# # Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved. # Copyright 2023 The vLLM team. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # This file is a part of the vllm-ascend project. # Adapted from vllm-project/vllm/vllm/worker/gpu_model_runner.py # import gc import os import time import types import weakref from contextlib import contextmanager, nullcontext from dataclasses import dataclass from typing import TYPE_CHECKING, Dict, List, Optional, Union import numpy as np import numpy.typing as npt import torch import torch._dynamo.cache_size import torch.distributed as dist import torch.nn as nn from torch.distributed import ReduceOp from vllm.attention import AttentionType, get_attn_backend from vllm.attention.layer import Attention from vllm.config import CompilationLevel, VllmConfig from vllm.distributed import get_tensor_model_parallel_world_size from vllm.distributed.parallel_state import get_dp_group, get_pp_group from vllm.forward_context import set_forward_context from vllm.inputs import INPUT_REGISTRY from vllm.logger import logger from vllm.model_executor.layers.fused_moe import FusedMoE from vllm.model_executor.layers.rotary_embedding import MRotaryEmbedding from vllm.model_executor.model_loader import get_model from vllm.multimodal import MULTIMODAL_REGISTRY from vllm.multimodal.inputs import MultiModalKwargs, PlaceholderRange from vllm.multimodal.utils import group_mm_inputs_by_modality from vllm.sampling_params import SamplingType from vllm.sequence import IntermediateTensors from vllm.utils import (STR_DTYPE_TO_TORCH_DTYPE, DeviceMemoryProfiler, LayerBlockType, LazyLoader, cdiv) from vllm.v1.core.encoder_cache_manager import compute_encoder_budget from vllm.v1.kv_cache_interface import (FullAttentionSpec, KVCacheConfig, KVCacheSpec) from vllm.v1.outputs import EMPTY_MODEL_RUNNER_OUTPUT, ModelRunnerOutput from vllm.v1.sample.metadata import SamplingMetadata from vllm.v1.sample.sampler import Sampler from vllm.v1.spec_decode.eagle import EagleProposer from vllm.v1.spec_decode.metadata import SpecDecodeMetadata from vllm.v1.spec_decode.ngram_proposer import NgramProposer from vllm.v1.spec_decode.utils import is_spec_decode_supported from vllm.v1.utils import bind_kv_cache from vllm.v1.worker.gpu_input_batch import CachedRequestState, InputBatch from vllm.v1.worker.lora_model_runner_mixin import LoRAModelRunnerMixin from vllm.v1.worker.utils import (gather_mm_placeholders, sanity_check_mm_encoder_outputs, scatter_mm_placeholders) from vllm_ascend.ascend_config import get_ascend_config from vllm_ascend.attention.attention import AttentionMaskBuilder from vllm_ascend.attention.attention_v1 import AscendAttentionState from vllm_ascend.attention.mla_v1 import CommonAttentionMetadata from vllm_ascend.platform import NPUPlatform from vllm_ascend.sample.rejection_sampler import AscendRejectionSampler from vllm_ascend.utils import ProfileExecuteDuration from vllm_ascend.worker.mtp_proposer_v1 import MtpProposer if TYPE_CHECKING: import xgrammar as xgr # type: ignore[import-untyped] from vllm.v1.core.sched.output import SchedulerOutput else: xgr = LazyLoader("xgr", globals(), "xgrammar") import vllm.envs as envs_vllm import vllm_ascend.envs as envs_ascend @dataclass class GraphCaptureContext: stream: torch.npu.Stream @contextmanager def graph_capture(device: torch.device): """ `graph_capture` is a context manager which should surround the code that is capturing the NPU graph. Its main purpose is to ensure that the some operations will be run after the graph is captured, before the graph is replayed. It returns a `GraphCaptureContext` object which contains the necessary data for the graph capture. Currently, it only contains the stream that the graph capture is running on. This stream is set to the current NPU stream when the context manager is entered and reset to the default stream when the context manager is exited. This is to ensure that the graph capture is running on a separate stream from the default stream, in order to explicitly distinguish the kernels to capture from other kernels possibly launched on background in the default stream. """ graph_capture_context = GraphCaptureContext( torch.npu.Stream(device=device)) stream = graph_capture_context.stream # we use nullcontext now maybe_ca_context = nullcontext() # ensure all initialization operations complete before attempting to # capture the graph on another stream curr_stream = torch.npu.current_stream() if curr_stream != stream: stream.wait_stream(curr_stream) with torch.npu.stream(stream), maybe_ca_context: yield graph_capture_context class NPUModelRunner(LoRAModelRunnerMixin): def __init__(self, vllm_config: VllmConfig, device: torch.device): self.vllm_config = vllm_config self.model_config = vllm_config.model_config self.cache_config = vllm_config.cache_config self.lora_config = vllm_config.lora_config self.scheduler_config = vllm_config.scheduler_config self.speculative_config = vllm_config.speculative_config ascend_config = get_ascend_config() if ascend_config.ascend_scheduler_config.enabled: self.chunked_prefill_enabled = self.scheduler_config.chunked_prefill_enabled else: self.chunked_prefill_enabled = True self.device = device self.is_multimodal_model = self.model_config.is_multimodal_model self.block_size = vllm_config.cache_config.block_size self.max_num_blocks_per_req = cdiv(self.model_config.max_model_len, self.block_size) self.max_num_tokens = self.scheduler_config.max_num_batched_tokens self.max_num_reqs = self.scheduler_config.max_num_seqs self.graph_block_tables = np.zeros( (self.vllm_config.scheduler_config.max_num_seqs, (self.model_config.max_model_len + self.block_size - 1) // self.block_size), dtype=np.int32) # Model-related. self.num_attn_layers = self.model_config.get_num_layers_by_block_type( vllm_config.parallel_config, LayerBlockType.attention) self.hidden_size = self.model_config.get_hidden_size() self.dtype = self.model_config.dtype cache_config = vllm_config.cache_config if cache_config.cache_dtype == "auto": self.kv_cache_dtype = self.dtype else: self.kv_cache_dtype = STR_DTYPE_TO_TORCH_DTYPE[ cache_config.cache_dtype] self.head_size = self.model_config.get_head_size() self.attn_backend = get_attn_backend( self.head_size, self.dtype, self.kv_cache_dtype, self.block_size, self.model_config.is_attention_free, use_mla=self.model_config.use_mla, ) if self.attn_backend is None: error_msg = ( f"Error with get_att_backend: {self.head_size=}, " f"{self.dtype=}, {self.kv_cache_dtype=}, {self.block_size=}, " f"{self.model_config.is_attention_free=}, " f"{self.model_config.use_mla=}") logger.error(error_msg) raise NotImplementedError( "Non-Attention backend is not supported by V1 NPUModelRunner.") self.attn_metadata_builder = self.attn_backend.get_builder_cls()( weakref.proxy(self)) # Multi-modal data support self.input_registry = INPUT_REGISTRY self.mm_registry = MULTIMODAL_REGISTRY self.uses_mrope = self.model_config.uses_mrope self.max_num_encoder_input_tokens, self.encoder_cache_size = compute_encoder_budget( model_config=self.model_config, scheduler_config=self.scheduler_config, mm_registry=self.mm_registry) # Lazy initialization # self.model: nn.Module # Set after load_model self.kv_caches: List[torch.Tensor] = [] # req_id -> (input_id -> encoder_output) self.encoder_cache: Dict[str, Dict[int, torch.Tensor]] = {} # Set up speculative decoding. self.use_spec_decode = False self.spec_attn_mask = None if self.speculative_config: self.use_spec_decode = True self.spec_attn_mask = torch.triu(torch.ones(2048, 2048, dtype=torch.bool), diagonal=1).to("npu") if get_pp_group().is_last_rank: if self.speculative_config.method == "ngram": self.drafter = NgramProposer(self.vllm_config) elif self.speculative_config.method == "eagle": self.drafter = EagleProposer(self.vllm_config, self.device) # type: ignore elif self.speculative_config.method == 'deepseek_mtp': self.drafter = MtpProposer(self.vllm_config, self) else: raise ValueError("Unknown speculative decoding method: " f"{self.speculative_config.method}") self.rejection_sampler = AscendRejectionSampler() # Request states. self.requests: Dict[str, CachedRequestState] = {} # Persistent batch. self.input_ids = torch.zeros(self.max_num_tokens, dtype=torch.int32, device=self.device) self.positions = torch.zeros(self.max_num_tokens, dtype=torch.int64, device=self.device) self.query_start_loc = torch.zeros(self.max_num_reqs + 1, dtype=torch.int32, device=self.device) self.seq_lens = torch.zeros(self.max_num_reqs, dtype=torch.int32, device=self.device) # None in the first PP rank. The rest are set after load_model. self.intermediate_tensors: Optional[IntermediateTensors] = None # Only relevant for models using M-RoPE (e.g, Qwen2-VL) if self.uses_mrope: # NOTE: `mrope_positions` is implemented with one additional dummy # position on purpose to make it non-contiguous so that it can work # with torch compile. # See detailed explanation in https://github.com/vllm-project/vllm/pull/12128#discussion_r1926431923 # NOTE: When M-RoPE is enabled, position ids are 3D regardless of # the modality of inputs. For text-only inputs, each dimension has # identical position IDs, making M-RoPE functionally equivalent to # 1D-RoPE. # See page 5 of https://arxiv.org/abs/2409.12191 self.mrope_positions = torch.zeros((3, self.max_num_tokens + 1), dtype=torch.int64, device=self.device) self.mrope_positions_cpu = torch.zeros( (3, self.max_num_tokens + 1), dtype=torch.int64, device="cpu", pin_memory=True) if self.is_multimodal_model: self.inputs_embeds = torch.zeros( (self.max_num_tokens, self.hidden_size), dtype=self.dtype, device=self.device) # OPTIMIZATION: Cache the tensors rather than creating them every step. self.arange_np: npt.NDArray[np.int32] = np.arange(max( self.max_num_reqs + 1, self.model_config.max_model_len, self.max_num_tokens), dtype=np.int32) # NOTE(woosuk): These tensors are "stateless", i.e., they are literally # a faster version of creating a new tensor every time. Thus, we should # not make any assumptions about the values in these tensors. self.input_ids_cpu = torch.zeros(self.max_num_tokens, dtype=torch.int32, device="cpu", pin_memory=True) self.positions_cpu = torch.zeros(self.max_num_tokens, dtype=torch.int64, device="cpu", pin_memory=True) self.positions_np = self.positions_cpu.numpy() self.slot_mapping_cpu = torch.zeros(self.max_num_tokens, dtype=torch.int32, device="cpu", pin_memory=True) self.slot_mapping_np = self.slot_mapping_cpu.numpy() self.query_start_loc_cpu = torch.zeros(self.max_num_reqs + 1, dtype=torch.int32, device="cpu", pin_memory=True) self.query_start_loc_np = self.query_start_loc_cpu.numpy() self.seq_lens_cpu = torch.zeros(self.max_num_reqs, dtype=torch.int32, device="cpu", pin_memory=True) self.seq_lens_np = self.seq_lens_cpu.numpy() self.input_positions_cpu = torch.arange(0, self.max_num_tokens, device="cpu") self.attn_mask = None self.attn_state = None self.use_aclgraph = (self.vllm_config.compilation_config.level == CompilationLevel.PIECEWISE and not self.model_config.enforce_eager) self.aclgraph_batch_sizes = list( reversed( self.vllm_config.compilation_config.cudagraph_capture_sizes)) # NOTE: Pre-construct a mask matrix to improve the efficiency of # attention mask construction during inference. # Note that the length of the matrix needs to be carefully balanced: a # matrix that is too large will consume excessive VRAM, while a matrix # that is too small will require dynamic concatenation during inference, # leading to performance degradation. # Therefore, an environment variable is added here to dynamically set # the size of the pre-constructed mask matrix based on requirements. mask_len = os.getenv("PAGED_ATTENTION_MASK_LEN", 10000) self.attn_mask_len = min(self.model_config.max_model_len, int(mask_len)) self.attn_mask_builder = AttentionMaskBuilder.initialize_from_len( self.attn_mask_len, self.dtype) self.sampler = Sampler() self.torchair_compiled_model = None # type: ignore self.torchair_compiled_models = {} # type: ignore ascend_config = get_ascend_config() self.torchair_graph_enabled = ascend_config.torchair_graph_config.enabled and self.vllm_config.model_config.use_mla self.use_cached_npu_graph = ascend_config.torchair_graph_config.use_cached_graph self.torchair_graph_batch_sizes = ascend_config.torchair_graph_config.graph_batch_sizes if ascend_config.torchair_graph_config.graph_batch_sizes_init: self.init_torchair_graph_batch_sizes() if len(self.torchair_graph_batch_sizes) == 0: # TODO(zzzzwwjj): check torchair_graph_batch_sizes init code self.torchair_graph_batch_sizes = [ self.scheduler_config.max_num_seqs ] torch._dynamo.cache_size.config.cache_size_limit += len( self.torchair_graph_batch_sizes) torch._dynamo.config.capture_dynamic_output_shape_ops = True torch._logging.set_logs( recompiles=envs_ascend.VLLM_ASCEND_TRACE_RECOMPILES) self.dp_size = vllm_config.parallel_config.data_parallel_size self.dp_rank = vllm_config.parallel_config.data_parallel_rank def _update_states(self, scheduler_output: "SchedulerOutput") -> None: """Update the cached states and the persistent batch with the scheduler output. The SamplingMetadata is updated and copied to the NPU if there is a new/resumed/paused/finished request in the batch. """ # Remove finished requests from the cached states. for req_id in scheduler_output.finished_req_ids: self.requests.pop(req_id, None) self.encoder_cache.pop(req_id, None) # Remove the finished requests from the persistent batch. # NOTE(woosuk): There could be an edge case where finished_req_ids and # scheduled_req_ids overlap. This happens when a request is aborted and # then resubmitted with the same ID. In this case, we treat them as two # distinct requests - clearing the cached states for the first request # and handling the second as a new request. removed_req_indices: List[int] = [] for req_id in scheduler_output.finished_req_ids: req_index = self.input_batch.remove_request(req_id) if req_index is not None: removed_req_indices.append(req_index) # Free the cached encoder outputs. for req_id, input_id in scheduler_output.free_encoder_input_ids: encoder_outputs = self.encoder_cache.get(req_id) if encoder_outputs is not None: encoder_outputs.pop(input_id, None) if not encoder_outputs: self.encoder_cache.pop(req_id, None) # Remove the unscheduled requests from the persistent batch. # NOTE(woosuk): The unscheduled requests are either preempted requests # or running requests that are not scheduled in this step. We remove # them from the persistent batch but keep their cached states since # they will be scheduled again sometime in the future. scheduled_req_ids = scheduler_output.num_scheduled_tokens.keys() cached_req_ids = self.input_batch.req_id_to_index.keys() unscheduled_req_ids = cached_req_ids - scheduled_req_ids # NOTE(woosuk): The persistent batch optimization assumes that # consecutive batches contain mostly the same requests. If batches # have low request overlap (e.g., alternating between two distinct # sets of requests), this optimization becomes very inefficient. for req_id in unscheduled_req_ids: req_index = self.input_batch.remove_request(req_id) assert req_index is not None removed_req_indices.append(req_index) req_ids_to_add: List[str] = [] # Add new requests to the cached states. for new_req_data in scheduler_output.scheduled_new_reqs: req_id = new_req_data.req_id sampling_params = new_req_data.sampling_params if sampling_params.sampling_type == SamplingType.RANDOM_SEED: generator = torch.Generator(device=self.device) generator.manual_seed(sampling_params.seed) else: generator = None self.requests[req_id] = CachedRequestState( req_id=req_id, prompt_token_ids=new_req_data.prompt_token_ids, mm_inputs=new_req_data.mm_inputs, mm_positions=new_req_data.mm_positions, sampling_params=sampling_params, generator=generator, block_ids=new_req_data.block_ids, num_computed_tokens=new_req_data.num_computed_tokens, output_token_ids=[], lora_request=new_req_data.lora_request, ) # Only relevant for models using M-RoPE (e.g, Qwen2-VL) if self.uses_mrope: image_grid_thw = [] video_grid_thw = [] second_per_grid_ts = [] audio_feature_lengths = [] use_audio_in_video = False for mm_input in self.requests[req_id].mm_inputs: if mm_input.get("image_grid_thw") is not None: image_grid_thw.extend( mm_input["image_grid_thw"].tolist()) if mm_input.get("video_grid_thw") is not None: video_grid_thw.extend( mm_input["video_grid_thw"].tolist()) if mm_input.get("second_per_grid_ts") is not None: second_per_grid_ts.extend( mm_input["second_per_grid_ts"]) if mm_input.get("audio_feature_lengths") is not None: audio_feature_lengths.extend( mm_input["audio_feature_lengths"]) if mm_input.get("use_audio_in_video") is True: use_audio_in_video = True hf_config = self.model_config.hf_config self.requests[req_id].mrope_positions, \ self.requests[req_id].mrope_position_delta = \ MRotaryEmbedding.get_input_positions_tensor( self.requests[req_id].prompt_token_ids, hf_config=hf_config, image_grid_thw=image_grid_thw, video_grid_thw=video_grid_thw, second_per_grid_ts=second_per_grid_ts, audio_feature_lengths=audio_feature_lengths, use_audio_in_video=use_audio_in_video, ) req_ids_to_add.append(req_id) # Update the states of the running/resumed requests. for req_data in scheduler_output.scheduled_cached_reqs: req_id = req_data.req_id req_state = self.requests[req_id] # Update the cached states. num_computed_tokens = req_data.num_computed_tokens req_state.num_computed_tokens = num_computed_tokens # Add the sampled token(s) from the previous step (if any). # This doesn't include "unverified" tokens like spec decode tokens. num_new_tokens = (num_computed_tokens + len(req_data.new_token_ids) - req_state.num_tokens) if num_new_tokens == 1: # Avoid slicing list in most common case. req_state.output_token_ids.append(req_data.new_token_ids[-1]) elif num_new_tokens > 0: req_state.output_token_ids.extend( req_data.new_token_ids[-num_new_tokens:]) # Update the block IDs. if not req_data.resumed_from_preemption: # Append the new blocks to the existing block IDs. for block_ids, new_block_ids in zip( # type: ignore[call-overload] req_state.block_ids, req_data.new_block_ids, strict=True): block_ids.extend(new_block_ids) else: # The request is resumed from preemption. # Replace the existing block IDs with the new ones. req_state.block_ids = req_data.new_block_ids req_index = self.input_batch.req_id_to_index.get(req_id) if req_index is None: # The request is not in the persistent batch. # The request was either preempted and resumed later, or was not # scheduled in the previous step and needs to be added again. req_ids_to_add.append(req_id) continue # Update the persistent batch. self.input_batch.num_computed_tokens_cpu[req_index] = ( num_computed_tokens) start_index = (len(req_state.block_ids) - len(req_data.new_block_ids)) self.input_batch.block_table.append_row(req_data.new_block_ids, req_index) # Add new_token_ids to token_ids_cpu. start_token_index = num_computed_tokens end_token_index = num_computed_tokens + len(req_data.new_token_ids) self.input_batch.token_ids_cpu[ req_index, start_token_index:end_token_index] = req_data.new_token_ids self.input_batch.num_tokens_no_spec[req_index] = end_token_index # Add spec_token_ids to token_ids_cpu. spec_token_ids = scheduler_output.scheduled_spec_decode_tokens.get( req_id, ()) if spec_token_ids: start_index = end_token_index end_token_index += len(spec_token_ids) self.input_batch.token_ids_cpu[ req_index, start_index:end_token_index] = spec_token_ids # NOTE(woosuk): `num_tokens` here may include spec decode tokens. self.input_batch.num_tokens[req_index] = end_token_index # Check if the batch has changed. If not, we can skip copying the # sampling metadata from CPU to GPU. batch_changed = len(removed_req_indices) > 0 or len(req_ids_to_add) > 0 # Add the new or resumed requests to the persistent batch. # The smaller empty indices are filled first. removed_req_indices = sorted(removed_req_indices, reverse=True) for req_id in req_ids_to_add: req_state = self.requests[req_id] if removed_req_indices: # Fill the empty index. req_index = removed_req_indices.pop() else: # Append to the end. req_index = None self.input_batch.add_request(req_state, req_index) # Condense the batched states if there are empty indices. if removed_req_indices: self.input_batch.condense(removed_req_indices) if batch_changed: self.input_batch.refresh_sampling_metadata() def _get_forward_metadata_across_dp( self, total_num_scheduled_tokens: int, with_prefill: bool) -> tuple[int, bool]: forward_metadata = torch.tensor( [total_num_scheduled_tokens, with_prefill], device="cpu", dtype=torch.int32) dist.all_reduce(forward_metadata, op=ReduceOp.MAX, group=get_dp_group().cpu_group) return int(forward_metadata[0]), bool(forward_metadata[1] > 0) def get_model(self) -> nn.Module: return self.model def _make_attention_mask(self, seq_lens, query_lens, position, attn_state) -> torch.Tensor: # Chunk Prefill situation. if attn_state == AscendAttentionState.ChunkedPrefill: return self.attn_mask_builder.get_splitfuse_attn_mask( seq_lens, query_lens, position, self.dtype, self.device) # Prefill without cache situation. elif attn_state == AscendAttentionState.PrefillNoCache: max_seq_len = max(seq_lens, default=0) return self.attn_mask_builder.get_attn_mask( max_seq_len, self.dtype, self.device) # Prefill with cache hit. elif attn_state == AscendAttentionState.PrefillCacheHit: return self.attn_mask_builder.get_attn_mask( 128, self.dtype, self.device) # Decode-only situation. else: return None def _calc_mrope_positions(self, scheduler_output: "SchedulerOutput"): mrope_pos_ptr = 0 for index, req_id in enumerate(self.input_batch.req_ids): req = self.requests[req_id] assert req.mrope_positions is not None num_computed_tokens = \ self.input_batch.num_computed_tokens_cpu[index] num_scheduled_tokens = \ scheduler_output.num_scheduled_tokens[req_id] num_prompt_tokens = len(req.prompt_token_ids) if num_computed_tokens + num_scheduled_tokens > num_prompt_tokens: prompt_part_len = max(0, num_prompt_tokens - num_computed_tokens) completion_part_len = max( 0, num_scheduled_tokens - prompt_part_len) else: prompt_part_len = num_scheduled_tokens completion_part_len = 0 assert num_scheduled_tokens == prompt_part_len + completion_part_len if prompt_part_len > 0: # prompt's mrope_positions are pre-computed dst_start = mrope_pos_ptr dst_end = mrope_pos_ptr + prompt_part_len src_start = num_computed_tokens src_end = num_computed_tokens + prompt_part_len self.mrope_positions_cpu[:, dst_start:dst_end] = \ req.mrope_positions[:,src_start:src_end] mrope_pos_ptr += prompt_part_len if completion_part_len > 0: # compute completion's mrope_positions on-the-fly dst_start = mrope_pos_ptr dst_end = mrope_pos_ptr + completion_part_len self.mrope_positions_cpu[:, dst_start:dst_end] = \ MRotaryEmbedding.get_next_input_positions_tensor( req.mrope_position_delta, context_len=num_computed_tokens + prompt_part_len, seq_len=num_computed_tokens + prompt_part_len + completion_part_len, ) mrope_pos_ptr += completion_part_len def _execute_mm_encoder(self, scheduler_output: "SchedulerOutput"): scheduled_encoder_inputs = scheduler_output.scheduled_encoder_inputs if not scheduled_encoder_inputs: return # Batch the multi-modal inputs. mm_inputs = list[MultiModalKwargs]() req_ids_pos = list[tuple[str, int, PlaceholderRange]]() for req_id, encoder_input_ids in scheduled_encoder_inputs.items(): req_state = self.requests[req_id] for mm_input_id in encoder_input_ids: mm_inputs.append(req_state.mm_inputs[mm_input_id]) req_ids_pos.append( (req_id, mm_input_id, req_state.mm_positions[mm_input_id])) # Batch mm inputs as much as we can: if a request in the batch has # multiple modalities or a different modality than the previous one, # we process it separately to preserve item order. # FIXME(ywang96): This is a hacky way to deal with multiple modalities # in the same batch while still being able to benefit from batching # multimodal inputs. The proper solution should be reordering the # encoder outputs. grouped_mm_inputs_list = group_mm_inputs_by_modality(mm_inputs) encoder_outputs = [] for grouped_mm_inputs in grouped_mm_inputs_list: batched_mm_inputs = MultiModalKwargs.batch(grouped_mm_inputs) batched_mm_inputs = MultiModalKwargs.as_kwargs(batched_mm_inputs, device=self.device) # Run the encoder. # `curr_group_outputs` is either of the following: # 1. A tensor of shape (num_items, feature_size, hidden_size) # in case feature_size is fixed across all multimodal items. # 2. A list or tuple (length: num_items) of tensors, each of shape # (feature_size, hidden_size) in case the feature size is dynamic # depending on the input multimodal items. curr_group_outputs = self.model.get_multimodal_embeddings( **batched_mm_inputs) sanity_check_mm_encoder_outputs( curr_group_outputs, expected_num_items=len(grouped_mm_inputs), ) for output in curr_group_outputs: encoder_outputs.append(output) # Cache the encoder outputs. for (req_id, input_id, pos_info), output in zip( req_ids_pos, encoder_outputs, ): if req_id not in self.encoder_cache: self.encoder_cache[req_id] = {} self.encoder_cache[req_id][input_id] = scatter_mm_placeholders( output, is_embed=pos_info.is_embed, ) def _gather_mm_embeddings( self, scheduler_output: "SchedulerOutput", ) -> list[torch.Tensor]: mm_embeds: list[torch.Tensor] = [] for req_id in self.input_batch.req_ids: num_scheduled_tokens = scheduler_output.num_scheduled_tokens[ req_id] req_state = self.requests[req_id] num_computed_tokens = req_state.num_computed_tokens mm_positions = req_state.mm_positions for i, pos_info in enumerate(mm_positions): start_pos = pos_info.offset num_encoder_tokens = pos_info.length # The encoder output is needed if the two ranges overlap: # [num_computed_tokens, # num_computed_tokens + num_scheduled_tokens) and # [start_pos, start_pos + num_encoder_tokens) if start_pos >= num_computed_tokens + num_scheduled_tokens: # The encoder output is not needed in this step. break if start_pos + num_encoder_tokens <= num_computed_tokens: # The encoder output is already processed and stored # in the decoder's KV cache. continue start_idx = max(num_computed_tokens - start_pos, 0) end_idx = min( num_computed_tokens - start_pos + num_scheduled_tokens, num_encoder_tokens) assert start_idx < end_idx assert req_id in self.encoder_cache assert i in self.encoder_cache[req_id] encoder_output = self.encoder_cache[req_id][i] if (is_embed := pos_info.is_embed) is not None: is_embed = is_embed[start_idx:end_idx] mm_embeds_item = gather_mm_placeholders( encoder_output[start_idx:end_idx], is_embed=is_embed, ) mm_embeds.append(mm_embeds_item) return mm_embeds def _process_reqs( self, scheduler_output: "SchedulerOutput", intermediate_tensors: Optional[IntermediateTensors] = None, ) -> tuple[SpecDecodeMetadata, torch.Tensor, SpecDecodeMetadata, torch.Tensor, int, torch.Tensor]: # Check input valid total_num_scheduled_tokens = scheduler_output.total_num_scheduled_tokens assert total_num_scheduled_tokens > 0 num_reqs = self.input_batch.num_reqs assert num_reqs > 0 if (self.use_aclgraph and total_num_scheduled_tokens <= self.aclgraph_batch_sizes[-1]): # Add padding to the batch size. num_input_tokens = self.vllm_config.pad_for_cudagraph( total_num_scheduled_tokens) else: # Eager mode. num_input_tokens = total_num_scheduled_tokens modified_batch = self.attn_metadata_builder.reorder_batch( self.input_batch, scheduler_output) if modified_batch: self.input_batch.refresh_sampling_metadata() # OPTIMIZATION: Start copying the block table first. # This way, we can overlap the copy with the following CPU operations. self.input_batch.block_table.commit(num_reqs) # Get the number of scheduled tokens for each request. # TODO: The Python loop can be slow. Optimize. num_scheduled_tokens = np.empty(num_reqs, dtype=np.int32) num_valid_tokens = np.empty(num_reqs, dtype=np.int32) max_num_scheduled_tokens = 0 for i, req_id in enumerate(self.input_batch.req_ids): num_tokens = scheduler_output.num_scheduled_tokens[req_id] num_scheduled_tokens[i] = num_tokens num_valid_tokens[i] = num_tokens - \ len(scheduler_output.scheduled_spec_decode_tokens.get(req_id, [])) max_num_scheduled_tokens = max(max_num_scheduled_tokens, num_tokens) # Hot-Swap lora model if self.lora_config: self.set_active_loras(self.input_batch, num_scheduled_tokens) # Prepare positions req_indices = np.repeat(self.arange_np[:num_reqs], num_scheduled_tokens) cu_num_tokens = np.cumsum(num_scheduled_tokens) cumsums_offsets = np.repeat(cu_num_tokens - num_scheduled_tokens, num_scheduled_tokens) sample_indices = cu_num_tokens - 1 sample_indices = torch.from_numpy(sample_indices).to(self.device, non_blocking=True) arange = self.arange_np[:total_num_scheduled_tokens] - cumsums_offsets positions_np = self.positions_np[:total_num_scheduled_tokens] np.add(self.input_batch.num_computed_tokens_cpu[req_indices], arange, out=positions_np) # Calculate M-RoPE positions. # Only relevant for models using M-RoPE (e.g, Qwen2-VL) if self.uses_mrope: self._calc_mrope_positions(scheduler_output) if self.uses_mrope: # Only relevant for models using M-RoPE (e.g, Qwen2-VL) self.mrope_positions[:, :total_num_scheduled_tokens].copy_( self.mrope_positions_cpu[:, :total_num_scheduled_tokens], non_blocking=True) self.positions[:total_num_scheduled_tokens].copy_( self.positions_cpu[:total_num_scheduled_tokens], non_blocking=True) positions = self.positions[:num_input_tokens] self.query_lens = torch.from_numpy(num_scheduled_tokens) self.seq_lens_np[:num_reqs] = ( self.input_batch.num_computed_tokens_cpu[:num_reqs] + num_scheduled_tokens) seq_lens = self.seq_lens_cpu[:num_reqs] block_table_indices = (req_indices * self.max_num_blocks_per_req + positions_np // self.block_size) block_table_cpu = self.input_batch.block_table[0].get_cpu_tensor() block_numbers = block_table_cpu.flatten()[block_table_indices].numpy() block_offsets = positions_np % self.block_size np.add(block_numbers * self.block_size, block_offsets, out=self.slot_mapping_np[:total_num_scheduled_tokens]) ascend_config = get_ascend_config() use_spec_decode = len( scheduler_output.scheduled_spec_decode_tokens) > 0 if np.array_equal(self.seq_lens_np[:num_reqs], num_scheduled_tokens): attn_state = AscendAttentionState.PrefillNoCache # We assume it is the decode stage, where prefill occurs but only one token is not hit in cache. elif np.all(num_scheduled_tokens == 1): attn_state = AscendAttentionState.DecodeOnly # Speculative decoding. elif np.all(num_valid_tokens == 1): attn_state = AscendAttentionState.SpecDecoding # splitfuse elif not ascend_config.ascend_scheduler_config.enabled or self.chunked_prefill_enabled: attn_state = AscendAttentionState.ChunkedPrefill else: attn_state = AscendAttentionState.PrefillCacheHit attn_mask = self._make_attention_mask(seq_lens=seq_lens, query_lens=num_scheduled_tokens, position=positions, attn_state=attn_state) self.attn_mask = attn_mask self.attn_state = attn_state # type: ignore extra_builder_kwargs = {} self.query_start_loc_np[0] = 0 self.query_start_loc_np[1:num_reqs + 1] = cu_num_tokens self.query_start_loc[:num_reqs + 1].copy_( self.query_start_loc_cpu[:num_reqs + 1], non_blocking=True) self.seq_lens[:num_reqs].copy_(self.seq_lens_cpu[:num_reqs], non_blocking=True) # Fill unused with -1. Needed for reshape_and_cache self.seq_lens[num_reqs:].fill_(0) self.query_start_loc[num_reqs + 1:].fill_(-1) query_start_loc = self.query_start_loc[:num_reqs + 1] seq_lens = self.seq_lens[:num_reqs] common_attn_metadata = CommonAttentionMetadata( query_start_loc=query_start_loc, seq_lens=seq_lens) with_prefill = attn_state not in [ AscendAttentionState.DecodeOnly, AscendAttentionState.SpecDecoding ] if self.dp_size > 1: max_num_tokens, with_prefill = self._get_forward_metadata_across_dp( total_num_scheduled_tokens, with_prefill) extra_builder_kwargs['max_num_tokens_across_dp'] = max_num_tokens extra_builder_kwargs['with_prefill_across_dp'] = with_prefill # Add graph_pad_size here if self.torchair_graph_enabled and not with_prefill: if self.dp_size > 1: padded_batch_size = self.select_torchair_padded_batch_size( max_num_tokens) else: padded_batch_size = self.select_torchair_padded_batch_size( total_num_scheduled_tokens) graph_pad_size = padded_batch_size - total_num_scheduled_tokens extra_builder_kwargs['graph_pad_size'] = graph_pad_size if self.vllm_config.model_config.use_mla: attn_metadata = self.attn_metadata_builder.build( # type: ignore num_reqs=num_reqs, num_actual_tokens=total_num_scheduled_tokens, max_query_len=max_num_scheduled_tokens, common_attn_metadata=common_attn_metadata, common_prefix_len=None, **extra_builder_kwargs, ) else: attn_metadata = self.attn_metadata_builder.build( # type: ignore num_reqs=num_reqs, num_actual_tokens=total_num_scheduled_tokens, max_query_len=max_num_scheduled_tokens, common_prefix_len=None, **extra_builder_kwargs, ) attn_metadata.num_input_tokens = num_input_tokens # Prepare input_ids token_indices = (positions_np + req_indices * self.input_batch.token_ids_cpu.shape[1]) torch.index_select(self.input_batch.token_ids_cpu_tensor.flatten(), 0, torch.from_numpy(token_indices), out=self.input_ids_cpu[:total_num_scheduled_tokens]) # Copy the tensors to the NPU. self.input_ids[:total_num_scheduled_tokens].copy_( self.input_ids_cpu[:total_num_scheduled_tokens], non_blocking=True) input_ids = self.input_ids[:num_input_tokens] # prepare the MRoPE for mllm if using multimodal num_input_tokens = total_num_scheduled_tokens # _prepare_inputs may reorder the batch, so we must gather multi # modal outputs after that to ensure the correct order if self.is_multimodal_model: # Run the multimodal encoder if any. self._execute_mm_encoder(scheduler_output) mm_embeds = self._gather_mm_embeddings(scheduler_output) else: mm_embeds = [] if self.is_multimodal_model: # NOTE(woosuk): To unify token ids and soft tokens (vision # embeddings), we always use embeddings (rather than token ids) # as input to the multimodal model, even when the input is text. input_ids = self.input_ids[:num_input_tokens] if mm_embeds: inputs_embeds = self.model.get_input_embeddings( input_ids, mm_embeds) else: inputs_embeds = self.model.get_input_embeddings(input_ids) # TODO(woosuk): Avoid the copy. Optimize. self.inputs_embeds[:num_input_tokens].copy_(inputs_embeds) inputs_embeds = self.inputs_embeds[:num_input_tokens] input_ids = None else: # For text-only models, we use token ids as input. # While it is possible to use embeddings as input just like the # multimodal models, it is not desirable for performance since # then the embedding layer is not included in the CUDA graph. input_ids = self.input_ids[:num_input_tokens] inputs_embeds = None if self.uses_mrope: positions = self.mrope_positions[:, :num_input_tokens] else: positions = self.positions[:num_input_tokens] if self.torchair_graph_enabled and not with_prefill: input_ids = self.input_ids[:padded_batch_size] positions = self.positions[:padded_batch_size] # Run forward pass with set_forward_context(attn_metadata, self.vllm_config, num_tokens=num_input_tokens): with ProfileExecuteDuration().capture_async("forward"): model_kwargs = {} if self.torchair_graph_enabled: model_kwargs["kv_caches"] = self.kv_caches model_kwargs["attn_metadata"] = attn_metadata if self.torchair_graph_enabled and not with_prefill: compiled_model = self._get_torchair_lazy_compiled_model( padded_batch_size) hidden_states = compiled_model( input_ids=input_ids, positions=positions, intermediate_tensors=intermediate_tensors, inputs_embeds=inputs_embeds, **model_kwargs, ) else: assert self.model is not None hidden_states = self.model( input_ids=input_ids, positions=positions, intermediate_tensors=intermediate_tensors, inputs_embeds=inputs_embeds, **model_kwargs, ) use_spec_decode = len( scheduler_output.scheduled_spec_decode_tokens) > 0 if not use_spec_decode: # NOTE(woosuk): Due to chunked prefills, the batch may contain # partial requests. While we should not sample any token # from these partial requests, we do so for simplicity. # We will ignore the sampled tokens from the partial requests. # TODO: Support prompt logprobs. spec_decode_metadata = None else: # Get the number of draft tokens for each request. # Iterate over the dictionary rather than all requests since not all # requests have draft tokens. num_draft_tokens = np.zeros(num_reqs, dtype=np.int32) for req_id, draft_token_ids in ( scheduler_output.scheduled_spec_decode_tokens.items()): req_idx = self.input_batch.req_id_to_index[req_id] num_draft_tokens[req_idx] = len(draft_token_ids) spec_decode_metadata = self._calc_spec_decode_metadata( num_draft_tokens, cu_num_tokens) sample_indices = spec_decode_metadata.logits_indices return (attn_metadata, hidden_states, spec_decode_metadata, positions, total_num_scheduled_tokens, sample_indices) def _calc_spec_decode_metadata( self, num_draft_tokens: np.ndarray, cu_num_scheduled_tokens: np.ndarray, ) -> SpecDecodeMetadata: # Inputs: # cu_num_scheduled_tokens: [ 4, 104, 107, 207, 209] # num_draft_tokens: [ 3, 0, 2, 0, 1] # Outputs: # cu_num_draft_tokens: [ 3, 3, 5, 5, 6] # logits_indices: [ 0, 1, 2, 3, 103, 104, 105, 106, # 206, 207, 208] # target_logits_indices: [ 0, 1, 2, 5, 6, 9] # bonus_logits_indices: [ 3, 4, 7, 8, 10] # Compute the logits indices. # [4, 1, 3, 1, 2] num_sampled_tokens = num_draft_tokens + 1 # Step 1. [4, 5, 8, 9, 11] cu_num_sampled_tokens = np.cumsum(num_sampled_tokens, dtype=np.int32) total_num_sampled_tokens = cu_num_sampled_tokens[-1] # Step 2. [0, 0, 0, 0, 4, 5, 5, 5, 8, 9, 9] cumsums_offsets = np.repeat(cu_num_sampled_tokens - num_sampled_tokens, num_sampled_tokens) # Step 3. [0, 1, 2, 3, 0, 0, 1, 2, 0, 0, 1] arange = self.arange_np[:total_num_sampled_tokens] - cumsums_offsets # Step 4. [0, 0, 0, 0, 103, 104, 104, 104, 206, 207, 207] logits_indices = np.repeat( cu_num_scheduled_tokens - num_sampled_tokens, num_sampled_tokens) # Step 5. [0, 1, 2, 3, 103, 104, 105, 106, 206, 207, 208] logits_indices += arange # Compute the bonus logits indices. bonus_logits_indices = cu_num_sampled_tokens - 1 # Compute the draft logits indices. # [3, 3, 5, 5, 6] cu_num_draft_tokens = np.cumsum(num_draft_tokens, dtype=np.int32) total_num_draft_tokens = cu_num_draft_tokens[-1] # [0, 0, 0, 3, 3, 5] cumsums_offsets = np.repeat(cu_num_draft_tokens - num_draft_tokens, num_draft_tokens) # [0, 1, 2, 0, 1, 0] arange = self.arange_np[:total_num_draft_tokens] - cumsums_offsets # [0, 0, 0, 5, 5, 9] target_logits_indices = np.repeat( cu_num_sampled_tokens - num_sampled_tokens, num_draft_tokens) # [0, 1, 2, 5, 6, 9] target_logits_indices += arange # TODO: Optimize the CPU -> NPU copy. cu_num_draft_tokens = torch.from_numpy(cu_num_draft_tokens).to( self.device, non_blocking=True) logits_indices = torch.from_numpy(logits_indices).to(self.device, non_blocking=True) target_logits_indices = torch.from_numpy(target_logits_indices).to( self.device, non_blocking=True) bonus_logits_indices = torch.from_numpy(bonus_logits_indices).to( self.device, non_blocking=True) # Compute the draft token ids. # draft_token_indices: [ 1, 2, 3, 105, 106, 208] draft_token_ids = self.input_ids[logits_indices] draft_token_ids = draft_token_ids[target_logits_indices + 1] metadata = SpecDecodeMetadata( draft_token_ids=draft_token_ids, num_draft_tokens=num_draft_tokens.tolist(), cu_num_draft_tokens=cu_num_draft_tokens, target_logits_indices=target_logits_indices, bonus_logits_indices=bonus_logits_indices, logits_indices=logits_indices, ) return metadata def apply_grammar_bitmask( self, scheduler_output: "SchedulerOutput", logits: torch.Tensor, ) -> torch.Tensor: # Serialization of np.ndarray is much more efficient than a tensor, # so we receive it in that format. grammar_bitmask = scheduler_output.grammar_bitmask if grammar_bitmask is None: return # We receive the structured output bitmask from the scheduler, but the # indices of the requests in the batch may not match the indices of # the bitmask since the scheduler doesn't know how the gpu runner is # ordering the requests in the batch. We need to sort the bitmask to # match the order of the requests used here. struct_out_req_batch_indices: dict[str, int] = {} indices_match = True for req_id in self.input_batch.req_ids: mask_index = scheduler_output.structured_output_request_ids.get( req_id) if mask_index is None: # not a structured output request continue batch_index = self.input_batch.req_id_to_index[req_id] if batch_index != mask_index: indices_match = False struct_out_req_batch_indices[req_id] = batch_index if not indices_match: # Sort the bitmask to match the order of the requests sorted_bitmask = np.zeros_like(grammar_bitmask) for req_id, batch_index in struct_out_req_batch_indices.items(): orig_index = scheduler_output.structured_output_request_ids[ req_id] sorted_bitmask[batch_index] = grammar_bitmask[orig_index] grammar_bitmask = sorted_bitmask grammar_bitmask = torch.from_numpy(grammar_bitmask) # TODO: compatibility with spec decode. # NOTE: # 1. XGrammar bitmask applying only supports CPU and GPU. # 2. The logits and bitmask should be on the same device. # 3. XGrammar logits on CPU only supports float32 dtype. logits_dtype = logits.dtype logits = logits.to("cpu").float() xgr.apply_token_bitmask_inplace( logits, grammar_bitmask, indices=list(struct_out_req_batch_indices.values()), ) return logits.to(self.device).to(logits_dtype) def _get_spec_token_ids( self, valid_sampled_token_ids: list[list[int]], sampling_metadata: SamplingMetadata, scheduler_output: "SchedulerOutput", spec_decode_metadata: SpecDecodeMetadata, positions: torch.Tensor, num_scheduled_tokens: int, hidden_states: torch.Tensor, attn_metadata: SpecDecodeMetadata, ) -> Optional[list[list[int]]]: if not self.use_spec_decode: # Speculative decoding is not enabled. spec_token_ids = None elif self.speculative_config.method == "ngram": assert isinstance(self.drafter, NgramProposer) spec_token_ids = self._generate_draft_token_ids( valid_sampled_token_ids, sampling_metadata) elif self.speculative_config.method == "eagle": raise NotImplementedError( "eagle method for spec decode doesn't work on vllm-ascend currently" ) elif self.speculative_config.method == 'deepseek_mtp': assert isinstance(self.drafter, MtpProposer) spec_token_ids = self._generate_mtp_token_ids( valid_sampled_token_ids, sampling_metadata, scheduler_output, spec_decode_metadata, positions, num_scheduled_tokens, hidden_states, attn_metadata) return spec_token_ids @torch.inference_mode() def execute_model( self, scheduler_output: "SchedulerOutput", intermediate_tensors: Optional[IntermediateTensors] = None, ) -> Union[ModelRunnerOutput, torch.Tensor]: with ProfileExecuteDuration().capture_async( "prepare input and forward"): self._update_states(scheduler_output) if not scheduler_output.total_num_scheduled_tokens: # Return empty ModelRunnerOuptut if there's no work to do. return EMPTY_MODEL_RUNNER_OUTPUT (attn_metadata, hidden_states, spec_decode_metadata, positions, num_scheduled_tokens, sample_indices) = (self._process_reqs(scheduler_output, intermediate_tensors)) with ProfileExecuteDuration().capture_async("post process"): logits = self.model.compute_logits(hidden_states[sample_indices], None) # Apply structured output bitmasks if present if scheduler_output.grammar_bitmask is not None: logits = self.apply_grammar_bitmask(scheduler_output, logits) # Sample the next token and get logprobs if needed. sampling_metadata = self.input_batch.sampling_metadata if spec_decode_metadata is None: sampler_output = self.sampler( logits=logits, sampling_metadata=sampling_metadata, ) else: # When indexing with a tensor (bonus_logits_indices), PyTorch # creates a new tensor with separate storage from the original # logits tensor. This means any in-place operations on bonus_logits # won't affect the original logits tensor. bonus_logits = logits[ spec_decode_metadata.bonus_logits_indices] sampler_output = self.sampler( logits=bonus_logits, sampling_metadata=sampling_metadata, ) bonus_token_ids = sampler_output.sampled_token_ids # Just like `bonus_logits`, `target_logits` is a new tensor with # separate storage from the original `logits` tensor. Therefore, # it is safe to update `target_logits` in place. target_logits = logits[ spec_decode_metadata.target_logits_indices] output_token_ids = self.rejection_sampler( spec_decode_metadata, None, # draft_probs target_logits, bonus_token_ids, sampling_metadata, ) sampler_output.sampled_token_ids = output_token_ids # TODO(woosuk): The following loop can be slow since it iterates over # the requests one by one. Optimize. discard_sampled_tokens_req_indices = [] for i, req_id in enumerate(self.input_batch.req_ids): req_state = self.requests[req_id] seq_len = (req_state.num_computed_tokens + scheduler_output.num_scheduled_tokens[req_id]) if seq_len < req_state.num_tokens: # Ignore the sampled token. # Rewind the generator state as if the token was not sampled. generator = self.input_batch.generators.get(i) if generator is not None: generator.set_offset(generator.get_offset() - 4) discard_sampled_tokens_req_indices.append(i) # NOTE: NPU -> CPU Sync happens here. # Move as many CPU operations as possible before this sync point. logprobs_tensors = sampler_output.logprobs_tensors logprobs_lists = logprobs_tensors.tolists() \ if logprobs_tensors is not None else None # Get the valid generated tokens. sampled_token_ids = sampler_output.sampled_token_ids max_gen_len = sampled_token_ids.shape[-1] if max_gen_len == 1: # No spec decode tokens. valid_sampled_token_ids = sampled_token_ids.tolist() else: # Includes spec decode tokens. valid_sampled_token_ids = self.rejection_sampler.parse_output( sampled_token_ids, self.input_batch.vocab_size, ) for i in discard_sampled_tokens_req_indices: valid_sampled_token_ids[i].clear() spec_token_ids = self._get_spec_token_ids( valid_sampled_token_ids, sampling_metadata, scheduler_output, spec_decode_metadata, positions, num_scheduled_tokens, hidden_states, attn_metadata, ) model_runner_output = ModelRunnerOutput( req_ids=self.input_batch.req_ids, req_id_to_index=self.input_batch.req_id_to_index, sampled_token_ids=valid_sampled_token_ids, spec_token_ids=spec_token_ids, logprobs=logprobs_lists, prompt_logprobs_dict={}, ) durations = ProfileExecuteDuration().pop_captured_sync() if durations: dr_str = [ f"[{tag}]:{duration:.2f}ms" for tag, duration in durations.items() ] captured_name = "Decode" if self.attn_state == AscendAttentionState.DecodeOnly else "Prefill" logger.info("Profile execute duration [%s]:%s", captured_name, " ".join(dr_str)) return model_runner_output def _profile_multimodal(self) -> None: # TODO: handle encoder-decoder models once we support them. # NOTE: Currently model is profiled with a single non-text # modality with the max possible input tokens even when # it supports multiple. if (not self.is_multimodal_model or self.max_num_encoder_input_tokens <= 0 or self.encoder_cache_size <= 0): return max_tokens_by_modality_dict = ( MULTIMODAL_REGISTRY.get_max_tokens_per_item_by_nonzero_modality( self.model_config)) dummy_data_modality, max_tokens_per_mm_item = max( max_tokens_by_modality_dict.items(), key=lambda item: item[1]) # Check how many items of this modality can be supported by # the encoder budget. encoder_budget = min(self.max_num_encoder_input_tokens, self.encoder_cache_size) max_num_mm_items_encoder_budget = cdiv(encoder_budget, max_tokens_per_mm_item) # Check how many items of this modality can be supported by # the decoder budget. max_mm_items_per_req = self.mm_registry.get_mm_limits_per_prompt( self.model_config)[dummy_data_modality] # NOTE: We do not consider max_num_batched_tokens on purpose # because the multimodal embeddings can be generated in advance # and chunked prefilled. max_num_mm_items_decoder_budget = self.max_num_reqs * \ max_mm_items_per_req max_num_mm_items = min(max_num_mm_items_encoder_budget, max_num_mm_items_decoder_budget) logger.info( "Encoder cache will be initialized with a budget of %s tokens," " and profiled with %s %s items of the maximum feature size.", encoder_budget, max_num_mm_items, dummy_data_modality) # Create dummy batch of multimodal inputs. dummy_request_data = self.input_registry.dummy_data_for_profiling( model_config=self.model_config, seq_len=self.max_num_tokens, mm_registry=self.mm_registry, ) dummy_mm_data = dummy_request_data.multi_modal_data if not isinstance(dummy_mm_data, MultiModalKwargs): # TODO: Delete this check once input mapper is fully removed. raise RuntimeError("Legacy input mapper is not supported in V1") # Dummy data definition in V0 may contain multiple multimodal items # (e.g, multiple images) for a single request, therefore here we # always replicate first item by max_num_mm_items times since in V1 # they are scheduled to be processed separately. dummy_mm_item = dummy_mm_data.get_item(modality=dummy_data_modality, item_index=0) dummy_mm_kwargs = MultiModalKwargs.from_items([dummy_mm_item]) batched_dummy_mm_inputs = MultiModalKwargs.batch([dummy_mm_kwargs] * max_num_mm_items) batched_dummy_mm_inputs = MultiModalKwargs.as_kwargs( batched_dummy_mm_inputs, device=self.device) # Run multimodal encoder. dummy_encoder_outputs = self.model.get_multimodal_embeddings( **batched_dummy_mm_inputs) assert len(dummy_encoder_outputs) == max_num_mm_items, ( "Expected dimension 0 of encoder outputs to match the number " f"of multimodal data items: {max_num_mm_items}, got " f"{len(dummy_encoder_outputs)=} instead. This is most likely " "due to the 'get_multimodal_embeddings' method of the model " "not implemented correctly.") # Cache the dummy encoder outputs. self.encoder_cache["tmp"] = dict(enumerate(dummy_encoder_outputs)) @torch.inference_mode() def _dummy_run( self, num_tokens: int, is_compile: bool = False, with_prefill: bool = True, ) -> torch.Tensor: # Set num_scheduled_tokens based on num_tokens and max_num_seqs # for dummy run with LoRA so that the num_reqs collectively # has num_tokens in total. assert num_tokens <= self.scheduler_config.max_num_batched_tokens max_num_reqs = self.scheduler_config.max_num_seqs num_reqs = max_num_reqs if num_tokens >= max_num_reqs else num_tokens min_tokens_per_req = num_tokens // num_reqs num_scheduled_tokens_list = [min_tokens_per_req] * num_reqs num_scheduled_tokens_list[-1] += num_tokens % num_reqs assert sum(num_scheduled_tokens_list) == num_tokens assert len(num_scheduled_tokens_list) == num_reqs num_scheduled_tokens = np.array(num_scheduled_tokens_list, dtype=np.int32) with self.maybe_dummy_run_with_lora(self.lora_config, num_scheduled_tokens): model = self.model if self.is_multimodal_model: input_ids = None inputs_embeds = self.inputs_embeds[:num_tokens] else: input_ids = self.input_ids[:num_tokens] inputs_embeds = None if self.uses_mrope: positions = self.mrope_positions[:, :num_tokens] else: positions = self.positions[:num_tokens] if get_pp_group().is_first_rank: intermediate_tensors = None else: if self.intermediate_tensors is None: self.intermediate_tensors = ( self.model.make_empty_intermediate_tensors( batch_size=num_tokens, dtype=self.dtype, device=self.device)) intermediate_tensors = IntermediateTensors({ k: v[:num_tokens] for k, v in self.intermediate_tensors.items() }) with set_forward_context(None, self.vllm_config, num_tokens=num_tokens): if self.torchair_graph_enabled and not with_prefill: attn_metadata = self.attn_metadata_builder.build_dummy( num_reqs=num_tokens, num_actual_tokens=1) # Only mark static while compiling if is_compile: torch._dynamo.mark_static(input_ids) torch._dynamo.mark_static(positions) torch._dynamo.mark_static( attn_metadata.decode.block_table) torch._dynamo.mark_static( attn_metadata.decode.input_positions) torch._dynamo.mark_static(attn_metadata.slot_mapping) for kv in self.kv_caches: assert isinstance( kv, tuple), "kv_cache must be a tuple" torch._dynamo.mark_static(kv[0]) torch._dynamo.mark_static(kv[1]) compiled_model = self._get_torchair_lazy_compiled_model( num_tokens) hidden_states = compiled_model( input_ids=input_ids, positions=positions, intermediate_tensors=intermediate_tensors, inputs_embeds=None, kv_caches=self.kv_caches, attn_metadata=attn_metadata, ) else: hidden_states = model( input_ids=input_ids, positions=positions, intermediate_tensors=intermediate_tensors, inputs_embeds=inputs_embeds) return hidden_states def profile_run(self) -> None: # FIXME Profile with multimodal encoder & encoder cache. # current _profile_multimodal() using PyTorch SDPA backend method not # support for window/full attn to reduce Memcpy operations, so will cause # Out Of Memory problem, so we currently don't use self._profile_multimodal() # self._profile_multimodal() # For profile, have maximum num_reqs and that collectively have # maximum num_tokens. num_reqs = self.scheduler_config.max_num_seqs num_tokens = self.max_num_tokens min_tokens_per_req = num_tokens // num_reqs num_scheduled_tokens_list = [min_tokens_per_req] * num_reqs num_scheduled_tokens_list[-1] += num_tokens % num_reqs assert sum(num_scheduled_tokens_list) == num_tokens assert len(num_scheduled_tokens_list) == num_reqs num_scheduled_tokens = np.array(num_scheduled_tokens_list, dtype=np.int32) logit_indices = np.cumsum(num_scheduled_tokens) - 1 # assert self.lora_manager is not None, "LoRA is not enabled" # TODO: call maybe_profile_with_lora() # Trigger compilation for general shape. hidden_states = self._dummy_run(self.max_num_tokens) if get_pp_group().is_last_rank: hidden_states = hidden_states[logit_indices] logits = self.model.compute_logits(hidden_states, None) else: logits = None NPUPlatform.synchronize() del hidden_states, logits self.encoder_cache.clear() gc.collect() def load_model(self) -> None: logger.info("Starting to load model %s...", self.model_config.model) with DeviceMemoryProfiler() as m: # noqa: SIM117 self.model = get_model(vllm_config=self.vllm_config) if hasattr(self, "drafter"): logger.info("Loading drafter model...") self.drafter.load_model() if self.lora_config: self.model = self.load_lora_model(self.model, self.model_config, self.scheduler_config, self.lora_config, self.device) logger.info("Loading model weights took %.4f GB", m.consumed_memory / float(2**30)) def _get_torchair_lazy_compiled_model(self, batch_size: int): if batch_size < 0 or batch_size > self.max_num_reqs: raise ValueError( f"Bad graph batch size:{batch_size}! max_num_reqs:{self.max_num_reqs}" ) compiled_model = self.torchair_compiled_models.get( batch_size ) if self.use_cached_npu_graph else self.torchair_compiled_model if compiled_model: return compiled_model import torchair # type: ignore from torchair import patch_for_hcom # type: ignore patch_for_hcom() config = torchair.CompilerConfig() config.experimental_config.frozen_parameter = True config.experimental_config.tiling_schedule_optimize = True config.experimental_config.enable_view_optimize = \ get_ascend_config().torchair_graph_config.enable_view_optimize torch.npu.set_compile_mode(jit_compile=False) if not self.use_cached_npu_graph: npu_backend = torchair.get_npu_backend(compiler_config=config) self.torchair_compiled_model = torch.compile( self.model, dynamic=True, fullgraph=envs_vllm.VLLM_TEST_DYNAMO_FULLGRAPH_CAPTURE, backend=npu_backend) return self.torchair_compiled_model else: # Generate a new forward proxy code object to prevent the invalidation of # compilation cache caused by dynamo retracing forward_proxy_name = f"{self.model.__class__.__name__}_forward_with_batch_size_{batch_size}" forward_fn = self.model.forward code = forward_fn.__code__ # Mark code object with a new proxy name modified_code = code.replace(co_name=forward_proxy_name, ) modified_func = types.FunctionType(modified_code, forward_fn.__globals__, name=forward_proxy_name, argdefs=forward_fn.__defaults__) self.model.__dict__[forward_proxy_name] = modified_func.__get__( self.model, nn.Module) self.torchair_compiled_models[ batch_size] = torchair.inference.cache_compile( self.model.__dict__[forward_proxy_name], dynamic=True, fullgraph=envs_vllm.VLLM_TEST_DYNAMO_FULLGRAPH_CAPTURE, config=config, ge_cache=False) return self.torchair_compiled_models[batch_size] def initialize_kv_cache(self, kv_cache_config: KVCacheConfig) -> None: """ Initialize KV cache based on `kv_cache_config`. Args: kv_cache_config: Configuration for the KV cache, including the KV cache size of each layer """ import torch_npu kv_caches: Dict[str, torch.Tensor] = {} self.input_batch = InputBatch( max_num_reqs=self.max_num_reqs, max_model_len=self.model_config.max_model_len, max_num_batched_tokens=self.max_num_tokens, device=self.device, pin_memory=True, vocab_size=self.model_config.get_vocab_size(), block_sizes=[self.cache_config.block_size], ) kv_cache_sizes = {} for kv_cache_tensor in kv_cache_config.kv_cache_tensors: assert len(kv_cache_tensor.shared_by) == 1, ( "KV cache tensor shared by multiple layers is not supported in " "NPU.") kv_cache_sizes[kv_cache_tensor.shared_by[0]] = kv_cache_tensor.size for kv_cache_group in kv_cache_config.kv_cache_groups: kv_cache_spec = kv_cache_group.kv_cache_spec for layer_name in kv_cache_group.layer_names: tensor_size = kv_cache_sizes[layer_name] assert tensor_size % kv_cache_spec.page_size_bytes == 0 num_blocks = tensor_size // kv_cache_spec.page_size_bytes # `num_blocks` is the number of blocks the model runner can use. # `kv_cache_config.num_blocks` is the number of blocks that # KVCacheManager may allocate. # Since different GPUs may have different number of layers and # different memory capacities, `num_blocks` can be different on # different GPUs, and `kv_cache_config.num_blocks` is set to # the min of all `num_blocks`. Verify it here. assert num_blocks >= kv_cache_config.num_blocks # TODO: remove this after the OOM issue is located and fixed, otherwise, some model may # encounter OOM issue if isinstance(kv_cache_spec, FullAttentionSpec): kv_cache_shape = self.attn_backend.get_kv_cache_shape( num_blocks, kv_cache_spec.block_size, kv_cache_spec.num_kv_heads, kv_cache_spec.head_size) dtype = kv_cache_spec.dtype if self.torchair_graph_enabled: layer_kv_cache_nope = torch.zeros( kv_cache_shape[:-1] + (self.model_config.hf_text_config.kv_lora_rank, ), dtype=self.dtype, pin_memory=True, device=self.device) layer_kv_cache_pe = torch.zeros( kv_cache_shape[:-1] + (self.model_config.hf_text_config.qk_rope_head_dim, ), dtype=self.dtype, pin_memory=True, device=self.device) kv_caches[layer_name] = (layer_kv_cache_nope, layer_kv_cache_pe) torch_npu.npu_format_cast(kv_caches[layer_name][0], 2) torch_npu.npu_format_cast(kv_caches[layer_name][1], 2) else: kv_caches[layer_name] = torch.zeros(kv_cache_shape, dtype=dtype, device=self.device) torch_npu.npu_format_cast(kv_caches[layer_name], 2) else: # TODO: add new branches when introducing more types of # KV cache specs. raise ValueError("Unknown KV cache spec type.") bind_kv_cache( kv_caches, self.vllm_config.compilation_config.static_forward_context, self.kv_caches) def get_kv_cache_spec(self) -> dict[str, KVCacheSpec]: """ Generates the KVCacheSpec by parsing the kv cache format from each Attention module in the static forward context. Returns: KVCacheSpec: A dictionary mapping layer names to their KV cache format. Layers that do not need KV cache are not included. """ forward_ctx = self.vllm_config.compilation_config.static_forward_context block_size = self.vllm_config.cache_config.block_size use_mla = self.vllm_config.model_config.use_mla kv_cache_spec: dict[str, KVCacheSpec] = {} for layer_name, attn_module in forward_ctx.items(): if isinstance(attn_module, FusedMoE): continue # TODO: Support other attention modules, e.g., sliding window, # cross-attention assert isinstance(attn_module, Attention) if attn_module.attn_type == AttentionType.DECODER: kv_cache_spec[layer_name] = FullAttentionSpec( block_size=block_size, num_kv_heads=attn_module.num_kv_heads, head_size=attn_module.head_size, dtype=attn_module.dtype, use_mla=use_mla) elif attn_module.attn_type in (AttentionType.ENCODER, AttentionType.ENCODER_ONLY): # encoder-only attention does not need KV cache. continue elif attn_module.attn_type == AttentionType.ENCODER_DECODER: raise NotImplementedError else: raise ValueError( f"Unknown attention type: {attn_module.attn_type}") return kv_cache_spec def capture_model(self) -> None: start_time = time.perf_counter() start_free_npu_memory = torch.npu.mem_get_info()[0] # TODO(NeverRaR): Calling graph_capture(device=self.device) in # torchair graph capture can cause some issues, so now we just # temporarily split the codepath for the two different graph patterns. if self.torchair_graph_enabled: torchair_graph_batch_sizes = self.torchair_graph_batch_sizes graph_num = len(torchair_graph_batch_sizes) logger.info( "Capturing torchair graph, this usually takes %.1f~%.1f mins.", 0.5 * graph_num, 1.5 * graph_num) # Trigger torchair graph capture for specific shapes. # Capture the large shapes first so that the smaller shapes # can reuse the memory pool allocated for the large shapes. for idx, num_tokens in enumerate( reversed(torchair_graph_batch_sizes)): for _ in range(self.vllm_config.compilation_config. cudagraph_num_of_warmups): self._dummy_run(num_tokens, is_compile=True, with_prefill=False) self._dummy_run(num_tokens, is_compile=True, with_prefill=False) logger.info("Batchsize %d is compiled successfully: %d/%d.", num_tokens, idx + 1, graph_num) elif self.use_aclgraph: # Trigger ACL graph capture for specific shapes. # Capture the large shapes first so that the smaller shapes # can reuse the memory pool allocated for the large shapes. with graph_capture(device=self.device): for num_tokens in reversed(self.aclgraph_batch_sizes): for _ in range(self.vllm_config.compilation_config. cudagraph_num_of_warmups): self._dummy_run(num_tokens) self._dummy_run(num_tokens) else: logger.info("Skipping NPU graph capture for eager mode.") return end_time = time.perf_counter() end_free_npu_memory = torch.npu.mem_get_info()[0] elapsed_time = end_time - start_time npu_graph_size = start_free_npu_memory - end_free_npu_memory # This usually takes 5~20 seconds. logger.info("Graph capturing finished in %.0f secs, took %.2f GiB", elapsed_time, npu_graph_size / (1 << 30)) def _generate_draft_token_ids( self, sampled_token_ids: list[list[int]], sampling_metadata: SamplingMetadata, ) -> list[list[int]]: # TODO(woosuk): Optimize. draft_token_ids: list[list[int]] = [] for i, sampled_ids in enumerate(sampled_token_ids): num_sampled_ids = len(sampled_ids) if not num_sampled_ids: # Skip speculative decoding. draft_token_ids.append([]) continue # Skip requests that require top-p, top-k, etc. req_id = self.input_batch.req_ids[i] if not is_spec_decode_supported(req_id, self.input_batch): draft_token_ids.append([]) continue # Add sampled_token_ids to token_ids_cpu. start_idx = self.input_batch.num_tokens_no_spec[i] end_idx = start_idx + num_sampled_ids self.input_batch.token_ids_cpu[i, start_idx:end_idx] = sampled_ids drafter_output = self.drafter.propose( self.input_batch.token_ids_cpu[i, :end_idx]) if drafter_output is None or len(drafter_output) == 0: draft_token_ids.append([]) else: draft_token_ids.append(drafter_output.tolist()) return draft_token_ids def _generate_mtp_token_ids( self, valid_sampled_token_ids: list[list[int]], sampling_metadata: SamplingMetadata, scheduler_output: "SchedulerOutput", spec_decode_metadata: SpecDecodeMetadata, positions: torch.Tensor, num_scheduled_tokens: int, hidden_states: torch.Tensor, attn_metadata: SpecDecodeMetadata, ): next_token_ids: list[int] = [] for i, token_ids in enumerate(valid_sampled_token_ids): if token_ids: # Common case. next_token_id = token_ids[-1] else: # Partial prefill (rare case). # Get the next token id from the request state. req_id = self.input_batch.req_ids[i] req_state = self.requests[req_id] seq_len = (req_state.num_computed_tokens + scheduler_output.num_scheduled_tokens[req_id]) next_token_id = req_state.get_token_id(seq_len) next_token_ids.append(next_token_id) next_token_ids = torch.tensor(next_token_ids, dtype=torch.int32, device=self.device) if spec_decode_metadata is None: # input_ids can be None for multimodal models. target_token_ids = self.input_ids[:num_scheduled_tokens] target_positions = positions[:num_scheduled_tokens] target_hidden_states = hidden_states[:num_scheduled_tokens] target_slot_mapping = attn_metadata.slot_mapping cu_num_tokens = attn_metadata.query_start_loc else: # TODO(woosuk): Refactor this. num_draft_tokens = spec_decode_metadata.num_draft_tokens num_rejected_tokens = [ n + 1 - len(valid_sampled_token_ids[i]) if n > 0 else 0 for i, n in enumerate(num_draft_tokens) ] num_rejected_tokens = torch.tensor( num_rejected_tokens, dtype=torch.int32, device=self.device, ) cu_num_tokens, token_indices = self.drafter.prepare_inputs( attn_metadata.query_start_loc, num_rejected_tokens, ) target_token_ids = self.input_ids[token_indices] target_positions = positions[token_indices] target_hidden_states = hidden_states[token_indices] target_slot_mapping = attn_metadata.slot_mapping[token_indices] draft_token_ids = self.drafter.propose( target_token_ids=target_token_ids, target_positions=target_positions, target_hidden_states=target_hidden_states, target_slot_mapping=target_slot_mapping, next_token_ids=next_token_ids, cu_num_tokens=cu_num_tokens, block_table=attn_metadata.block_tables, sampling_metadata=sampling_metadata, ) spec_token_ids = draft_token_ids.tolist() return spec_token_ids def init_torchair_graph_batch_sizes(self): start_graph_batch_size = 4 tp_size = get_tensor_model_parallel_world_size() # NOTE: When use all2all | mc2, We need to slice the `num_tokens` dimension into `tp_size` blocks start_graph_batch_size = max(start_graph_batch_size, tp_size) while (start_graph_batch_size <= self.scheduler_config.max_num_seqs): self.torchair_graph_batch_sizes.append(start_graph_batch_size) start_graph_batch_size *= 2 def select_torchair_padded_batch_size(self, batch_size: int): selected_batch_size = self.max_num_reqs for padded_batch_size in self.torchair_graph_batch_sizes: if batch_size <= padded_batch_size < selected_batch_size: selected_batch_size = padded_batch_size return selected_batch_size