Functions — github.com/EverMind-AI/MSA

hub / github.com/EverMind-AI/MSA / functions

Functions265 in github.com/EverMind-AI/MSA

⨍ Functions 265◇ Types & classes 54

Methodget_bench_files

(self)

src/benchmarks.py:52

Methodget_current_usage

获取当前显存使用量（不停止监控）

src/utils/gpu_monitor.py:194

Functionget_data

(index)

src/utils/data_utils.py:72

Methodget_decoder

(self)

src/msa/model.py:329

Functionget_eval_response

(prompt)

src/evaluation/llm_judge.py:83

Methodget_kvcache

获取指定层的KV缓存（反量化后） Args: layer_idx: 层索引 Returns: Tuple[torch.Tensor, torch.Tensor]: (key_cache, value

src/utils/cache.py:208

Methodget_layer_length

(self)

src/utils/cache.py:43

Methodget_layer_length

(self)

src/utils/cache.py:199

Functionget_length

(index)

src/utils/data_utils.py:58

Functionget_local_rank

()

src/utils/misc.py:42

Methodget_padded_gather_list

(tensor, target_seqlen)

src/msa_service.py:1114

Methodget_peak_usage

获取当前记录的峰值显存使用量（不停止监控）

src/utils/gpu_monitor.py:198

Methodget_seq_length

返回缓存状态的序列长度 Args: layer_idx: 可选的层索引 Returns: int: 序列长度

src/utils/cache.py:282

Functionget_sequence_length

(dataset, num_worker=16)

src/utils/data_utils.py:61

Functionget_sha

()

src/utils/misc.py:51

Methodget_template_prefix_kvcaches

(self, layer_idx: int)

src/msa_service.py:261

Methodgpu_select

(self, scores, k, rk, v, pooled_doc_ids, total_docs)

src/msa_service.py:995

Methodis_running

检查监控是否在运行

src/utils/gpu_monitor.py:203

Functionis_transformers_version_equal_to_4_46

()

src/utils/data_utils.py:103

Functionis_transformers_version_greater_than

(content: str)

src/utils/data_utils.py:99

Functionmanual_deepcopy_kv_cache

Manually performs a deep copy of a custom KV cache object, avoiding the issues with quanto's __deepcopy__.

src/utils/cache.py:353

Methodmap_tensor_to_group_ids

(a: torch.Tensor)

src/msa/memory_sparse_attention.py:100

Methodmemory_path

(self)

src/benchmarks.py:49

Functionmonitor_gpu_memory_example

使用示例

src/utils/gpu_monitor.py:218

Methodnext_bucket

()

src/msa_service.py:1467

Methodnr_docs

(self)

src/msa_service.py:230

Methodon_log

(self, args, state, control, logs=None, **kwargs)

src/utils/callbacks.py:38

Methodon_log

(self, args, state, control, logs=None, **kwargs)

src/utils/callbacks.py:74

Methodon_step_begin

(self, args, state, control, **kwargs)

src/utils/callbacks.py:30

Methodon_step_begin

(self, args, state, control, **kwargs)

src/utils/callbacks.py:64

Methodon_step_end

(self, args, state, control, **kwargs)

src/utils/callbacks.py:34

Methodon_step_end

(self, args, state, control, **kwargs)

src/utils/callbacks.py:69

Methodon_step_end

(self, args, state, control, **kwargs)

src/utils/callbacks.py:109

Methodon_step_end

(self, args, state, control, **kwargs)

src/utils/callbacks.py:156

Methodon_step_end

(self, args, state, control, **kwargs)

src/utils/callbacks.py:171

Methodon_train_begin

(self, args, state, control, **kwargs)

src/utils/callbacks.py:27

Methodon_train_begin

(self, args, state, control, **kwargs)

src/utils/callbacks.py:60

Functionpad_sequence

(examples, cutoff_len, tokenizer)

src/utils/data_utils.py:177

Functionparse_match_result

(text)

src/evaluation/llm_judge.py:10

Functionpatch_cosine_with_warmup_schedule

(minimal_lr=0.0)

src/utils/misc.py:71

Functionpatch_torch_clip_grad_norm

()

src/utils/misc.py:174

Functionpatch_transformer_logging

()

src/utils/common.py:90

Methodprefill_stage2_1

(self, layer_idx: int, query_states: torch.Tensor, query_mask: torch.Tensor)

src/msa_service.py:830

Methodprefill_worker_main

(gpu_id: int, request_queue: mp.Queue, response_queue: mp.Queue, model_path: str,

src/prefill.py:112

Functionpreprocess_packed_supervised_dataset

(examples, tokenizer, cutoff_len)

src/utils/data_utils.py:136

Methodprint

(self)

src/utils/tools.py:110

Functionprint_trainable_params

(model: torch.nn.Module)

src/utils/common.py:14

Methodquery_path

(self)

src/benchmarks.py:45

Functionquick_monitor

快速启动监控的便捷函数 Args: gpu_index: GPU索引 interval: 监控间隔 unit: 单位 wait_for_input: 是否等待用户输入 Returns:

src/utils/gpu_monitor.py:310

Functionread_lmdb

(lmdb_path)

src/utils/data_utils.py:44

Methodreceive_request

(self)

src/msa_service.py:1579

Methodreceive_response

(self)

src/msa_service.py:1772

Methodrecord_kwargs

记录层的元数据信息 Args: layer_idx: 层索引 kwargs: 包含路由层信息的字典

src/utils/cache.py:186

Methodservice_main

(gpu_id: int, request_queue: mp.Queue, response_queue: mp.Queue,

src/msa_service.py:1591

Methodset_decoder

(self, decoder)

src/msa/model.py:326

Methodset_input_embeddings

(self, value)

src/msa/model.py:317

Methodset_output_embeddings

(self, new_embeddings)

src/msa/model.py:323

Functionsmart_tokenizer_and_embedding_resize

Resize tokenizer and embedding. Note: This is the unoptimized version that may make your embedding size not be divisible by 64.

src/utils/common.py:68

Functionsp_split

(examples, sequence_parallel_size, sequence_parallel_mode="ulysses")

src/utils/data_utils.py:222

Methodstop_all_and_get_max

停止所有监控并返回最大峰值

src/utils/gpu_monitor.py:299

Functiontorch_dataset_to_hf_dataset

(torch_dataset, num_worker=16)

src/utils/data_utils.py:77

Methodupdate

更新缓存 Args: key_states: 新的key状态 value_states: 新的value状态 layer_idx: 层索引 cache_kwargs:

src/utils/cache.py:222

Methodupdate_router_kcache

( self, key_states: torch.Tensor, layer_idx: int, )

src/utils/cache.py:165

Methodworld

(self)

src/config/memory_config.py:25

Functionwrite_lmdb

(output_dir, name)

src/utils/data_utils.py:21

← previous201–265 of 265, ranked by callers