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
Method
backup_tags
(self)
slime/slime/utils/tensor_backper.py:23
Method
backup_tags
(self)
slime/slime/utils/tensor_backper.py:48
Method
backup_tags
(self)
slime/slime/utils/tensor_backper.py:85
Function
backward
backward host function
Megatron-LM/megatron/core/fusions/linear_cross_entropy/blackwell/entry.py:309
Method
backward
(ctx, grad_out)
openclaw-combine/hint_opd_loss.py:134
Method
backward
(ctx, grad_out)
openclaw-combine/hint_opd_loss.py:198
Method
backward
(ctx, grad_output)
Megatron-LM/pretrain_ict.py:62
Method
backward
(ctx, output_grads)
Megatron-LM/megatron/legacy/model/fused_softmax.py:33
Method
backward
(ctx, output_grads)
Megatron-LM/megatron/legacy/model/fused_softmax.py:69
Method
backward
(ctx, grad_output)
Megatron-LM/megatron/legacy/model/fused_bias_gelu.py:39
Method
backward
No-op
Megatron-LM/megatron/core/utils.py:682
Method
backward
Backward pass for scaled upper-triangular masked softmax. Args: ctx: Autograd context containing saved tensors from forward.
Megatron-LM/megatron/core/fusions/fused_softmax.py:40
Method
backward
Backward pass for scaled masked softmax. Args: ctx: Autograd context containing saved tensors from forward. output_gr
Megatron-LM/megatron/core/fusions/fused_softmax.py:90
Method
backward
Backward implementation for the cross entropy loss.
Megatron-LM/megatron/core/fusions/fused_cross_entropy.py:124
Method
backward
Backward function for IndicesToMultihot Convert multihot probs representation to indices. indices is ignored in the backward function
Megatron-LM/megatron/core/fusions/fused_indices_converter.py:239
Method
backward
Backward function for ApplyMLARotaryEmbQ. Args: grad: [seq_len, batch_size, head_num, qk_head_dim + emb_dim]
Megatron-LM/megatron/core/fusions/fused_mla_yarn_rope_apply.py:285
Method
backward
Backward function for ApplyMLARotaryEmbKV. Args: dk: [seq_len, batch_size, head_num, emb_dim + k_dim] or
Megatron-LM/megatron/core/fusions/fused_mla_yarn_rope_apply.py:692
Method
backward
Backward pass of biased GEGLU activation. Args: ctx: Autograd context object containing saved tensors from forward pass.
Megatron-LM/megatron/core/fusions/fused_bias_geglu.py:104
Method
backward
Backward pass of GEGLU activation. Args: ctx: Autograd context object containing saved tensors from forward pass. gra
Megatron-LM/megatron/core/fusions/fused_bias_geglu.py:138
Method
backward
Backward pass of weighted Quick-GEGLU. Args: ctx: Autograd context object containing saved tensors from forward pass.
Megatron-LM/megatron/core/fusions/fused_bias_geglu.py:333
Method
backward
Backward pass of weighted Quick-GEGLU with bias. Args: ctx: Autograd context object containing saved tensors from forward pass.
Megatron-LM/megatron/core/fusions/fused_bias_geglu.py:387
Method
backward
(ctx, grad_output)
Megatron-LM/megatron/core/fusions/fused_bias_gelu.py:44
Method
backward
The backward pass of the Linear Cross Entropy. Args: dlogprobs (torch.Tensor): The gradient of the cross entropy, with sh
Megatron-LM/megatron/core/fusions/fused_linear_cross_entropy.py:184
Method
backward
backward method for `WeightedSquaredReLUFunction` Args: ctx : context object to store intermediate tensors. grad_outp
Megatron-LM/megatron/core/fusions/fused_weighted_squared_relu.py:79
Method
backward
Backward pass of biased SwiGLU activation. Args: ctx: Autograd context object containing saved tensors from forward pass.
Megatron-LM/megatron/core/fusions/fused_bias_swiglu.py:128
Method
backward
Backward pass of SwiGLU activation. Args: ctx: Autograd context object containing saved tensors from forward pass. gr
Megatron-LM/megatron/core/fusions/fused_bias_swiglu.py:173
Method
backward
(ctx, grad_output)
Megatron-LM/megatron/core/fusions/fused_bias_swiglu.py:202
Method
backward
Backward with frozen weight.
Megatron-LM/megatron/core/tensor_parallel/layers.py:356
Method
backward
Backward.
Megatron-LM/megatron/core/tensor_parallel/layers.py:503
Method
backward
Vocab parallel cross entropy backward function.
Megatron-LM/megatron/core/tensor_parallel/cross_entropy.py:192
Method
backward
Backward pass.
Megatron-LM/megatron/core/tensor_parallel/random.py:579
Method
backward
Backward function.
Megatron-LM/megatron/core/tensor_parallel/mappings.py:212
Method
backward
Backward function.
Megatron-LM/megatron/core/tensor_parallel/mappings.py:231
Method
backward
Backward function.
Megatron-LM/megatron/core/tensor_parallel/mappings.py:251
Method
backward
Backward function.
Megatron-LM/megatron/core/tensor_parallel/mappings.py:271
Method
backward
Backward function.
Megatron-LM/megatron/core/tensor_parallel/mappings.py:291
Method
backward
Backward function.
Megatron-LM/megatron/core/tensor_parallel/mappings.py:328
Method
backward
Backward function.
Megatron-LM/megatron/core/tensor_parallel/mappings.py:368
Method
backward
Backward function.
Megatron-LM/megatron/core/tensor_parallel/mappings.py:395
Method
backward
Backward function.
Megatron-LM/megatron/core/tensor_parallel/mappings.py:415
Method
backward
Backward pass of the RegisterFSDPBackwardFunction function.
Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/megatron_fsdp.py:1256
Method
backward
Passes through gradients unchanged in the backward pass.
Megatron-LM/megatron/core/pipeline_parallel/utils.py:139
Method
backward
(ctx, *grad_output)
Megatron-LM/megatron/core/pipeline_parallel/fine_grained_activation_offload.py:503
Method
backward
(ctx, grad_output)
Megatron-LM/megatron/core/pipeline_parallel/fine_grained_activation_offload.py:542
Method
backward
(ctx, grad_output)
Megatron-LM/megatron/core/extensions/transformer_engine.py:1210
Method
backward
If this is the first bwd pass of this runner, record that a bwd graph needs to be created.
Megatron-LM/megatron/core/transformer/cuda_graphs.py:426
Method
backward
Replay the backward graph of the passed runner.
Megatron-LM/megatron/core/transformer/cuda_graphs.py:495
Method
backward
Backward pass for the native attention function with context parallelism
Megatron-LM/megatron/core/transformer/dot_product_attention_context_parallel.py:237
Method
backward
Compute and scale the gradient for mtp loss.. Args: grad_output (torch.Tensor): The gradient of the output. Returns:
Megatron-LM/megatron/core/transformer/multi_token_prediction.py:569
Method
backward
Compute and scale the gradient for indexer loss. Args: grad_output: The gradient of the output. Returns: Tup
Megatron-LM/megatron/core/transformer/experimental_attention_variant/dsa.py:279
Method
backward
Compute and scale the gradient for auxiliary loss.. Args: grad_output (torch.Tensor): The gradient of the output. Return
Megatron-LM/megatron/core/transformer/moe/moe_utils.py:194
Method
backward
Backward pass propagates the gradient for logits.
Megatron-LM/megatron/core/transformer/moe/moe_utils.py:940
Method
backward
Backward pass of the RouterGatingLinearFunction function.
Megatron-LM/megatron/core/transformer/moe/moe_utils.py:987
Method
backward
Backward pass of fused dispatch.
Megatron-LM/megatron/core/transformer/moe/fused_a2a.py:140
Method
backward
Backward pass of fused combine.
Megatron-LM/megatron/core/transformer/moe/fused_a2a.py:191
Method
backward
Backward pass of fused dispatch of the HybridEP backend
Megatron-LM/megatron/core/transformer/moe/fused_a2a.py:398
Method
backward
Backward pass of fused combine of the HybridEP backend
Megatron-LM/megatron/core/transformer/moe/fused_a2a.py:429
Method
backward
(ctx, g_logprobs: torch.Tensor)
Megatron-LM/tests/unit_tests/fusions/test_fused_linear_cross_entropy.py:1094
Method
backward
Backward pass for GateUpProjFunction - Pure Python implementation. Forward: output = input @ w1 (without topk_weight multiplication)
slime/tests/test_fused_experts_backward.py:56
Method
backward
(ctx, *grads)
slime/slime_plugins/models/hf_attention.py:67
Method
backward
(ctx, grad_output: torch.Tensor)
slime/slime/utils/ppo_utils.py:185
Method
backward
Backward pass for GateUpProjFunction using Triton kernels. Args: grad_output: shape (num_tokens * topk, N) Retu
slime/slime/backends/fsdp_utils/kernels/fused_experts.py:96
Method
backward
(ctx, grad_output)
slime/slime/backends/fsdp_utils/kernels/fused_experts.py:188
Method
backward
(ctx, grad_output)
slime/slime/backends/fsdp_utils/kernels/fused_experts.py:379
Method
backward_dw
Compute weight gradients during the backward pass if delay_wgrad_compute is enabled.
Megatron-LM/megatron/core/extensions/transformer_engine.py:656
Method
backward_dw
Compute weight gradients during the backward pass if delay_wgrad_compute is enabled.
Megatron-LM/megatron/core/extensions/transformer_engine.py:755
Method
backward_dw
Compute weight gradients during the backward pass if delay_wgrad_compute is enabled.
Megatron-LM/megatron/core/extensions/transformer_engine.py:855
Method
backward_dw
Compute weight gradients during the backward pass if delay_wgrad_compute is enabled.
Megatron-LM/megatron/core/extensions/transformer_engine.py:1562
Method
backward_dw
(self)
Megatron-LM/megatron/core/transformer/mlp.py:254
Method
backward_dw
Performs backward pass for weight gradients in TEGroupedMLP. This method executes the backward pass for weight gradients by calling b
Megatron-LM/megatron/core/transformer/moe/experts.py:1029
Method
backward_dw
Backward pass for weight gradients in SequentialMLP.
Megatron-LM/megatron/core/transformer/moe/experts.py:1152
Method
backward_dw
Compute weight gradients for experts and shared experts.
Megatron-LM/megatron/core/transformer/moe/moe_layer.py:329
Method
backward_impl
Implements the backward pass for the transformer layer node.
Megatron-LM/megatron/core/models/gpt/fine_grained_callables.py:277
Method
base_generate
(self, request: InferenceRequest)
Megatron-LM/megatron/rl/inference/megatron.py:166
Method
base_generate
(self, request: ChatInferenceRequest)
Megatron-LM/megatron/rl/inference/inference_interface.py:111
Method
base_generate
(request: ChatInferenceRequest)
Megatron-LM/megatron/rl/server/inference/inference_interface_server.py:72
Method
batch
Pad/stack individual samples into a single batch dict.
Megatron-LM/examples/mimo/data/energon_avlm_task_encoder.py:288
Method
batch
Pad/stack individual samples into a single batch dict.
Megatron-LM/examples/mimo/data/energon_vlm_task_encoder.py:195
Method
batch
(self, samples: List[Union[ImageTaskSample, ImageTaskSamplePacked]])
Megatron-LM/examples/multimodal/dataset_helpers.py:791
Function
bert_vocab
(odir)
Megatron-LM/tests/unit_tests/data/test_preprocess_data.py:207
Method
bias_act_func
(intermediate_parallel, bias_parallel, permuted_probs)
Megatron-LM/megatron/core/transformer/moe/experts.py:901
Function
bias_dropout_add_fused_inference
(x: torch.Tensor, bias: Optional[torch.Tensor],
Megatron-LM/megatron/legacy/model/transformer.py:868
Function
bias_dropout_add_fused_inference
( x_with_bias: Tuple[torch.Tensor, Optional[torch.Tensor]], residual: torch.Tensor, prob: float )
Megatron-LM/megatron/core/fusions/fused_bias_dropout.py:75
Function
bias_dropout_add_fused_train
(x: torch.Tensor, bias: Optional[torch.Tensor],
Megatron-LM/megatron/legacy/model/transformer.py:860
Method
bos
Returns id of beginning of sentence token.
Megatron-LM/megatron/core/tokenizers/text/text_tokenizer.py:186
Method
bos
The BOS token id Raises: NotImplementedError: Non-abstract, optional attribute
Megatron-LM/megatron/core/datasets/megatron_tokenizer.py:138
Method
bos
(self)
Megatron-LM/megatron/training/tokenizer/tokenizer.py:186
Method
bos
Id of the beginning of sentence token in the vocabulary.
Megatron-LM/megatron/training/tokenizer/tokenizer.py:291
Method
bos
(self)
Megatron-LM/megatron/training/tokenizer/tokenizer.py:518
Method
bos
(self)
Megatron-LM/megatron/training/tokenizer/tokenizer.py:749
Method
bos
Beginning of sequence token ID.
Megatron-LM/megatron/training/tokenizer/sft_tokenizer.py:174
Method
bos_id
Get the beginning-of-sequence ID.
Megatron-LM/megatron/core/tokenizers/text/libraries/bytelevel_tokenizer.py:139
Method
bos_id
Returns id of beggining of sentence token.
Megatron-LM/megatron/core/tokenizers/text/libraries/huggingface_tokenizer.py:286
Method
bos_id
Returns id of begginning of sentence token.
Megatron-LM/megatron/core/tokenizers/text/libraries/sentencepiece_tokenizer.py:323
Method
bos_id
Property alias to match MegatronTokenizer; returns bos_id if available.
Megatron-LM/megatron/core/tokenizers/text/libraries/abstract_tokenizer.py:129
Method
bos_id
Returns id of beginning of sentence token.
Megatron-LM/megatron/core/tokenizers/text/libraries/tiktoken_tokenizer.py:241
Method
bos_token
Beginning of sentence token id
Megatron-LM/megatron/training/tokenizer/tokenizer.py:306
Method
bound_device_id
(self)
slime/slime/utils/reloadable_process_group.py:257
Method
broadcast_params
Syncs parameters across all DP ranks.
Megatron-LM/megatron/core/distributed/fsdp/src/megatron_fsdp/megatron_fsdp.py:1213
Method
broadcast_params
All rank broadcast updated local params.
Megatron-LM/megatron/core/optimizer/layer_wise_optimizer.py:185
← previous
next →
6,701–6,800 of 10,561, ranked by callers