MCPcopy Create free account

hub / github.com/bytedance/ABQ-LLM / functions

Functions14,558 in github.com/bytedance/ABQ-LLM

⨍ Functions 14,558◇ Types & classes 4,979
↓ 5 callersMethodname
(self)
fastertransformer/3rdparty/cutlass/tools/library/scripts/rt.py:407
↓ 5 callersMethodnvcc
(self)
fastertransformer/3rdparty/cutlass/tools/library/scripts/pycutlass/src/pycutlass/compiler.py:143
↓ 5 callersFunctionparam2distributed
( pt_param, layer_id, save_dir, dest_key, n_inference_gpus, split_axis, )
fastertransformer/examples/pytorch/gptj/utils/gptj_ckpt_convert.py:96
↓ 5 callersFunctionparam2distributed
( pt_param, layer_id, save_dir, dest_key, n_inference_gpus, split_axis, )
fastertransformer/examples/pytorch/gptj/utils/huggingface_gptj_ckpt_convert.py:26
↓ 5 callersFunctionpositional_embedding
(pos_seq, inv_freq, bsz=None)
fastertransformer/examples/tensorflow/xlnet/modeling.py:54
↓ 5 callersFunctionpretrain
Main training program. This function will run the followings in the order provided: 1) initialize Megatron. 2) setup model, optim
fastertransformer/3rdparty/Megatron-LM/megatron/training.py:64
↓ 5 callersMethodqualified_name
Returns a string name for debugging
fastertransformer/3rdparty/cutlass/tools/profiler/src/problem_space.h:215
↓ 5 callersMethodregisterCallback
fastertransformer/src/fastertransformer/triton_backend/transformer_triton_backend.hpp:273
↓ 5 callersFunctionrelatively_equal_float
fastertransformer/3rdparty/cutlass/include/cutlass/relatively_equal.h:56
↓ 5 callersMethodrelease
Graceful shutdown
fastertransformer/3rdparty/cutlass/tools/library/src/manifest.cpp:75
↓ 5 callersMethodrun
Runs the kernel
fastertransformer/3rdparty/cutlass/tools/library/src/gemm_operation.h:290
↓ 5 callersMethodrun
(self, problem_count: int, alpha: float = 1.0, beta: float = 0.0)
fastertransformer/3rdparty/cutlass/tools/library/scripts/pycutlass/src/pycutlass/test/gemm_grouped_testbed.py:108
↓ 5 callersMethodrun
fastertransformer/src/fastertransformer/kernels/llama/flash_attention2/utils.h:206
↓ 5 callersMethodrun_moe_fc
fastertransformer/src/fastertransformer/kernels/moe_kernels.h:220
↓ 5 callersFunctionserialize
fastertransformer/src/fastertransformer/models/bert_fp8/serialize.hpp:13
↓ 5 callersFunctionserialize_d2h
fastertransformer/src/fastertransformer/models/bert_fp8/serialize.hpp:33
↓ 5 callersMethodsetMaxSeqLen
fastertransformer/src/fastertransformer/models/llama/LlamaWeight.h:73
↓ 5 callersMethodset_kgroup_index
Notify the iterator which k-group it is currently pointing to. This does not advance the iterator. Rather, it overrides its internal tracking with co
fastertransformer/3rdparty/cutlass/include/cutlass/gemm/warp/mma_tensor_op_tile_iterator_sm80.h:2439
↓ 5 callersMethodset_residual_tile
fastertransformer/3rdparty/cutlass/examples/42_fused_multi_head_attention/iterators/predicated_tile_access_iterator_residual_last.h:803
↓ 5 callersMethodsize
fastertransformer/src/fastertransformer/utils/Tensor.h:325
↓ 5 callersFunctionsmooth_fc_fc_inplace
(fc1, fc2, scales,shifts=None, self_attn=None)
algorithm/models/transformation.py:105
↓ 5 callersMethodstart
engine/common/timer.h:36
↓ 5 callersMethodstart
(self, tag)
fastertransformer/examples/pytorch/gpt/utils/profiler.py:10
↓ 5 callersMethodstop
(self, tag)
fastertransformer/examples/pytorch/gpt/utils/profiler.py:13
↓ 5 callersFunctionstore_with_byte_offset
Stores a fragment to memory
fastertransformer/3rdparty/cutlass/examples/42_fused_multi_head_attention/iterators/epilogue_predicated_tile_iterator.h:412
↓ 5 callersMethodstore_with_pointer_offset
Store a fragment to memory
fastertransformer/3rdparty/cutlass/examples/42_fused_multi_head_attention/iterators/predicated_tile_iterator_residual_last.h:893
↓ 5 callersMethodstore_with_pointer_offset
Stores a fragment to memory at the location pointed to by the iterator
fastertransformer/3rdparty/cutlass/include/cutlass/gemm/warp/mma_tensor_op_tile_iterator_wmma.h:761
↓ 5 callersMethodstore_with_pointer_offset
Store a fragment to memory
fastertransformer/3rdparty/cutlass/include/cutlass/transform/threadblock/regular_tile_iterator_tensor_op_sm70.h:519
↓ 5 callersMethodstore_with_pointer_offset
Store a fragment to memory
fastertransformer/3rdparty/cutlass/include/cutlass/transform/threadblock/predicated_tile_iterator.h:798
↓ 5 callersMethodstore_with_pointer_offset
Stores a fragment
fastertransformer/3rdparty/cutlass/include/cutlass/transform/threadblock/regular_tile_iterator_pitch_linear.h:357
↓ 5 callersFunctionstrided_dgrad_tile_m_per_filter
/////////////////////////////////////////////////////////////////////////////////////////////// Strided dgrad helper functions
fastertransformer/3rdparty/cutlass/include/cutlass/conv/conv2d_problem_size.h:580
↓ 5 callersMethodto_half
(self)
fastertransformer/examples/pytorch/t5/utils/ft_decoding.py:413
↓ 5 callersMethodto_half
(self)
fastertransformer/examples/pytorch/t5/utils/ft_encoder.py:347
↓ 5 callersMethodtokenize
(self, text)
fastertransformer/3rdparty/Megatron-LM/megatron/tokenizer/tokenizer.py:96
↓ 5 callersMethodtokenize
Tokenizes a piece of text.
fastertransformer/examples/tensorflow/bert/tensorflow_bert/bert/tokenization.py:196
↓ 5 callersMethodtransposeWeight
fastertransformer/src/fastertransformer/models/gpt_fp8/GptFP8Weight.cc:222
↓ 5 callersMethodtranspose_for_scores
(self, x)
fastertransformer/examples/pytorch/vit/ViT-quantization/vit_int8.py:170
↓ 5 callersMethodtransposed_problem
Returns arguments for the transposed problem
fastertransformer/3rdparty/cutlass/include/cutlass/gemm/kernel/gemm_universal.h:229
↓ 5 callersMethodunRegisterCallback
fastertransformer/src/fastertransformer/triton_backend/transformer_triton_backend.hpp:279
↓ 5 callersMethodupdate
(self, val, n=1)
fastertransformer/examples/pytorch/vit/ViT-quantization/eval_engine.py:59
↓ 5 callersMethodwarn
(self, s)
fastertransformer/examples/pytorch/vit/ViT-quantization/quant_utils.py:32
↓ 4 callersFunctionCreateConv2dFixedChannelsOperator
(manifest, layout, tile_descriptions, data_type, channel_counts, \ conv_kinds = [ConvKind.Fprop, ConvKind.Dg
fastertransformer/3rdparty/cutlass/tools/library/scripts/generator.py:432
↓ 4 callersFunctionEncodeOperator
fastertransformer/3rdparty/cutlass/test/unit/conv/device/cache_testbed_output.h:255
↓ 4 callersMethodGetSerializeSize
fastertransformer/src/fastertransformer/models/vit/ViTWeight.h:156
↓ 4 callersFunctionOffsetBytes
fastertransformer/3rdparty/cutlass/include/cutlass/fast_math.h:548
↓ 4 callersFunctionPackDevicePointer
(host_workspace, offset, value)
fastertransformer/3rdparty/cutlass/tools/library/scripts/rt.py:74
↓ 4 callersFunctionPackInteger
(host_workspace, offset, value)
fastertransformer/3rdparty/cutlass/tools/library/scripts/rt.py:67
↓ 4 callersFunctionTensorFillSymmetricRandomGaussian
< Layout function
fastertransformer/3rdparty/cutlass/tools/util/include/cutlass/util/reference/host/tensor_fill.h:431
↓ 4 callersFunctionTensorFillSymmetricRandomUniform
< Layout function
fastertransformer/3rdparty/cutlass/tools/util/include/cutlass/util/reference/host/tensor_fill.h:844
↓ 4 callersFunctionTensorRelativeErrorMetric
fastertransformer/3rdparty/cutlass/tools/util/include/cutlass/util/reference/host/error_metrics.h:51
↓ 4 callersMethod__init__
(self, optimizer, last_epoch=-1)
fastertransformer/examples/pytorch/bert/bert-quantization-sparsity/schedulers.py:22
↓ 4 callersMethod__iter__
(self)
fastertransformer/3rdparty/Megatron-LM/megatron/data/data_samplers.py:94
↓ 4 callersFunction_cpp_extention_load_helper
(name, sources, extra_cuda_flags)
fastertransformer/3rdparty/Megatron-LM/megatron/fused_kernels/__init__.py:46
↓ 4 callersFunction_ensure_var_is_initialized
Make sure the input variable is not None.
fastertransformer/3rdparty/Megatron-LM/megatron/global_vars.py:186
↓ 4 callersFunction_get_best_indices
Get the n-best logits from a list.
fastertransformer/examples/pytorch/bert/bert-quantization-sparsity/run_squad.py:683
↓ 4 callersMethod_initialize_model
(self, force_init=False)
fastertransformer/examples/pytorch/gpt/utils/gpt_decoder.py:338
↓ 4 callersMethod_is_number
(self, text)
algorithm/lm_eval/tasks/drop.py:245
↓ 4 callersMethod_map
(self, func)
fastertransformer/examples/pytorch/llama2/llamav2.py:113
↓ 4 callersMethod_map
(self, func)
fastertransformer/examples/pytorch/baichuan2/baichuanv2.py:112
↓ 4 callersMethod_map_int8
(self, func)
fastertransformer/examples/pytorch/gpt/utils/gpt_decoder.py:188
↓ 4 callersMethod_mask_padded_vocab_weights
(self, weight: torch.Tensor)
fastertransformer/examples/pytorch/gpt/utils/gpt_decoder.py:974
↓ 4 callersFunction_transpose_first_dim
(t, num_splits, num_splits_first, model)
fastertransformer/3rdparty/Megatron-LM/megatron/checkpointing.py:230
↓ 4 callersFunction_zero_grad_group_helper
Zero out the gradient for a group of parameters. Note: copied from torch.optim.optimizer.
fastertransformer/3rdparty/Megatron-LM/megatron/optimizer/optimizer.py:33
↓ 4 callersMethodaccum_ref
fastertransformer/3rdparty/cutlass/examples/42_fused_multi_head_attention/mma_from_smem.h:112
↓ 4 callersFunctionadd_byte_offset_
Adds a pointer offset in units of element
fastertransformer/3rdparty/cutlass/include/cutlass/conv/threadblock/conv2d_dgrad_output_gradient_tile_access_iterator_optimized.h:673
↓ 4 callersFunctionadd_pointer_offset
Adds a pointer offset in units of Element
fastertransformer/3rdparty/cutlass/include/cutlass/transform/threadblock/regular_tile_access_iterator_tensor_op_sm80.h:163
↓ 4 callersMethodany
fastertransformer/src/fastertransformer/utils/Tensor.h:258
↓ 4 callersMethodat
Accesses a bit within the predicate vector.
fastertransformer/3rdparty/cutlass/include/cutlass/predicate_vector.h:455
↓ 4 callersMethodback
fastertransformer/3rdparty/cutlass/include/cutlass/array.h:396
↓ 4 callersMethodbench_helper
(self, act_type, quant_type, rtol, atol)
fastertransformer/tests/gemm_dequantize/th_gemm_dequantize.py:171
↓ 4 callersMethodbench_helper
(self, act_type, quant_type, rtol, atol)
fastertransformer/tests/gemm_dequantize/single_test.py:174
↓ 4 callersFunctionbert_extended_attention_mask
(attention_mask)
fastertransformer/3rdparty/Megatron-LM/megatron/model/bert_model.py:32
↓ 4 callersFunctionbg
(e)
fastertransformer/3rdparty/cutlass/docs/jquery.js:23
↓ 4 callersFunctionbuild_data_loader
Data loader. Note that batch-size is the local (per GPU) batch-size.
fastertransformer/3rdparty/Megatron-LM/tasks/finetune_utils.py:84
↓ 4 callersFunctionbuild_optimizer
Build optimizer, set weight decay of normalization to 0 by default.
fastertransformer/examples/pytorch/swin/Swin-Transformer-Quantization/SwinTransformer/optimizer.py:11
↓ 4 callersFunctionbv
(bR,bW,bV,bZ,bX,bY)
fastertransformer/3rdparty/cutlass/docs/jquery.js:23
↓ 4 callersMethodcan_implement
Determines whether kernel satisfies alignment
fastertransformer/3rdparty/cutlass/examples/13_two_tensor_op_fusion/kernel/b2b_gemm.h:169
↓ 4 callersFunctioncastToFloat
fastertransformer/examples/cpp/xlnet/xlnet_correctness_example.cc:62
↓ 4 callersFunctioncast_cudnn_compute_type_to_float
Cudnn compute type seems to be hardcoded to float (To handle a possible cudnn issue)
fastertransformer/3rdparty/cutlass/tools/profiler/src/cudnn_helpers.cpp:205
↓ 4 callersFunctioncheck_quant_type_allowed
fastertransformer/src/fastertransformer/th_op/common/WeightOnlyQuantOps.cc:30
↓ 4 callersMethodchunks
(data, size=len(self.articles))
fastertransformer/examples/pytorch/bert/bert-quantization-sparsity/data/TextSharding.py:71
↓ 4 callersMethodclear
< Efficiently disables all accesses guarded by mask
fastertransformer/3rdparty/cutlass/include/cutlass/epilogue/threadblock/predicated_tile_iterator_v2.h:148
↓ 4 callersFunctionclear_mask
Clears the predicates
fastertransformer/3rdparty/cutlass/include/cutlass/conv/threadblock/conv2d_dgrad_output_gradient_tile_access_iterator_optimized.h:386
↓ 4 callersFunctioncode_to_language
(code)
algorithm/lm_eval/tasks/translation.py:241
↓ 4 callersMethodcompute_predicates_
Computes predicates based on internally tracked per-thread offset.
fastertransformer/3rdparty/cutlass/include/cutlass/transform/threadblock/predicated_tile_access_iterator.h:132
↓ 4 callersMethodconcat_and_pad_tokens
Concat with special tokens and pad sequence to self.max_seq_length
fastertransformer/3rdparty/Megatron-LM/megatron/data/ict_dataset.py:142
↓ 4 callersFunctionconvert_by_vocab
Converts a sequence of [tokens|ids] using the vocab.
fastertransformer/3rdparty/Megatron-LM/megatron/tokenizer/bert_tokenization.py:136
↓ 4 callersFunctionconvert_by_vocab
Converts a sequence of [tokens|ids] using the vocab.
fastertransformer/examples/tensorflow/bert/tensorflow_bert/bert/tokenization.py:136
↓ 4 callersFunctionconvert_layout_acc_rowcol
fastertransformer/src/fastertransformer/kernels/llama/flash_attention2/utils.h:230
↓ 4 callersFunctionconvert_megatron_to_HF_naming_style_multiple
(saved_key, name_mapping)
fastertransformer/examples/pytorch/t5/utils/megatron_t5_ckpt_convert.py:112
↓ 4 callersFunctioncopy
fastertransformer/3rdparty/cutlass/tools/util/include/cutlass/util/device_memory.h:86
↓ 4 callersMethodcopy_to_host
fastertransformer/3rdparty/cutlass/tools/util/include/cutlass/util/device_memory.h:315
↓ 4 callersFunctioncreate_logger
(output_dir, dist_rank=0, name='')
fastertransformer/examples/pytorch/swin/Swin-Transformer-Quantization/SwinTransformer/logger.py:16
↓ 4 callersFunctioncreate_record_worker
(filename_prefix, shard_id, output_format='tfrecord')
fastertransformer/examples/pytorch/bert/bert-quantization-sparsity/data/bertPrep.py:149
↓ 4 callersFunctioncross_check
(name, tf_val, op_val, atol_threshold)
fastertransformer/examples/tensorflow/bert/utils/common.py:90
↓ 4 callersMethodcuFuncSetAttribute
fastertransformer/3rdparty/common/cudaDriverWrapper.cpp:76
↓ 4 callersMethodcuModuleGetFunction
fastertransformer/3rdparty/common/cudaDriverWrapper.cpp:107
↓ 4 callersMethodcuModuleLoadData
fastertransformer/3rdparty/common/cudaDriverWrapper.cpp:96
↓ 4 callersFunctioncustomMatmulRun
fastertransformer/src/fastertransformer/utils/gemm_test/encoder_igemm_func.cc:205
↓ 4 callersFunctiondeallocate_output_tensor
Pseudo-deallocate (i.e., set to scalar) the output tensor's '.data' field. This method should be called right after the output tensor has been
fastertransformer/3rdparty/Megatron-LM/megatron/schedules.py:46
← previousnext →801–900 of 14,558, ranked by callers