Code
Hub
Workspaces
Following
Trending
Connect
MCP
copy
Create free account
hub
/
github.com/Gen-Verse/OpenClaw-RL
/ functions
Functions
10,561 in github.com/Gen-Verse/OpenClaw-RL
⨍
Functions
10,561
◇
Types & classes
1,405
↳
Endpoints
144
↓ 1 callers
Method
train_critic
(self, rollout_id: int, rollout_data: RolloutBatch)
slime/slime/backends/megatron_utils/actor.py:1138
↓ 1 callers
Function
train_on_embeddings
Train index on embedded DB chunks. Args: config (RetroPreprocessingConfig): Retro preprocessing config.
Megatron-LM/megatron/core/datasets/retro/index/build.py:185
↓ 1 callers
Function
train_one_step
Execute a single pipeline-parallel training step. Runs forward/backward over ``num_microbatches``, applies optimizer step and one scheduler s
slime/slime/backends/megatron_utils/model.py:299
↓ 1 callers
Function
train_step
Run one training step: forward_backward_custom + optim_step.
openclaw-fireworks/fireworks_training_loop.py:101
↓ 1 callers
Function
transformer_flops
Calculate FLOPs for a standard Transformer model.
Megatron-LM/megatron/training/training.py:259
↓ 1 callers
Function
triton_append_key_value_cache
Append to KV cache using a high-performance, standalone Triton kernel. Args: layer_number (int): Layer number (1-based). key
Megatron-LM/megatron/core/inference/contexts/fused_kv_append_kernel.py:80
↓ 1 callers
Function
truncate
( weights: torch.Tensor, loss_mask: torch.Tensor, metrics: dict[str, list[torch.Tensor]], uppe
slime/examples/train_infer_mismatch_helper/mis.py:103
↓ 1 callers
Method
uncompress_kv_from_cache
Take a compressed kv and uncompress them
Megatron-LM/megatron/core/transformer/multi_latent_attention.py:894
↓ 1 callers
Function
unfused_dsa_fn
Unfused sparse attention implementation.
Megatron-LM/megatron/core/transformer/experimental_attention_variant/dsa.py:642
↓ 1 callers
Method
unpack_batch
Unpacks the packed samples into a list of sub-samples. Since each sub-sample may be routed to different DPxCP ranks, we unpac
Megatron-LM/megatron/core/datasets/data_schedule.py:245
↓ 1 callers
Method
unpad_input_prompt_tokens
Truncates the given input tensor back to the original prompt size before padding. Args: padded_batch_prompt_tokens (torch.Tensor)
Megatron-LM/megatron/core/inference/text_generation_controllers/text_generation_controller.py:473
↓ 1 callers
Function
unshard_row_parallel_state
(saved_state, out_features, in_features, tp)
Megatron-LM/tests/functional_tests/python_test_utils/test_optimizer_grads_match.py:144
↓ 1 callers
Function
upcycle_state_dict
Convert a dense model's state_dict to a MoE model's state_dict. This function facilitates the conversion of the state_dict from a dense mode
Megatron-LM/megatron/core/transformer/moe/upcycling_utils.py:301
↓ 1 callers
Method
update
Update number of microbatches depending on batch size rampup.
Megatron-LM/megatron/core/num_microbatches_calculator.py:293
↓ 1 callers
Method
update
Update number of microbatches. Args: consumed_samples (int): Number of samples consumed. consistency_check (bool): Op
Megatron-LM/megatron/core/num_microbatches_calculator.py:441
↓ 1 callers
Method
update_bucket_weights
(self, named_tensors, weight_version=None)
slime/slime/backends/fsdp_utils/update_weight_utils.py:84
↓ 1 callers
Method
update_center
Update center used for teacher output.
Megatron-LM/megatron/legacy/model/vision/dino.py:73
↓ 1 callers
Function
update_chunk_counts
Set n_chunks_train & n_chunks sampled for each individual DB. Args: config (RetroPreprocessingConfig): Retro preprocessing config.
Megatron-LM/megatron/core/datasets/retro/db/build.py:409
↓ 1 callers
Method
update_cudagraph_mapping
Updates the dedicated CUDA graph mapping tensor with the indices of currently active requests. Args: active_mamb
Megatron-LM/megatron/core/inference/contexts/attention_context/mamba_metadata.py:73
↓ 1 callers
Method
update_exit_status_table
(self)
swe-rl/mini-swe-agent/src/minisweagent/run/extra/utils/batch_progress.py:99
↓ 1 callers
Method
update_fp32_param_by_new_param
Update the fp32 parameters by the new parameters.
Megatron-LM/megatron/core/optimizer/cpu_offloading/hybrid_optimizer.py:376
↓ 1 callers
Method
update_from_meta_info
Update the sample with new information from meta_info returned by the rollout engine. And extract
slime/slime/utils/types.py:149
↓ 1 callers
Method
update_generation_status
Checks which prompts have reached an end condition We check which prompts have reached an end condition and set the corresponding fla
Megatron-LM/megatron/core/inference/text_generation_controllers/text_generation_controller.py:391
↓ 1 callers
Function
update_image_size_
根据 min_tokens, max_tokens 更新 image_ele 的尺寸信息 Args: image_ele (dict): - image_ele["image"]: str 图片路径 - image_ele["
gui-rl/agents/utils/qwen_vl_utils.py:48
↓ 1 callers
Method
update_metadata
(self, metadata: dict)
slime/slime/rollout/data_source.py:203
↓ 1 callers
Method
update_momentum
(self, iteration)
Megatron-LM/megatron/legacy/model/vision/dino.py:286
↓ 1 callers
Function
update_pg_timeout
Update the timeout for all process groups or a specific process group. Synchronize the process groups before updating the timeout. Args:
Megatron-LM/megatron/core/parallel_state.py:180
↓ 1 callers
Method
update_sampling_client
(self, client)
openclaw-tinker/rollout.py:89
↓ 1 callers
Function
update_trajectory
()
swe-rl/mini-swe-agent/tests/test_data/update_outputs_github_issue.py:12
↓ 1 callers
Function
update_trajectory
()
swe-rl/mini-swe-agent/tests/test_data/update_outputs_local.py:12
↓ 1 callers
Function
upload_statistics
(pipeline_id: int)
Megatron-LM/tests/test_utils/python_scripts/dashboard.py:182
↓ 1 callers
Function
validate
Validation method for validating saved neighbor IDs. Args: f (h5py.File): File containing save neighbor IDs.
Megatron-LM/megatron/core/datasets/retro/query/query.py:345
↓ 1 callers
Method
validate
(f)
Megatron-LM/tools/bert_embedding/embed.py:266
↓ 1 callers
Function
validate_args
(args)
slime/slime/backends/sglang_utils/arguments.py:115
↓ 1 callers
Function
validate_cuda_graph_request_counts
Validate `cuda_graph_request_count` usage across tests. For each test (i.e., each `num_cuda_graphs`), we validate how many times each cuda gr
Megatron-LM/tests/functional_tests/test_cases/gpt/gpt_dynamic_inference_tp1_pp1_583m_cuda_graphs_validation/cuda_graphs.py:115
↓ 1 callers
Function
validate_equation
Validate that equation only uses available numbers and each number once.
Megatron-LM/examples/rl/environments/countdown/countdown.py:39
↓ 1 callers
Function
validate_latency_proxies
Validate that the latency 'proxy' decreases as we increase the number of cuda graphs. Latency proxy is computed as the sum of `cuda_graph_request
Megatron-LM/tests/functional_tests/test_cases/gpt/gpt_dynamic_inference_tp1_pp1_583m_cuda_graphs_validation/cuda_graphs.py:189
↓ 1 callers
Method
validate_layer_layout
Check whether the layout is valid.
Megatron-LM/megatron/core/transformer/pipeline_parallel_layer_layout.py:85
↓ 1 callers
Function
validate_logprobs
Validate the logprob tensors. The logprobs should remain bitwise equal whether cuda graphs are enabled or not. Args: result_map (dic
Megatron-LM/tests/functional_tests/test_cases/gpt/gpt_dynamic_inference_tp1_pp1_583m_cuda_graphs_validation/cuda_graphs.py:206
↓ 1 callers
Method
validate_metadata_integrity
Codifies the constraints on metadata attributes. Meeting those constraints is guaranteed when instantiating a ShardedTensor class wit
Megatron-LM/megatron/core/dist_checkpointing/mapping.py:97
↓ 1 callers
Method
validate_metadata_integrity
(self)
Megatron-LM/megatron/core/dist_checkpointing/mapping.py:566
↓ 1 callers
Function
validate_model_config_args_from_heterogeneous_config
Validate model config arguments from heterogeneous config. This function takes model arguments and validates them based on a heterogeneous layer
Megatron-LM/megatron/training/arguments.py:131
↓ 1 callers
Function
validate_results
Validate test results, by running various checks.
Megatron-LM/tests/functional_tests/test_cases/gpt/gpt_dynamic_inference_tp1_pp1_583m_cuda_graphs_validation/cuda_graphs.py:219
↓ 1 callers
Method
validate_state_dict
Method that validate a checkpoint state dict before loading it. Args: state_dict: the state dict saved in the checkpoint and orig
Megatron-LM/megatron/core/rerun_state_machine.py:797
↓ 1 callers
Function
validate_step_counts
Validate engine step counts. This value should be identical across all tests, regardless of whether cuda graphs are enabled or not. Args
Megatron-LM/tests/functional_tests/test_cases/gpt/gpt_dynamic_inference_tp1_pp1_583m_cuda_graphs_validation/cuda_graphs.py:159
↓ 1 callers
Function
validate_uneven_dtensor
Validates the chunk metadata of an uneven DTensor to ensure correctness and boundary coverage. Notes: - `gather_and_compute_chunk_metada
Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/uneven_dtensor.py:140
↓ 1 callers
Function
validate_yaml
(args, defaults={})
Megatron-LM/megatron/training/yaml_arguments.py:41
↓ 1 callers
Function
verify_and_get_config_attr_descs
(config_cls, strict_docstring_match=True)
Megatron-LM/tools/retro/config_utils.py:487
↓ 1 callers
Function
verify_global_md_reuse
Verifies that global metadata reuse is possible by checking the loaded plans from the checkpoint are consistent, which means we have the sam
Megatron-LM/megatron/core/dist_checkpointing/strategies/state_dict_saver.py:171
↓ 1 callers
Function
verify_transformers_version
()
Megatron-LM/tools/checkpoint/loader_mixtral_hf.py:63
↓ 1 callers
Function
verify_transformers_version
()
Megatron-LM/tools/checkpoint/loader_llama_mistral.py:46
↓ 1 callers
Method
verify_vocabs_match
If both --true-vocab-size and --vocab-file are specified, verify they match. Return False (and exit) if they don't match; True otherw
Megatron-LM/tools/checkpoint/loader_base.py:167
↓ 1 callers
Function
visit
(obj, path, depth=0, is_root=False)
Megatron-LM/scripts/check_api_backwards_compatibility.py:104
↓ 1 callers
Function
vqav2_eval
Run VQAv2 evaluation.
Megatron-LM/examples/multimodal/evaluation/evaluate_vqav2.py:148
↓ 1 callers
Function
wait_for_weight_update
Wait until weight_version changes from current_wv.
openclaw-rl/oel/eval/gsm8k_personal_agent.py:545
↓ 1 callers
Method
wake_up
Resume CUDA memory for all tracked tensors.
slime/slime/backends/fsdp_utils/actor.py:313
↓ 1 callers
Function
walk
(module)
Megatron-LM/megatron/core/inference/utils.py:60
↓ 1 callers
Function
wandb_finalize_fn
()
Megatron-LM/megatron/training/checkpointing.py:719
↓ 1 callers
Function
weight_dequant
(x: torch.Tensor, s: torch.Tensor, block_size: int = 128)
slime/tools/fp8_cast_bf16.py:29
↓ 1 callers
Function
weighted_bias_quick_geglu
Token-wise weighted Quick-GEGLU activation with bias. Args: y: Input tensor before bias addition. bias: Bias tensor broadcastable
Megatron-LM/megatron/core/fusions/fused_bias_geglu.py:259
↓ 1 callers
Function
weighted_bias_quick_geglu_back
Backward helper for weighted Quick-GEGLU with bias. Returns gradients w.r.t input `y`, `bias`, and `weights`.
Megatron-LM/megatron/core/fusions/fused_bias_geglu.py:279
↓ 1 callers
Function
weighted_quick_geglu
Token-wise-weighted Quick-GEGLU activation. The weights tensor is expected to have the same first-dimension length as ``y`` and a trailing si
Megatron-LM/megatron/core/fusions/fused_bias_geglu.py:206
↓ 1 callers
Function
weighted_quick_geglu_back
Backward helper for weighted Quick-GEGLU. Returns gradient w.r.t input `y` and `weights`.
Megatron-LM/megatron/core/fusions/fused_bias_geglu.py:240
↓ 1 callers
Function
weighted_squared_relu
Element-wise weight applied after Squared-ReLU. Args: x (torch.Tensor): Input tensor. weights (torch.Tensor): Weight tensor that
Megatron-LM/megatron/core/fusions/fused_weighted_squared_relu.py:14
↓ 1 callers
Function
weighted_squared_relu_back
Backward for weighted Squared-ReLU. Returns gradients w.r.t ``x`` and ``weights``.
Megatron-LM/megatron/core/fusions/fused_weighted_squared_relu.py:41
↓ 1 callers
Function
weighted_swiglu
(y, weights)
Megatron-LM/megatron/core/fusions/fused_bias_swiglu.py:45
↓ 1 callers
Function
weighted_swiglu_back
(g, y, weights)
Megatron-LM/megatron/core/fusions/fused_bias_swiglu.py:90
↓ 1 callers
Function
white_space_fix
(text)
Megatron-LM/tools/retro/text_generation/evaluate.py:21
↓ 1 callers
Function
white_space_fix
(text)
slime/examples/search-r1/qa_em_format.py:25
↓ 1 callers
Function
white_space_fix
(text)
slime/slime/rollout/rm_hub/f1.py:11
↓ 1 callers
Function
window_reverse
Args: windows: (num_windows*B, window_size, window_size, C) window_size (int): Window size H (int): Height of image
Megatron-LM/megatron/legacy/model/vision/esvit_swin_backbone.py:59
↓ 1 callers
Function
window_reverse
Args: windows: (num_windows*B, window_size, window_size, C) window_size (int): Window size H (int): Height of image
Megatron-LM/megatron/legacy/model/vision/swin_backbone.py:54
↓ 1 callers
Method
worker_status
( self, worker_url: str, timeout: float = 10.0 )
terminal-rl/router_server.py:167
↓ 1 callers
Function
wrap_model_provider_with_freeze
(original_provider, args)
slime/slime/backends/megatron_utils/model_provider.py:299
↓ 1 callers
Function
wrapped_func
(*args, **kwargs)
Megatron-LM/megatron/core/utils.py:149
↓ 1 callers
Function
write_args_to_tensorboard
Write arguments to tensorboard.
Megatron-LM/megatron/training/initialize.py:430
↓ 1 callers
Function
write_eval_to_tensorboard
Write evaluation data to Tensorboard.
Megatron-LM/examples/multimodal/train.py:353
↓ 1 callers
Function
write_json
(text, path)
Megatron-LM/tools/checkpoint/loader_llama_mistral.py:70
↓ 1 callers
Function
write_jsonl
(dataset, out_path, data_source)
swe-rl/data/preprocess_swe_dataset.py:41
↓ 1 callers
Function
z_loss_func
Encourages the router's logits to remain small to enhance stability. Please refer to the ST-MoE paper (https://arxiv.org/pdf/2202.08906.pdf) for d
Megatron-LM/megatron/core/transformer/moe/moe_utils.py:123
↓ 1 callers
Method
zero_grad
Zero out the underlying grad_buffer and reset all buckets in preparation for the next iteration of training.
Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/param_and_grad_buffer.py:2384
↓ 1 callers
Method
zero_grad
Zeroes grads for the model related parameters, i.e., model_float16_groups and model_fp32_groups. We additionally zero the remaining g
Megatron-LM/megatron/core/optimizer/distrib_optimizer.py:2253
↓ 1 callers
Method
zero_grad
Zero or zero to none the gradients of all the parameters in the model.
Megatron-LM/megatron/core/optimizer/cpu_offloading/hybrid_optimizer.py:440
↓ 1 callers
Method
zero_grad_buffer
Zeros out all grad buffers. Needs to be called at the beginning of each training iteration alongside optimizer.zero_grad().
Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/megatron_fsdp.py:1196
↓ 1 callers
Method
zero_out_tensors
Replace all tensors inside arg, kwargs with zeroed copies.
Megatron-LM/megatron/core/transformer/cuda_graphs.py:983
↓ 1 callers
Function
zip_strict
Alternative to Python's builtin zip(..., strict=True) (available in 3.10+). Apart from providing functionality in earlier versions of Python
Megatron-LM/megatron/core/dist_checkpointing/utils.py:25
Method
__call__
(self, tokens, position_ids, attention_mask)
Megatron-LM/tools/run_vlm_text_generation.py:183
Method
__call__
(self, img)
Megatron-LM/megatron/legacy/data/vit_dataset.py:26
Method
__call__
(self, img)
Megatron-LM/megatron/legacy/data/vit_dataset.py:45
Method
__call__
(self, input)
Megatron-LM/megatron/legacy/data/vit_dataset.py:76
Method
__call__
(self, input)
Megatron-LM/megatron/legacy/data/vit_dataset.py:142
Method
__call__
(self, image)
Megatron-LM/megatron/legacy/data/vit_dataset.py:208
Method
__call__
Define call method for ImageNetPolicy class.
Megatron-LM/megatron/legacy/data/autoaugment.py:104
Method
__call__
Define call method for SubPolicy class.
Megatron-LM/megatron/legacy/data/autoaugment.py:311
Method
__call__
Call timer with name and log level.
Megatron-LM/megatron/core/timers.py:230
Method
__call__
Returns the value, the rank, and unit as a Tuple Returns: Tuple[float, int, str]: value, rank, unit
Megatron-LM/megatron/core/utils.py:1225
Method
__call__
Callable for the instance. Set context state, Useful when the context is used for cpu timers only when bdata=True Args:
Megatron-LM/megatron/core/utils.py:1896
Method
__call__
(self, inputs, stage, microbatch)
Megatron-LM/megatron/core/full_cuda_graph.py:65
Method
__call__
(self, *args, **kwargs)
Megatron-LM/megatron/core/full_cuda_graph.py:139
Method
__call__
( self, hidden: cute.Tensor, weight: cute.Tensor, labels: cute
Megatron-LM/megatron/core/fusions/linear_cross_entropy/blackwell/fwd_mainloop.py:547
← previous
next →
5,301–5,400 of 10,561, ranked by callers