to set up process groups for optimizer and DDP with proper validation and fallbacks. Args: pg_coll
Megatron-LM/megatron/core/process_groups_config.py:253
↓ 1 callersFunctionsetup_seed
(seed)
Megatron-LM/tests/unit_tests/test_optimizer_cpu_offloading.py:42
↓ 1 callersFunctionsetup_seed
(seed)
Megatron-LM/tests/unit_tests/distributed/test_mcore_fully_sharded_data_parallel.py:62
↓ 1 callersMethodshard_params
Shard all params into lists by rank.
Megatron-LM/megatron/core/optimizer/layer_wise_optimizer.py:92
↓ 1 callersMethodsharded_param_state_dp_reshardable
Sharded state dict where each noncontiguous buffer is a separate ShardedTensor. Results in fully parallel save and load without any inter-pro
Megatron-LM/megatron/core/optimizer/distrib_optimizer.py:1526
↓ 1 callersMethodsharded_param_state_dp_zero
Naive implementation which reuses gather/scatter from the legacy ckpt format. During saving, gathers the parameters state on DP rank 0 and sa
Megatron-LM/megatron/core/optimizer/distrib_optimizer.py:1358
↓ 1 callersMethodsharded_param_state_fs_model_space
Sharded state dict where each buffer is mapped to corresponding model param. In this approach the optimizer state tensors are directly relate
Megatron-LM/megatron/core/optimizer/distrib_optimizer.py:1673
↓ 1 callersMethodsharded_param_state_fsdp_dtensor
Sharded state dict where each parameter is a separate PyTorch DTensor.
Megatron-LM/megatron/core/optimizer/distrib_optimizer.py:1334
↓ 1 callersMethodsharded_param_state_fully_reshardable
Exchange based format in model space representation. `fully_reshardable` format involves gathering the tensors on DP rank 0 during save.
Megatron-LM/megatron/core/optimizer/distrib_optimizer.py:1391
← previousnext →5,001–5,100 of 10,561, ranked by callers