MCPcopy Create free account

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

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

↓ 2 callersMethodrecord_fwd_graph
Record a fwd graph to 'cudagraph_record
Megatron-LM/megatron/core/transformer/cuda_graphs.py:194
↓ 2 callersMethodrecord_policy_turn
(self, *, action_text: str, response: str, screenshot_bytes: bytes)
gui-rl/agents/qwen35_agent.py:461
↓ 2 callersMethodrecord_tool_result
( self, turn_idx: int, tool_call_request: Any, raw_result: Any )
terminal-rl/rollout_agent.py:43
↓ 2 callersFunctionrecover_qkv
(new_tensor, num_head, head_dim)
Megatron-LM/tools/checkpoint/saver_hf_llava.py:16
↓ 2 callersMethodrecv_backward
Receive tensor from next rank in pipeline (backward receive).
Megatron-LM/megatron/core/pipeline_parallel/p2p_communication.py:434
↓ 2 callersMethodrecv_forward
Receive forward activation tensor. Args: tensor_shape: Expected tensor shape (None if using shape communication) Returns
Megatron-LM/megatron/core/pipeline_parallel/bridge_communicator.py:320
↓ 2 callersMethodrecycle_unused_buckets
Recycle the unused buckets.
Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/param_and_grad_buffer.py:3570
↓ 2 callersMethodreduce_gradients
Reduce the gradients for the given parameters. Args: params (List[torch.Tensor]): The parameters. suggested_queue_capa
Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/param_and_grad_buffer.py:2949
↓ 2 callersMethodreduce_scatter
(self, *a, **kw)
slime/slime/utils/reloadable_process_group.py:217
↓ 2 callersMethodreformat_datum
(self, datum: dict)
Megatron-LM/examples/rl/environments/math/dapo_agent.py:18
↓ 2 callersMethodreformat_datum
(self, datum: dict)
Megatron-LM/examples/rl/environments/math/gsm8k_agent.py:30
↓ 2 callersFunctionreformat_prompt
(query, neighbours, dataset_name, ft_neighbours, \ max_output_len, tokenizer, max_seq_leng
Megatron-LM/tools/retro/sft/dataset_conv.py:331
↓ 2 callersFunctionregister
(api: OpenClawPluginApi)
extensions/rl-training-headers/index.ts:20
↓ 2 callersMethodrelease_bucket
Release the bucket.
Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/param_and_grad_buffer.py:3550
↓ 2 callersMethodreload
Reload.
Megatron-LM/megatron/core/pipeline_parallel/fine_grained_activation_offload.py:236
↓ 2 callersFunctionremove_boxed
Remove the LaTeX boxed command from a string. Args: s: String with format "\\boxed{content}" Returns: The content inside the
slime/slime/rollout/rm_hub/math_dapo_utils.py:50
↓ 2 callersFunctionreorder_inner_param_groups
(optimizer_state_dict)
Megatron-LM/megatron/training/training.py:595
↓ 2 callersMethodreplay_graph_capture
Replay the fwd cuda graph with autograd.
Megatron-LM/megatron/core/transformer/cuda_graphs.py:899
↓ 2 callersFunctionreport_theoretical_memory
(args, num_microbatches=None, verbose=False)
Megatron-LM/megatron/training/theoretical_memory_usage.py:337
↓ 2 callersMethodrequires_global_app_metadata
Whether a given strict option requires global metadata for validation.
Megatron-LM/megatron/core/dist_checkpointing/validation.py:95
↓ 2 callersMethodreset
( self, run_lease_id: str, task_meta: dict[str, Any], run_ctx_payload: dict[st
terminal-rl/remote/pool_server.py:232
↓ 2 callersMethodreset
(self, lease_id: str, task_config: dict[str, Any] | None = None)
gui-rl/env_pool_server.py:215
↓ 2 callersMethodreset
Zero out the underlying grad_buffer.
Megatron-LM/megatron/core/distributed/param_and_grad_buffer.py:927
↓ 2 callersMethodreset
Reset manager state for a new training iteration.
Megatron-LM/megatron/core/pipeline_parallel/fine_grained_activation_offload.py:63
↓ 2 callersMethodreset
(self)
slime/slime/rollout/sglang_rollout.py:81
↓ 2 callersMethodreset_global_aux_loss_tracker
Reset the global aux loss tracker.
Megatron-LM/megatron/core/transformer/moe/router.py:548
↓ 2 callersFunctionreset_model_temporary_tensors
Reset the temporary tensors of the model.
Megatron-LM/megatron/core/distributed/finalize_model_grads.py:278
↓ 2 callersMethodreset_parameters
Reset the router parameters.
Megatron-LM/megatron/core/transformer/moe/router.py:66
↓ 2 callersFunctionresolve_client_template_kwargs
(args: Any)
terminal-rl/rollout_agent.py:48
↓ 2 callersFunctionrestore_nd_flattened_tensors_formulation
Restores the original state dict from a reformulated form. Inverse of `apply_nd_flattened_tensors_reformulation`. Args: state_dict (
Megatron-LM/megatron/core/dist_checkpointing/strategies/resharding.py:166
↓ 2 callersMethodresume
Resume health checking. Called when engines are onloaded.
slime/slime/utils/health_monitor.py:93
↓ 2 callersMethodresume_engines
Sends a signal to unpause all inference engines.
Megatron-LM/megatron/core/inference/inference_client.py:220
↓ 2 callersFunctionretro_generate_and_post_process
Run inference and post-process outputs, i.e., detokenize, move to cpu and convert to list.
Megatron-LM/tools/retro/text_generation/retro_api.py:96
↓ 2 callersMethodrewind
Method to rewind the data iterator to the first microbatch of the iteration.
Megatron-LM/megatron/core/rerun_state_machine.py:1067
↓ 2 callersMethodrollout_from_response
( self, request: RolloutRequest, response: InferenceResponse, golden: Any )
Megatron-LM/megatron/rl/agent/reward_only_agent.py:86
↓ 2 callersFunctionround_by_factor
返回最接近 number 的且能被 factor 整除的整数
gui-rl/agents/utils/qwen_vl_utils.py:4
↓ 2 callersMethodround_up_requests
Round up to nearest multiple of `REQUEST_ROUNDER` (above) that is also divisible by tensor model parallel size.
Megatron-LM/megatron/core/inference/contexts/dynamic_context.py:698
↓ 2 callersMethodround_up_tokens
Round up to nearest multiple of `TOKEN_ROUNDER` (above) that is also divisible by tensor model parallel size.
Megatron-LM/megatron/core/inference/contexts/dynamic_context.py:644
↓ 2 callersMethodroute
Compute token routing for preprocessing. This method uses the router to determine which experts to send each token to, producing rout
Megatron-LM/megatron/core/transformer/moe/moe_layer.py:181
↓ 2 callersMethodrouted_experts_compute
Computes the output of the routed experts on the dispatched tokens. This method first post-processes the dispatched input to get permuted tok
Megatron-LM/megatron/core/transformer/moe/moe_layer.py:242
↓ 2 callersMethodrow_parallel_linear
Which row parallel linear module TE backend uses
Megatron-LM/megatron/core/models/backends.py:148
↓ 2 callersMethodrun
Run the server.
Megatron-LM/megatron/core/inference/text_generation_server/text_generation_server.py:209
↓ 2 callersMethodrun
(self, input_file, rollout_func, reward_func)
slime/slime_plugins/rollout_buffer/generator/base_generator.py:192
↓ 2 callersFunctionrunSession
(sessionId: string, trigger: string)
extensions/rl-training-headers/index.test.ts:43
↓ 2 callersFunctionrun_evaluation_loop
Common evaluation loop used by both online evaluation during training and standalone evaluation. Args: model: The model to evaluate
Megatron-LM/examples/multimodal/run_text_generation.py:804
↓ 2 callersMethodrun_identity
(self)
terminal-rl/custom_types.py:16
↓ 2 callersFunctionrun_model
(model, input, optim, fp8_recipe, fp8_recipe_settings)
Megatron-LM/tests/unit_tests/test_optimizer.py:230
↓ 2 callersMethodrun_self_attention
(self, pg_collection)
Megatron-LM/tests/unit_tests/transformer/test_attention.py:422
↓ 2 callersFunctionrun_tokenizer_tests
(tok)
Megatron-LM/tests/unit_tests/test_tokenizer.py:138
↓ 2 callersFunctionsafe_open_image_with_retry
Safely open an image file with retry mechanism for handling truncated files
gui-rl/desktop_env/evaluators/metrics/gimp.py:601
↓ 2 callersFunctionsafely_set_viewless_tensor_data
Safely set tensor's '.data' field. Check first that the tensor is viewless (i.e., '._base' not set). If not, raise an exception.
Megatron-LM/megatron/core/utils.py:724
↓ 2 callersFunctionsample_blocks
Sample a random subset of all blocks. Args: _blocks (List[Optional[Dict]]): List of all blocks. Returns:
Megatron-LM/megatron/core/datasets/retro/utils.py:301
↓ 2 callersMethodsave
(self, sharded_state_dict: ShardedStateDict, checkpoint_dir: Path)
Megatron-LM/megatron/core/dist_checkpointing/strategies/fully_parallel.py:96
↓ 2 callersFunctionsave_checkpoint_with_pickle_protocol
(state_dict, output_dir, pickle_protocol=4)
Megatron-LM/tools/checkpoint/checkpoint_inspector.py:289
↓ 2 callersFunctionsave_preprocess
Preprocesses the given state dictionary by applying factories, discarding non-persistent data and extracting the common state dictionary. Opti
Megatron-LM/megatron/core/dist_checkpointing/state_dict_utils.py:20
↓ 2 callersFunctionsave_to_aux_losses_tracker
Save the auxiliary loss for logging. Args: name (str): The name of the loss. loss (torch.Tensor): The loss tensor. layer_n
Megatron-LM/megatron/core/transformer/moe/moe_utils.py:726
↓ 2 callersMethodscaled_orthogonalize_fn
( grad: torch.Tensor, tp_group: torch.distributed.ProcessGroup, partition_
Megatron-LM/megatron/core/optimizer/muon.py:66
↓ 2 callersFunctionschedule_instance_termination
(region: str, instance_id: str, ttl_seconds: int, role_arn: str, logger)
gui-rl/desktop_env/providers/aws/scheduler_utils.py:107
↓ 2 callersMethodschedule_waiting_requests
Tries to schedule any requests in the waiting pool.
Megatron-LM/megatron/core/inference/engines/dynamic_engine.py:906
↓ 2 callersFunctionselect_best_hint
Select the longest positive hint from voting results.
openclaw-tinker/scorers.py:155
↓ 2 callersFunctionselective_log_softmax
Taken from: https://github.com/huggingface/trl/blob/26d86757a7c7e24e397ea44f57ecce6031dfac01/trl/trainer/utils.py#L1659. A memory-efficient imple
Megatron-LM/megatron/rl/rl_utils.py:664
↓ 2 callersMethodsend_backward
Send tensor to previous rank in pipeline (backward send).
Megatron-LM/megatron/core/pipeline_parallel/p2p_communication.py:486
↓ 2 callersFunctionsend_do_generate
Broadcasts a message to perform a generation to all tensor parallel ranks.
Megatron-LM/megatron/core/inference/text_generation_server/endpoints/common.py:11
↓ 2 callersMethodsend_forward
Send tensor to next rank in pipeline (forward send).
Megatron-LM/megatron/core/pipeline_parallel/p2p_communication.py:465
↓ 2 callersMethodsend_forward
Send forward activation tensor. Args: tensor_to_send: The tensor to send to the destination grid
Megatron-LM/megatron/core/pipeline_parallel/bridge_communicator.py:292
↓ 2 callersMethodsend_forward_backward_recv_forward_backward
Batched send and recv with previous and next ranks in pipeline.
Megatron-LM/megatron/core/pipeline_parallel/p2p_communication.py:624
↓ 2 callersMethodsend_llm_over_queue
Using self.all_models, extract model parameters and send them over the queue.
Megatron-LM/tools/checkpoint/loader_base.py:257
↓ 2 callersMethodsend_metadata_over_queue
(self)
Megatron-LM/tools/checkpoint/loader_base.py:246
↓ 2 callersFunctionsend_to_slack
Send message to Slack using webhook.
Megatron-LM/tests/test_utils/python_scripts/auto_reminder.py:242
↓ 2 callersFunctionset_bias_value
(dense_model)
Megatron-LM/tests/unit_tests/transformer/moe/test_upcycling.py:152
↓ 2 callersFunctionset_default_megatron_args
(args)
slime/slime/backends/megatron_utils/arguments.py:28
↓ 2 callersMethodset_document_indices
Set the document indices This method is slated for deprecation. Args: document_indices (numpy.ndarray): The document ind
Megatron-LM/megatron/core/datasets/indexed_dataset.py:791
↓ 2 callersMethodset_epoch
Change the seed offset so each epoch produces different randomization. Args: epoch: The epoch number to use as the seed
Megatron-LM/megatron/training/datasets/data_samplers.py:233
↓ 2 callersFunctionset_experimental_flag
Set the experimental flag to the given value.
Megatron-LM/megatron/core/config.py:6
↓ 2 callersFunctionset_input_tensor
(input_tensor)
Megatron-LM/tests/unit_tests/pipeline_parallel/test_schedules.py:152
↓ 2 callersMethodset_input_tensor
See megatron.model.transformer.set_input_tensor()
Megatron-LM/megatron/core/models/T5/t5_model.py:437
↓ 2 callersMethodset_prm_teacher_log_probs
Broadcast PRM teacher log-probs (computed on a separate GPU) to all actor ranks.
slime/slime/ray/actor_group.py:147
↓ 2 callersMethodset_shared_experts
Set shared expert to the dispatcher.
Megatron-LM/megatron/core/transformer/moe/token_dispatcher.py:204
↓ 2 callersMethodset_states
Set the rng states. For efficiency purposes, we do not check the size of seed for compatibility.
Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/utils.py:309
↓ 2 callersMethodset_states
Set the internal RNG state.
Megatron-LM/megatron/core/extensions/transformer_engine.py:2012
↓ 2 callersFunctionset_tp_pp_vpp
(tp, pp, vpp=None, pp_layout=None, destroy_first=True)
Megatron-LM/tests/unit_tests/pipeline_parallel/test_pipeline_layout.py:222
↓ 2 callersFunctionset_upcycling_args
(ep, granularity, num_experts=8)
Megatron-LM/tests/unit_tests/transformer/moe/test_upcycling.py:140
↓ 2 callersFunctionsetup
(app)
slime/docs/conf.py:192
↓ 2 callersMethodsetup
Args: config (List[Dict[str, Any]]): list of dict like {str: Any}. each config dict has the structure like
gui-rl/desktop_env/controllers/setup.py:58
↓ 2 callersMethodsetup
Setup the NVML Handler.
Megatron-LM/megatron/core/energy_monitor.py:37
↓ 2 callersMethodsharded_state_dict
Generate a sharded state dictionary for the multi token prediction layer. Args: prefix (str, optional): Prefix to be add
Megatron-LM/megatron/core/transformer/multi_token_prediction.py:996
↓ 2 callersMethodsharded_state_dict
( self, model_sharded_state_dict: ShardedStateDict, is_loading: bool = False,
Megatron-LM/megatron/core/optimizer/optimizer.py:1012
↓ 2 callersMethodshared_embedding_or_output_weight
This is a convenience method to surface the language model's word embeddings, which is necessary for `finalize_model_grads._allreduce_word_emb
Megatron-LM/megatron/core/models/multimodal/llava_model.py:382
↓ 2 callersMethodshared_experts_compute
Computes the output of the shared experts. If a shared expert is configured and not overlapped with communication, it is computed her
Megatron-LM/megatron/core/transformer/moe/moe_layer.py:215
↓ 2 callersFunctionshould_disable_forward_pre_hook
Block forward pre-hook for certain configurations.
Megatron-LM/tests/unit_tests/test_fp8_param.py:59
↓ 2 callersMethodshould_run_forward_backward
Method instructing whether to (re)run the forward-backward pass. Args: data_iterator: data iterator or list of data iterators use
Megatron-LM/megatron/core/rerun_state_machine.py:252
↓ 2 callersFunctionslice_cp_and_concat
( values: list[torch.Tensor], total_lengths: list[int], response_lengths: list[int] )
slime/examples/train_infer_mismatch_helper/mis.py:360
↓ 2 callersMethodsome_method
(self)
Megatron-LM/tests/unit_tests/test_utils.py:31
↓ 2 callersMethodsome_static_method
(cls)
Megatron-LM/tests/unit_tests/test_utils.py:35
↓ 2 callersFunctionsort_chunks_by_idxs
Split and sort the input tensor based on the split_sizes and sorted indices.
Megatron-LM/megatron/core/transformer/moe/moe_utils.py:404
↓ 2 callersFunctionspdocvqa_eval
Run SPDocVQA evaluation.
Megatron-LM/examples/multimodal/evaluation/evaluate_spdocvqa.py:35
↓ 2 callersFunctionsplit_dtensor
Splits a DTensor into smaller DTensors along a specified dimension. This function manages uneven sharding by accurately assigning chunk meta
Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/uneven_dtensor.py:396
↓ 2 callersFunctionsplit_swiglu_linear_fc1
Split the SWiGLU linear_fc1 parameter into two parts: weight_w and weight_v.
Megatron-LM/megatron/core/transformer/fsdp_dtensor_checkpoint.py:234
↓ 2 callersFunctionsplit_tuple
Split the elements in a tuple/interval, while handling well-formatted commas in large numbers
slime/slime/rollout/rm_hub/math_utils.py:360
↓ 2 callersFunctionsquared_relu
Squared ReLU activation
Megatron-LM/megatron/core/activations.py:9
← previousnext →2,301–2,400 of 10,561, ranked by callers