Functions — github.com/HazyResearch/HipKittens

CodeHub MCPcopy Create free account

hub / github.com/HazyResearch/HipKittens / functions

Functions1,340 in github.com/HazyResearch/HipKittens

⨍ Functions 1,340◇ Types & classes 190↳ Endpoints 1

(self, dataset_name, tokenizer_name, dataset_config_name=None, max_length=1024, cache_dir=Non

training/llama/train/datamodules/language_modeling_hf.py:42

tokens should be a numpy array

training/llama/train/datamodules/datasets/lm_dataset.py:12

training/llama/train/datamodules/datasets/indexed_dataset.py:228

(self, out_file, dtype=np.int32)

training/llama/train/datamodules/datasets/indexed_dataset.py:290

(self, path, skip_warmup=False)

training/llama/train/datamodules/datasets/indexed_dataset.py:404

(self, path, skip_warmup=False)

training/llama/train/datamodules/datasets/indexed_dataset.py:472

(self, out_file, dtype=np.int64)

training/llama/train/datamodules/datasets/indexed_dataset.py:572

(self, tokenizer, n_classes: int)

training/bert/tasks.py:155

(self, config, layer_idx=None, deterministic: bool = False)

training/bert/models/hipkittens.py:369

(self, config, layer_idx: Optional[int] = None)

training/bert/models/base.py:7

(self, config, layer_idx=None, deterministic: bool = False)

training/bert/models/aiter.py:12

Method__init_subclass__

(cls, **kwargs)

training/llama/train/datamodules/base.py:175

training/llama/train/datamodules/fault_tolerant_sampler.py:43

training/llama/train/datamodules/fault_tolerant_sampler.py:86

Function__launch_bounds__

analysis/attn/bkwd/benchmark/attn_bkwd_causal_HBN.cpp:49

Function__launch_bounds__

analysis/attn/bkwd/benchmark/attn_bkwd_non_causal_8_warps.cpp:105

Function__launch_bounds__

analysis/attn/bkwd/benchmark/attn_bkwd_prep.cpp:164

Function__launch_bounds__

analysis/attn/bkwd/benchmark/attn_fwd_causal.cpp:160

Function__launch_bounds__

analysis/attn/bkwd/benchmark/attn_bkwd_causal_HNB.cpp:49

Function__launch_bounds__

analysis/attn/bkwd/benchmark/attn_fwd_non_causal.cpp:80

Function__launch_bounds__

analysis/attn/bkwd/benchmark/attn_bkwd_non_causal_4_warps.cpp:118

Function__launch_bounds__

analysis/attn/bkwd/benchmark/attn_bkwd_non_causal_HBN.cpp:48

Function__launch_bounds__

analysis/attn/bkwd/benchmark/attn_bkwd_non_causal_HNB.cpp:48

Function__launch_bounds__

analysis/attn/fwd/benchmark/attn_fwd_causal.cpp:181

Function__launch_bounds__

analysis/attn/fwd/benchmark/attn_fwd_causal_d64.cpp:181

Function__launch_bounds__

analysis/attn/fwd/benchmark/attn_fwd_non_causal.cpp:101

Function__launch_bounds__

analysis/attn/fwd/benchmark/attn_fwd_non_causal_d64.cpp:102

Function__launch_bounds__

kernels/attn/gqa_backwards/attn_bkwd_prep.cpp:164

Function__launch_bounds__

kernels/attn/gqa_backwards/attn_bkwd_non_causal.cpp:48

Function__launch_bounds__

kernels/attn/gqa_backwards/attn_fwd_non_causal.cpp:80

Function__launch_bounds__

kernels/attn/gqa_backwards/archive/GQA_bkwd_asm.cpp:34

Function__launch_bounds__

kernels/attn/gqa_backwards/archive/GQA_bkwd_4warps.cpp:105

Function__launch_bounds__

kernels/attn/gqa_backwards/archive/GQA_bkwd_8warps.cpp:92

Function__launch_bounds__

kernels/attn/gqa_backwards/archive/GQA_fwd.cpp:54

Function__launch_bounds__

kernels/attn/gqa_backwards/archive/GQA_bkwd_prep.cpp:150

Function__launch_bounds__

kernels/attn/gqa_causal_backwards/attn_bkwd_causal.cpp:49

Function__launch_bounds__

kernels/attn/gqa_causal_backwards/attn_bkwd_prep.cpp:164

Function__launch_bounds__

kernels/attn/gqa_causal_backwards/attn_fwd_causal.cpp:160

Function__launch_bounds__

kernels/attn/gqa/kernel.cpp:102

Function__launch_bounds__

kernels/attn/gqa/kernel_d64.cpp:102

Function__launch_bounds__

kernels/attn/gqa_causal/kernel.cpp:181

Function__launch_bounds__

kernels/attn/gqa_causal/kernel_d64.cpp:181

Function__launch_bounds__

training/llama/csrc/attn_bkwd_causal_HBN.cpp:49

Function__launch_bounds__

training/llama/csrc/attn_bkwd_prep.cpp:164

Function__launch_bounds__

training/llama/csrc/attn_fwd_causal.cpp:160

Function__launch_bounds__

training/llama/csrc/attn_bkwd_causal_HNB.cpp:49

training/llama/train/datamodules/datasets/lm_dataset.py:25

training/llama/train/datamodules/datasets/indexed_dataset.py:207

training/llama/train/datamodules/datasets/indexed_dataset.py:469

training/llama/train/datamodules/datasets/indexed_dataset.py:506

training/bert/tasks.py:89

(cls, input_array, shm=None)

training/llama/train/datamodules/language_modeling_hf.py:31

Method__setstate__

training/llama/train/datamodules/datasets/indexed_dataset.py:484

training/llama/train/datamodules/base.py:234

Function_attn_bwd

(Q, K, V, sm_scale, # DO, # DQ, DK, DV, # M, D, # s

analysis/baselines/attn/triton_baseline_v01.py:311

Function_attn_bwd

(Q, K, V, sm_scale, alibi_slopes, DO, DQ, DK, DV, M, D, # shared by Q/K/V/DO. stri

analysis/baselines/attn/triton_baseline_v02.py:920

Function_attn_bwd_preprocess

(O, DO, # Delta, # Z, H, N_CTX, #

analysis/baselines/attn/triton_baseline_v01.py:193

Function_attn_bwd_preprocess

( Out, DO, Delta, stride_oz, stride_oh, stride_om, stride_on, stride_doz,

analysis/baselines/attn/triton_baseline_v02.py:770

Function_attn_fwd

(Q, K, V, sm_scale, M, Out, # stride_qz, stride_qh, stride_qm, stride_qk, # stri

analysis/baselines/attn/triton_baseline_v01.py:102

Method_cache_dir_name

training/llama/train/datamodules/language_modeling_hf.py:250

Method_collate_callback

(cls, x, resolution=None)

training/llama/train/datamodules/base.py:107

Method_collate_callback

(cls, x, resolution=None, img_size=None, channels_last=True)

training/llama/train/datamodules/base.py:136

Method_collate_fn

Default collate function. Generally accessed by the dataloader() methods to pass into torch DataLoader Arguments:

training/llama/train/datamodules/base.py:71

Function_deprecated

(*args, **kwargs)

training/llama/train/datamodules/base.py:11

Method_eval_dataloader

(self, dataset, train_resolution=None, eval_resolutions=None, **kwargs)

training/llama/train/datamodules/base.py:245

Method_forward_reduce_state_update

forward computation using single call to `update` to calculate the metric value on the current batch and accumulate global state. This

training/llama/train/metrics/num_tokens.py:39

Function_init_weights

(module, n_layer, initializer_range=0.02, rescale_prenorm_residual=True, use_weight_init=True)

training/llama/llama/models/gpt.py:149

Function_layer_norm_bwd_kernel

( X, # pointer to the input W, # pointer to the weights B, # pointer to the biases Y, # po

training/llama/llama/ops/triton/layer_norm.py:407

Function_layer_norm_fwd_1pass_kernel

( X, # pointer to the input Y, # pointer to the output W, # pointer to the weights B, # po

training/llama/llama/ops/triton/layer_norm.py:145

Method_return_callback

(cls, return_value, resolution=None)

training/llama/train/datamodules/base.py:123

Method_return_callback

(cls, return_value, resolution=None, img_size=None, channels_last=True)

training/llama/train/datamodules/base.py:154

Method_train_dataloader

(self, dataset, train_resolution=None, eval_resolutions=None, **kwargs)

training/llama/train/datamodules/base.py:239

(self, np_array)

training/llama/train/datamodules/datasets/indexed_dataset.py:299

(self, np_array)

training/llama/train/datamodules/datasets/indexed_dataset.py:582

Methodallocate_inference_cache

(self, batch_size, max_seqlen, dtype=None)

training/llama/llama/models/mha.py:146

Methodallocate_inference_cache

(self, batch_size, max_seqlen, dtype=None, **kwargs)

training/llama/llama/models/block.py:76

Methodallocate_inference_cache

(self, batch_size, max_seqlen, dtype=None, **kwargs)

training/llama/llama/models/gpt.py:313

Functionapply_rotary_emb

Arguments: x: (batch_size, seqlen, nheads, headdim) if cu_seqlens is None else (total_seqlen, nheads, headdim) cos, s

training/llama/llama/models/rotary.py:93

Functionapply_rotary_emb_torch

x: (batch_size, seqlen, nheads, headdim) cos, sin: (seqlen, rotary_dim / 2) or (batch_size, seqlen, rotary_dim / 2)

training/llama/llama/models/rotary.py:23

Functionatomic_pk_add_bf16_with_warpid

analysis/attn/bkwd/benchmark/attn_bkwd_non_causal_8_warps.cpp:58

Functionatomic_pk_add_bf16_with_warpid

analysis/attn/bkwd/benchmark/utils.cpp:6

Functionatomic_pk_add_bf16_with_warpid

analysis/attn/bkwd/benchmark/attn_bkwd_non_causal_4_warps.cpp:57

Functionatomic_pk_add_bf16_with_warpid

kernels/attn/gqa_backwards/utils.cpp:6

Functionatomic_pk_add_bf16_with_warpid

kernels/attn/gqa_backwards/archive/GQA_bkwd_4warps.cpp:44

Functionatomic_pk_add_bf16_with_warpid

kernels/attn/gqa_backwards/archive/GQA_bkwd_8warps.cpp:45

Functionatomic_pk_add_bf16_with_warpid

kernels/attn/gqa_backwards/archive/utils.cpp:6

Functionatomic_pk_add_bf16_with_warpid

kernels/attn/gqa_causal_backwards/utils.cpp:6

Functionatomic_pk_add_bf16_with_warpid

training/llama/csrc/utils.cpp:6

Functionattn_fwd

(Q, K, V, bias, SM_SCALE: tl.constexpr, L, Out, stride_qz, stride_qh, stride_qm, stride_qk, stride_kz,

analysis/baselines/attn/triton_baseline_v02.py:443

(ctx, grad_losses, grad_z_losses)

training/llama/llama/ops/triton/cross_entropy.py:251

(ctx, dy, *args)

training/llama/llama/ops/triton/layer_norm.py:799

(ctx, dout, *args)

training/llama/llama/ops/triton/layer_norm.py:1018

training/llama/llama/models/rotary.py:223

training/llama/llama/models/rotary.py:287

(ctx, dO_bnhd: torch.Tensor)

training/llama/llama/models/attentions/hipkittens.py:65

Functionbench_flash_attention

(BATCH, H, N_CTX, HEAD_DIM, causal, mode, provider, device="cuda")

analysis/baselines/attn/triton_baseline_v01.py:602

Functionbench_flash_attention

(BATCH, HQ, HK, N_CTX_Q, N_CTX_K, D_HEAD, dtype, causal, mode, provider, device="cuda",

analysis/baselines/attn/triton_baseline_v02.py:1712

Functionbenchmark

(M, N, K, provider, fp8_inputs)

analysis/baselines/gemm/triton_gemm_v01.py:425

analysis/attn/bkwd/benchmark/attn_bkwd_prep.cpp:200

kernels/attn/gqa_backwards/attn_bkwd_prep.cpp:200

← previous next →801–900 of 1,340, ranked by callers