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

Functiondispatch_micro

kernels/gemm/bf16fp32/archive/micros/test_load_to_lds/kernel.cpp:60

Functiondispatch_micro

kernels/gemm/bf16fp32/archive/micros/transpose_matmul/256_256_64_16.cpp:86

Functiondispatch_micro

kernels/layernorm/kernel.cpp:130

Functiondispatch_micro

kernels/attn/gqa/kernel.cpp:577

Functiondispatch_micro

kernels/attn/gqa/kernel_d64.cpp:577

Functiondispatch_micro

kernels/attn/gqa_causal/kernel.cpp:694

Functiondispatch_micro

kernels/attn/gqa_causal/kernel_d64.cpp:694

Functiondispatch_prep

analysis/attn/bkwd/benchmark/attn_bkwd_prep.cpp:189

Functiondispatch_prep

kernels/attn/gqa_backwards/attn_bkwd_prep.cpp:189

Functiondispatch_prep

kernels/attn/gqa_backwards/archive/GQA_bkwd_prep.cpp:175

Functiondispatch_prep

kernels/attn/gqa_causal_backwards/attn_bkwd_prep.cpp:189

Functiondispatch_prep

training/llama/csrc/attn_bkwd_prep.cpp:189

Functiondispatch_rotary

analysis/rotary/mi350x/kernel_4096.cpp:77

Functiondispatch_rotary

analysis/rotary/mi350x/kernel_1024.cpp:77

Functiondispatch_rotary

analysis/rotary/mi350x/kernel_2048.cpp:77

Functiondispatch_rotary

analysis/rotary/mi350x/kernel_16384.cpp:77

Functiondispatch_rotary

analysis/rotary/mi350x/kernel_8192.cpp:77

Functiondispatch_rotary

kernels/rotary/kernel.cpp:77

Functiondispatch_softmax

kernels/softmax/kernel.cpp:63

training/llama/train/datamodules/datasets/indexed_dataset.py:462

training/llama/train/datamodules/datasets/indexed_dataset.py:551

training/llama/train/datamodules/datasets/indexed_dataset.py:579

Methoddynamic_shared_memory

analysis/attn/bkwd/benchmark/attn_bkwd_prep.cpp:201

Methoddynamic_shared_memory

kernels/attn/gqa_backwards/attn_bkwd_prep.cpp:201

Methoddynamic_shared_memory

kernels/attn/gqa_backwards/archive/GQA_bkwd_prep.cpp:186

Methoddynamic_shared_memory

kernels/attn/gqa_causal_backwards/attn_bkwd_prep.cpp:201

Methoddynamic_shared_memory

training/llama/csrc/attn_bkwd_prep.cpp:201

Methodend_document

training/llama/train/datamodules/datasets/indexed_dataset.py:307

Methodend_document

training/llama/train/datamodules/datasets/indexed_dataset.py:587

training/llama/train/datamodules/datasets/indexed_dataset.py:565

A couple of optional utilities, controlled by main config file: - disabling warnings - forcing debug friendly configuration - verifying ex

training/llama/train/utils/utils.py:50

analysis/paper_experiments/swizzle_plot_scripts/swizzle_diagram.py:199

(self, index_file)

training/llama/train/datamodules/datasets/indexed_dataset.py:330

(self, index_file)

training/llama/train/datamodules/datasets/indexed_dataset.py:602

Makes sure everything closed properly.

training/llama/train/utils/utils.py:134

(ctx, a: torch.Tensor, b: torch.Tensor, c: torch.Tensor, bias: torch.Tensor, P: torch.Tensor,

analysis/baselines/gemm/triton_gemm_v02.py:119

(ctx, a: torch.Tensor, b: torch.Tensor, c: torch.Tensor, bias: torch.Tensor, P: torch.Tensor,

analysis/baselines/gemm/utils/gemm_wrapper.py:113

(ctx, q, k, v, causal, sm_scale)

analysis/baselines/attn/triton_baseline_v01.py:443

(ctx, q, k, v, o, metadata: MetaData)

analysis/baselines/attn/triton_baseline_v02.py:1087

( ctx, logits, labels, smoothing=0.0, logit_scale=1.0, lse_squ

training/llama/llama/ops/triton/cross_entropy.py:146

( ctx, x, weight, bias, residual=None, x1=None, weight

training/llama/llama/ops/triton/layer_norm.py:701

(self, x, residual=None, prenorm=False, residual_in_fp32=False)

training/llama/llama/ops/triton/layer_norm.py:943

( ctx, x, norm_weight, norm_bias, linear_weight, linear_bias,

training/llama/llama/ops/triton/layer_norm.py:959

Arguments: x: (batch, seqlen, hidden_dim) (where hidden_dim = num heads * head dim) if max_seqlen is None, else (

training/llama/llama/models/mha.py:170

hidden_states: (B, S, D) if batch_first else (S, B, D)

training/llama/llama/models/seq_common.py:60

hidden_states: (B, S, D) if batch_first else (S, B, D)

training/llama/llama/models/seq_common.py:84

hidden_states: (B, S, D) if batch_first else (S, B, D)

training/llama/llama/models/seq_common.py:114

hidden_states: (B, S, D) if batch_first else (S, B, D)

training/llama/llama/models/seq_common.py:140

r"""Inputs of forward function Args: x: the sequence fed to the positional encoder model (required). Shape: x:

training/llama/llama/models/seq_common.py:192

training/llama/llama/models/seq_common.py:227

training/llama/llama/models/seq_common.py:259

training/llama/llama/models/seq_common.py:282

training/llama/llama/models/seq_common.py:311

training/llama/llama/models/seq_common.py:335

r"""Pass the input through the encoder layer. Args: hidden_states: the sequence to the encoder layer (required). resi

training/llama/llama/models/block.py:79

training/llama/llama/models/mlp.py:34

training/llama/llama/models/mlp.py:69

( ctx, x, cos, sin, interleaved=False, inplace=False,

training/llama/llama/models/rotary.py:40

( ctx, qkv, cos, sin, cos_k=None, sin_k=None, interlea

training/llama/llama/models/rotary.py:196

(ctx, kv, cos, sin, interleaved=False, seqlen_offsets: Union[int, torch.Tensor] = 0)

training/llama/llama/models/rotary.py:270

qkv: (batch, seqlen, 3, nheads, headdim) or (batch, seqlen, num_heads_q + 2 * num_heads_k, headdim) if kv is none, else it's just

training/llama/llama/models/rotary.py:429

(self, input_ids, position_ids=None, inference_params=None, attention_mask=None, stream=None)

training/llama/llama/models/gpt.py:243

input_ids: (batch, seqlen) int tensor inference_params: for generation. Adapted from Megatron-LM (and Apex) https://github.co

training/llama/llama/models/gpt.py:318

input_ids: (batch, seqlen) position_ids: (batch, seqlen)

training/llama/llama/models/embedding.py:44

(ctx, q_bnhd: torch.Tensor, k_bnhd: torch.Tensor, v_bnhd: torch.Tensor)

training/llama/llama/models/attentions/hipkittens.py:18

Implements the multihead softmax attention. Arguments --------- qkv: The tensor containing the query, key, and value. (B,

training/llama/llama/models/attentions/hipkittens.py:147

Implements the multihead softmax attention. Arguments --------- q: The tensor containing the query. (B, Sq, H, D)

training/llama/llama/models/attentions/hipkittens.py:183

Implements the multihead softmax attention. Arguments --------- qkv: The tensor containing the query, key, and value. (B,

training/llama/llama/models/attentions/base.py:26

Implements the multihead softmax attention. Arguments --------- q: The tensor containing the query. (B, Sq, H, D)

training/llama/llama/models/attentions/base.py:78

Implements the multihead softmax attention. Arguments --------- qkv: The tensor containing the query, key, and value. (B,

training/llama/llama/models/attentions/aiter.py:30

Implements the multihead softmax attention. Arguments --------- q: The tensor containing the query. (B, Sq, H, D)

training/llama/llama/models/attentions/aiter.py:73

Arguments: input: (batch, vocab_size) target: (batch,) Returns: losses: (batch,) if reduction is

training/llama/train/losses/cross_entropy.py:47

(self, input_ids, attention_mask)

training/bert/tasks.py:182

(ctx, q_bnhd: torch.Tensor, k_bnhd: torch.Tensor, v_bnhd: torch.Tensor)

training/bert/models/hipkittens.py:39

( self, hidden_states: torch.Tensor, attention_mask: Optional[torch.FloatTensor] = Non

training/bert/models/hipkittens.py:390

( self, hidden_states: torch.Tensor, attention_mask: Optional[torch.FloatTensor] = Non

training/bert/models/base.py:33

( self, hidden_states: torch.Tensor, attention_mask: Optional[torch.FloatTensor] = Non

training/bert/models/aiter.py:34

Methodfrom_pretrained_hf

(cls, pretrained_model_name, device=None, **kwargs)

training/llama/llama/models/gpt.py:134

Functiongemm_async_kernel

kernels/gemm/bf16fp32/gfx1250/gemm_async.cpp:17

Functiongemm_double_buf_kernel

kernels/gemm/bf16fp32/gfx1250/gemm_double_buf.cpp:15

Functiongemm_expert_kernel

kernels/gemm/bf16fp32/gfx1250/gemm_expert.cpp:20

Functiongemm_naive_kernel

kernels/gemm/bf16fp32/gfx1250/gemm_naive.cpp:21

Functiongemm_segment_kernel

kernels/gemm/bf16fp32/gfx1250/gemm_segment.cpp:16

Functiongemm_split_bar_kernel

kernels/gemm/bf16fp32/gfx1250/gemm_split_bar.cpp:16

Functiongemm_tdm_arrive_kernel

kernels/gemm/bf16fp32/gfx1250/gemm_tdm_arrive.cpp:46

Functionget_autotune_config

analysis/baselines/gemm/triton_gemm_v01.py:225

Methodget_context

distributed-kernels/iris_py.cpp:97

Functionget_default_tuning_result_filename

analysis/baselines/gemm/utils/utils.py:83

Methodget_doc_idx

training/llama/train/datamodules/datasets/indexed_dataset.py:554

Methodget_dtype

distributed-kernels/iris_py.cpp:30

Functionget_filename_compile_driver

analysis/baselines/gemm/utils/utils.py:73

Functionget_filename_myKernels

analysis/baselines/gemm/utils/utils.py:62

Functionget_filename_profile_driver

(M, N, K, job_id)

analysis/baselines/gemm/utils/utils.py:78

Functionget_filename_without_extension

analysis/baselines/gemm/utils/utils.py:67

Functionget_gfx_version

analysis/baselines/attn/triton_baseline_v02.py:359

Functionget_num_sms

analysis/baselines/gemm/utils/utils.py:34

Functionget_output

(x, residual, norm, dropout_p=DROPOUT_P)

kernels/layernorm/test_python.py:38

Methodget_shape

distributed-kernels/iris_py.cpp:29

analysis/attn/bkwd/benchmark/attn_bkwd_prep.cpp:199

kernels/attn/gqa_backwards/attn_bkwd_prep.cpp:199

← previous next →1,001–1,100 of 1,340, ranked by callers