MCPcopy Create free account

hub / github.com/MLSysU/TD-Pipe / functions

Functions572 in github.com/MLSysU/TD-Pipe

⨍ Functions 572◇ Types & classes 128↳ Endpoints 6
↓ 2 callersMethodschedule
(self)
TD_Pipe/core/scheduler.py:416
↓ 2 callersFunctionset_cuda_visible_devices
(device_ids: List[int])
TD_Pipe/utils.py:70
↓ 2 callersFunctionset_random_seed
(seed: int)
TD_Pipe/model_executor/utils.py:9
↓ 2 callersMethodverify_with_parallel_config
( self, parallel_config: "ParallelConfig", )
TD_Pipe/config.py:170
↓ 1 callersMethod__init__
(self, num_embeddings: int, embedding_dim: int, bias: bool
TD_Pipe/model_executor/layers/vocab_parallel_embedding.py:119
↓ 1 callersMethod__init__
(self, request_id: str)
TD_Pipe/engine/async_llm_engine.py:46
↓ 1 callersMethod_allocate
(self, seq_group: SequenceGroup)
TD_Pipe/core/scheduler.py:458
↓ 1 callersMethod_append_slot
( self, seq_group: SequenceGroup, blocks_to_copy: Dict[int, List[int]], )
TD_Pipe/core/scheduler.py:463
↓ 1 callersFunction_apply_logits_processors
( logits: torch.Tensor, sampling_metadata: SamplingMetadata, )
TD_Pipe/model_executor/layers/sampler.py:105
↓ 1 callersFunction_apply_min_p
Adapted from https://github.com/oobabooga/text-generation-webui/blob/3146124ec01f02c8fb1650a6517cf1b60b537aaf/modules/sampler_hijack.py#L16C1
TD_Pipe/model_executor/layers/sampler.py:184
↓ 1 callersFunction_apply_penalties
(logits: torch.Tensor, prompt_tokens_tensor: torch.Tensor, output_tokens_tensor: torch.Te
TD_Pipe/model_executor/layers/sampler.py:129
↓ 1 callersFunction_apply_top_p_top_k
( logits: torch.Tensor, p: torch.Tensor, k: torch.Tensor, )
TD_Pipe/model_executor/layers/sampler.py:152
↓ 1 callersFunction_beam_search_sample
( selected_seq_groups: List[Tuple[List[int], SamplingParams]], is_prompts: List[bool], seq_data: D
TD_Pipe/model_executor/layers/sampler.py:247
↓ 1 callersFunction_build_sampler_output
( sample_results: List[Tuple[List[int], List[int]]], sampling_metadata: SamplingMetadata, prompt_l
TD_Pipe/model_executor/layers/sampler.py:503
↓ 1 callersMethod_check_beam_search_early_stopping
( self, early_stopping: Union[bool, str], sampling_params: SamplingParams, bes
TD_Pipe/engine/llm_engine.py:391
↓ 1 callersFunction_check_if_gpu_supports_dtype
(torch_dtype: torch.dtype)
TD_Pipe/worker/worker.py:216
↓ 1 callersMethod_check_stop
Stop the finished sequences.
TD_Pipe/engine/llm_engine.py:833
↓ 1 callersFunction_check_use_ref_attention
()
TD_Pipe/model_executor/layers/attention/backends/xformers.py:248
↓ 1 callersMethod_compute_cos_sin_cache
Compute the cos and sin cache.
TD_Pipe/model_executor/layers/rotary_embedding.py:84
↓ 1 callersMethod_compute_inv_freq
(self, scaling_factor: float)
TD_Pipe/model_executor/layers/rotary_embedding.py:299
↓ 1 callersMethod_decode_sequence
Decodes the new token for a sequence.
TD_Pipe/engine/llm_engine.py:813
↓ 1 callersMethod_engine_abort
(self, request_ids: Iterable[str])
TD_Pipe/engine/async_llm_engine.py:348
↓ 1 callersFunction_get_and_verify_dtype
( config: PretrainedConfig, dtype: Union[str, torch.dtype], )
TD_Pipe/config.py:402
↓ 1 callersFunction_get_and_verify_max_len
Get and verify the model's maximum length.
TD_Pipe/config.py:453
↓ 1 callersMethod_get_compute_intensity
(self, batch_size)
TD_Pipe/core/scheduler.py:264
↓ 1 callersMethod_get_current_batch_requests
(self, total_requests)
TD_Pipe/core/scheduler.py:441
↓ 1 callersMethod_get_decode_batch_time
(self, batch_size)
TD_Pipe/core/scheduler.py:277
↓ 1 callersFunction_get_dtype_size
(dtype: torch.dtype)
TD_Pipe/worker/cache_engine.py:173
↓ 1 callersFunction_get_logits
(hidden_states: torch.Tensor, embedding: torch.Tensor, embedding_bias: Optional[torch.Tensor],
TD_Pipe/model_executor/layers/projection.py:45
↓ 1 callersFunction_get_logprobs
( logprobs: torch.Tensor, sampling_metadata: SamplingMetadata, sample_results: List[Tuple[List[int
TD_Pipe/model_executor/layers/sampler.py:389
↓ 1 callersFunction_get_model_architecture
(config: PretrainedConfig)
TD_Pipe/model_executor/model_loader.py:24
↓ 1 callersFunction_greedy_sample
( selected_seq_groups: List[Tuple[List[int], SamplingParams]], samples: torch.Tensor, )
TD_Pipe/model_executor/layers/sampler.py:201
↓ 1 callersMethod_init_cache
Profiles the memory usage and initializes the KV cache.
TD_Pipe/engine/llm_engine.py:272
↓ 1 callersFunction_init_distributed_environment
Initialize the distributed environment.
TD_Pipe/worker/worker.py:185
↓ 1 callersMethod_init_engine
(self, *args, **kwargs)
TD_Pipe/engine/async_llm_engine.py:298
↓ 1 callersMethod_init_prefill_data
(self)
TD_Pipe/core/scheduler.py:117
↓ 1 callersMethod_init_workers
(self)
TD_Pipe/engine/llm_engine.py:139
↓ 1 callersMethod_init_workers_ray
(self, placement_group: "PlacementGroup", **ray_remote_kwargs)
TD_Pipe/engine/llm_engine.py:161
↓ 1 callersMethod_log_system_stats
( self, prompt_run: bool, num_batched_tokens: int, )
TD_Pipe/engine/llm_engine.py:740
↓ 1 callersFunction_make_alibi_bias
( alibi_slopes: torch.Tensor, num_kv_heads: int, dtype: torch.dtype, input_metadata: InputMeta
TD_Pipe/model_executor/layers/attention/backends/xformers.py:214
↓ 1 callersFunction_make_tensor_with_pad
( x: List[List[int]], max_len: int, pad: int, dtype: torch.dtype, device: Union[str, torch
TD_Pipe/worker/model_runner.py:435
↓ 1 callersFunction_multinomial
( probs: torch.Tensor, num_samples: int, )
TD_Pipe/model_executor/layers/sampler.py:305
↓ 1 callersFunction_pad_to_max
(x: List[int], max_len: int, pad: int)
TD_Pipe/worker/model_runner.py:430
↓ 1 callersFunction_partition_pipeline_model
(model_config: ModelConfig, parallel_config: ParallelConfig,
TD_Pipe/worker/worker.py:228
↓ 1 callersMethod_preempt_by_swap
( self, seq_group: SequenceGroup, blocks_to_swap_out: Dict[int, int], )
TD_Pipe/core/scheduler.py:519
↓ 1 callersMethod_prepare_decode
( self, seq_group_metadata_list: List[SequenceGroupMetadata], )
TD_Pipe/worker/model_runner.py:144
↓ 1 callersMethod_prepare_prompt
( self, seq_group_metadata_list: List[SequenceGroupMetadata], )
TD_Pipe/worker/model_runner.py:59
↓ 1 callersMethod_prepare_sample
( self, seq_group_metadata_list: List[SequenceGroupMetadata], prompt_lens: List[int],
TD_Pipe/worker/model_runner.py:266
↓ 1 callersMethod_process_sequence_group_outputs
(self, seq_group: SequenceGroup, outputs: SequenceGroupOutput)
TD_Pipe/engine/llm_engine.py:435
↓ 1 callersFunction_prune_hidden_states
( hidden_states: torch.Tensor, sampling_metadata: SamplingMetadata, is_prompt: bool, )
TD_Pipe/model_executor/layers/projection.py:59
↓ 1 callersFunction_random_sample
( selected_seq_groups: List[Tuple[List[int], SamplingParams]], is_prompts: List[bool], random_samp
TD_Pipe/model_executor/layers/sampler.py:220
↓ 1 callersFunction_ref_masked_attention
( query: torch.Tensor, key: torch.Tensor, value: torch.Tensor, num_heads: int, num_kv_head
TD_Pipe/model_executor/layers/attention/backends/xformers.py:256
↓ 1 callersMethod_run_memory_efficient_xformer_forward
Attention for 1D query of multiple prompts. Multiple prompt tokens are flattened in to `query` input. Args: output: shape
TD_Pipe/model_executor/layers/attention/backends/xformers.py:143
↓ 1 callersFunction_sample
( probs: torch.Tensor, logprobs: torch.Tensor, sampling_metadata: SamplingMetadata, )
TD_Pipe/model_executor/layers/sampler.py:322
↓ 1 callersMethod_schedule
Sched: Three states: waiting, running, ready Beam search is not considered; each seq_group contains only one seq
TD_Pipe/core/scheduler.py:280
↓ 1 callersFunction_set_default_torch_dtype
Sets the default torch dtype to the given dtype.
TD_Pipe/model_executor/model_loader.py:16
↓ 1 callersFunction_setup_logger
()
TD_Pipe/logger.py:29
↓ 1 callersFunction_shared_pointers
(tensors)
TD_Pipe/model_executor/weight_utils.py:36
↓ 1 callersMethod_swap_out
(self, seq_group: SequenceGroup, blocks_to_swap_out: Dict[int, int],
TD_Pipe/core/scheduler.py:537
↓ 1 callersMethod_validate_config
(self)
TD_Pipe/transformers_utils/configs/mpt.py:165
↓ 1 callersMethod_verify_args
(self)
TD_Pipe/config.py:285
↓ 1 callersMethod_verify_args
(self)
TD_Pipe/config.py:375
↓ 1 callersMethod_verify_args
(self)
TD_Pipe/sampling_params.py:160
↓ 1 callersMethod_verify_args
(self)
TD_Pipe/engine/llm_engine.py:268
↓ 1 callersMethod_verify_beam_search
(self)
TD_Pipe/sampling_params.py:196
↓ 1 callersMethod_verify_greedy_sampling
(self)
TD_Pipe/sampling_params.py:221
↓ 1 callersMethod_verify_load_format
(self)
TD_Pipe/config.py:101
↓ 1 callersMethod_verify_non_beam_search
(self)
TD_Pipe/sampling_params.py:211
↓ 1 callersMethod_verify_quantization
(self)
TD_Pipe/config.py:137
↓ 1 callersMethod_verify_tokenizer_mode
(self)
TD_Pipe/config.py:129
↓ 1 callersFunction_yarn_find_correction_range
(low_rot: int, high_rot: int, dim: int,
TD_Pipe/model_executor/layers/rotary_embedding.py:237
↓ 1 callersFunction_yarn_get_mscale
(scale: float = 1)
TD_Pipe/model_executor/layers/rotary_embedding.py:262
↓ 1 callersFunction_yarn_linear_ramp_mask
(low: float, high: float, dim: int, dtype: torch.dtype,
TD_Pipe/model_executor/layers/rotary_embedding.py:250
↓ 1 callersMethodabort_seq_group
Aborts a sequence group with the given ID. Check if the sequence group with the given ID is present in any of the state queue.
TD_Pipe/core/scheduler.py:139
↓ 1 callersFunctionadd_global_metrics_labels
(**kwargs)
TD_Pipe/engine/metrics.py:32
↓ 1 callersMethodadd_request
Add a request to the engine's request pool. The request is added to the request pool and will be processed by the scheduler as `engin
TD_Pipe/engine/llm_engine.py:326
↓ 1 callersMethodadd_request
( self, request_id: str, prompt: Optional[str], sampling_params: SamplingParam
TD_Pipe/engine/async_llm_engine.py:363
↓ 1 callersMethodadd_seq_group
(self, seq_group: SequenceGroup)
TD_Pipe/core/scheduler.py:135
↓ 1 callersMethodappend_slot
Allocate a physical slot for a new token.
TD_Pipe/core/block_manager.py:150
↓ 1 callersMethodappend_tokens
(self, token_ids: List[int])
TD_Pipe/block.py:36
↓ 1 callersMethodcache_swap
( self, blocks_to_swap_in: Dict[int, int], blocks_to_swap_out: Dict[int, int],
TD_Pipe/worker/worker.py:132
↓ 1 callersMethodcan_allocate
(self, seq_group: SequenceGroup)
TD_Pipe/core/block_manager.py:103
↓ 1 callersMethodcan_swap_out
(self, seq_group: SequenceGroup)
TD_Pipe/core/block_manager.py:236
↓ 1 callersFunctioncompletion_full_generator
()
TD_Pipe/entrypoints/openai/api_server.py:382
↓ 1 callersFunctiondetokenize_incrementally
( tokenizer: Union[PreTrainedTokenizer, PreTrainedTokenizerFast], all_input_ids: List[int], prev_t
TD_Pipe/transformers_utils/tokenizer.py:107
↓ 1 callersMethodengine_step
Kick the engine to process the waiting requests. Returns True if there are in-progress requests.
TD_Pipe/engine/async_llm_engine.py:317
↓ 1 callersFunctionensure_divisibility
Ensure that numerator is divisible by the denominator.
TD_Pipe/model_executor/parallel_utils/utils.py:10
↓ 1 callersMethodexecute_model
( self, seq_group_metadata_list: Optional[List[SequenceGroupMetadata]], kv_caches: Lis
TD_Pipe/worker/model_runner.py:355
↓ 1 callersFunctionfake_stream_generator
()
TD_Pipe/entrypoints/openai/api_server.py:706
↓ 1 callersMethodformat
(self, record)
TD_Pipe/logger.py:17
↓ 1 callersMethodforward
( self, input_ids: torch.Tensor, positions: torch.Tensor, kv_caches: List[KVCa
TD_Pipe/model_executor/models/opt.py:298
↓ 1 callersMethodforward_decode
( query: torch.Tensor, key_cache: torch.Tensor, value_cache: torch.Tensor, inp
TD_Pipe/model_executor/layers/attention/ops/paged_attn.py:35
↓ 1 callersMethodfree_finished_seq_groups
(self)
TD_Pipe/core/scheduler.py:452
↓ 1 callersMethodfrom_lists
(cls, temperatures: List[float], top_ps: List[float], top_ks: List[int], min_ps: List[float
TD_Pipe/model_executor/sampling_metadata.py:144
↓ 1 callersMethodfrom_sampling_metadata
( cls, sampling_metadata: "SamplingMetadata", vocab_size: int, device: torch.device,
TD_Pipe/model_executor/sampling_metadata.py:71
↓ 1 callersMethodfrom_seq_group
(cls, seq_group: SequenceGroup)
TD_Pipe/outputs.py:78
↓ 1 callersFunctionget_act_fn
Get an activation function by name.
TD_Pipe/model_executor/layers/activation.py:120
↓ 1 callersMethodget_block_table
(self, seq: Sequence)
TD_Pipe/core/block_manager.py:285
↓ 1 callersMethodget_cache_block_size
( block_size: int, model_config: ModelConfig, parallel_config: ParallelConfig, )
TD_Pipe/worker/cache_engine.py:157
↓ 1 callersFunctionget_config
(model: str, trust_remote_code: bool, revision: Optional[str] = None)
TD_Pipe/transformers_utils/config.py:19
← previousnext →101–200 of 572, ranked by callers