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
Functionstore_acc
kernels/gemm/bf16fp32/gfx1250/common.h:87
Functionstore_global_float
analysis/bf16_gemm/mi325x/utils.cpp:234
Functionstore_linear
kernels/gemm/bf16fp32/archive/utils.cpp:447
Functionstore_registers_to_shared
analysis/bf16_gemm/mi325x/utils.cpp:93
Functionstore_registers_to_shared
kernels/gemm/bf16fp32/archive/utils.cpp:105
Functionstore_shuffled
analysis/attn/bkwd/benchmark/attn_bkwd_prep.cpp:88
Functionstore_shuffled
kernels/attn/gqa_backwards/attn_bkwd_prep.cpp:88
Functionstore_shuffled
kernels/attn/gqa_backwards/archive/GQA_bkwd_prep.cpp:75
Functionstore_shuffled
kernels/attn/gqa_causal_backwards/attn_bkwd_prep.cpp:88
Functionstore_shuffled
training/llama/csrc/attn_bkwd_prep.cpp:88
Functionstore_transposed
analysis/bf16_gemm/mi325x/utils.cpp:252
Functionstreamk_gemm
( A, B, C, bias_ptr, P, locks, M, N, K, stride_am, stride_ak,
analysis/baselines/gemm/streamk_kernel.py:6
Methodsupports_prefetch
(self)
training/llama/train/datamodules/datasets/indexed_dataset.py:223
Methodsupports_prefetch
(self)
training/llama/train/datamodules/datasets/indexed_dataset.py:234
Methodsupports_prefetch
(self)
training/llama/train/datamodules/datasets/indexed_dataset.py:561
Methodtest_dataloader
The test dataloader
training/llama/train/datamodules/language_modeling_hf.py:276
Functiontest_op
(Z, H, N_CTX, HEAD_DIM, causal, dtype=torch.bfloat16)
analysis/baselines/attn/triton_baseline_v01.py:527
Functiontest_op_bwd
(Z, H, N_CTX, D_HEAD, qseqlen_not_equal_kseqlen, causal, torch_sdpa_test, use_alibi, dtype=tor
analysis/baselines/attn/triton_baseline_v02.py:1564
Functiontest_op_fwd
(Z, HQ, HK, N_CTX_Q, N_CTX_K, D_HEAD, causal, use_alibi, layout, dtype=torch.bfloat16)
analysis/baselines/attn/triton_baseline_v02.py:1363
Functiontest_op_fwd_bias
(Z, H, N_CTX_Q, N_CTX_K, D_HEAD, causal, use_bias, dtype)
analysis/baselines/attn/triton_baseline_v02.py:1513
Functiontest_op_persistent_fwd
(Z, HQ, HK, N_CTX_Q, N_CTX_K, D_HEAD, causal, use_alibi, layout, persistent, dtype=
analysis/baselines/attn/triton_baseline_v02.py:1437
Methodtest_step
(self, batch: Any, batch_idx: int)
training/llama/train/tasks/seq.py:131
Functiontk_fused_rotary
analysis/rotary/mi350x/kernel_4096.cpp:45
Functiontk_fused_rotary
analysis/rotary/mi350x/kernel_1024.cpp:45
Functiontk_fused_rotary
analysis/rotary/mi350x/kernel_2048.cpp:45
Functiontk_fused_rotary
analysis/rotary/mi350x/kernel_16384.cpp:45
Functiontk_fused_rotary
analysis/rotary/mi350x/kernel_8192.cpp:45
Functiontk_fused_rotary
kernels/rotary/kernel.cpp:45
Methodtokenize_concat
(examples)
training/llama/train/datamodules/language_modeling_hf.py:155
Methodtrain_dataloader
(self, **kwargs)
training/llama/train/datamodules/base.py:215
Methodtrain_dataloader
The train dataloader
training/llama/train/datamodules/language_modeling_hf.py:253
Methodtraining_step
(self, batch: Any, batch_idx: int)
training/llama/train/tasks/seq.py:125
Methodupdate
Compute and store intermediate statistics for Perplexity. Args: preds: Probabilities assigned to each token in a s
training/llama/train/metrics/perplexity.py:51
Methodval_dataloader
The val dataloader
training/llama/train/datamodules/language_modeling_hf.py:272
Methodvalidation_step
(self, batch: Any, batch_idx: int)
training/llama/train/tasks/seq.py:128
Functionwikitext_detokenize
Wikitext is whitespace tokenized and we remove these whitespaces. Taken from https://github.com/NVIDIA/Megatron-LM/blob/main/tasks/zeroshot_g
training/llama/train/datamodules/datasets/detokenizer.py:10
Methodwrite_ids_to_disk
(example, filename)
training/llama/train/datamodules/language_modeling_hf.py:199
Methodwrite_ids_to_shm
(example, shm_name, array_len)
training/llama/train/datamodules/language_modeling_hf.py:170
Method~IrisInstance
distributed-kernels/iris_py.cpp:64
Method~IrisTensor
distributed-kernels/iris_py.cpp:22
← previous1,301–1,340 of 1,340, ranked by callers