# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project """ # MLA Common Components This file implements common components for MLA implementations. First we define: Sq as Q sequence length Skv as KV sequence length MLA has two possible ways of computing, a data-movement friendly approach and a compute friendly approach, we generally want to use the compute friendly approach for "prefill" (i.e. the ratio Sq / Skv is "small", is near 1) and the data-movement friendly approach for "decode" (i.e. the ratio Sq / Skv is "large"). NOTE what we deem small and large is currently determined by if its labelled prefill or decode by the scheduler, but this is something we should probably tune. Main reference: DeepseekV2 paper, and FlashInfer Implementation (https://arxiv.org/abs/2405.04434 and https://github.com/flashinfer-ai/flashinfer/pull/551). Deepseek's MLA attention works the following way: * Use a single latent vector to represent the per-token entry of the KV cache. * For decode (i.e. the memory friendly approach) the attention "simulates" a multi-head attention, while the compute is similar to multi-query attention. Below is example of both paths assuming batchsize = 1 ## More Extent Definitions: C Context length, `Skv - Sq` H hidden size N number of attention heads Lq latent dimension for Q 1536 in DSV3 Lkv latent dimension for K/V 512 in DSV3 P nope dimension, no rope. 128 in DSV3 R rope dimension, goes through rope. 64 in DSV3 V V head dim. 128 in DSV3 ## Vector/Matrix Definitions h_t hidden states (input to attention) shape [Sq, H] q_c latent/compressed Q shape [Sq, Lq] q_nope uncompressed Q (no-rope) shape [Sq, N, P] q_pe uncompressed Q (rope) shape [Sq, N, R] kv_c latent/compressed KV shape [Skv, Lkv] k_pe decoupled k position embeddings shape [Skv, R] new_kv_c new kv_c from current iter shape [Sq, Lkv] new_k_pe new k_pe from current iter shape [Sq, R] cache_kv_c cached k_c from previous iters shape [C, Lkv] cache_k_pe cached k_pe from previous iters shape [C, R] W_DQ project h_t to q_c shape [H, Lq] W_UQ project q_c to q_nope shape [Lq, N * P] W_QR project q_c to q_pe shape [Lq, N * R] W_DKV project h_t to kv_c shape [H, Lkv] W_UK project kv_c to k_nope shape [Lkv, N, P] W_KR project h_t to k_pe shape [H, R] W_UV project kv_c to v shape [Lkv, N, V] W_O project v to h_t shape [N * V, H] ## Compute Friendly Approach (i.e. "_forward_prefill"): q_c = h_t @ W_DQ q_nope = (q_c @ W_UQ).view(Sq, N, P) q_pe = RoPE(q_c @ W_QR).view(Sq, N, R) new_kv_c = h_t @ W_DKV new_k_pe = RoPE(h_t @ W_KR) kv_c = torch.cat([new_kv_c, cache_kv_c], dim=0) k_pe = torch.cat([new_k_pe, cache_k_pe], dim=0) k_nope = (kv_c @ W_UK.view(Lkv, N * P)).view(Skv, N, P) v = (kv_c @ W_UV.view(Lkv, N * V)).view(Skv, N, V) // MHA with QK headdim = P + R // V headdim = V // spda_o shape [Sq, N, V] spda_o = scaled_dot_product_attention( torch.cat([q_nope, q_pe], dim=-1), torch.cat([k_nope, k_pe.unsqueeze(1).expand(-1, N, -1)], dim=-1), v ) return spda_o @ W_O NOTE: in the actual code, `kv_b_proj` is [W_UK; W_UV] concatenated per head `q_b_proj` is [W_UQ; W_QR] concatenated per head `out_proj` is W_O ## Data-Movement Friendly Approach (i.e. "_forward_decode"): Runtime q_c = h_t @ W_DQ q_nope = (q_c @ W_UQ).view(-1, N, P) ql_nope = einsum("snh,lnh->snl", q, W_UK) q_pe = RoPE(q_c @ W_QR).view(Sq, N, R) new_kv_c = h_t @ W_DKV new_k_pe = RoPE(h_t @ W_KR) kv_c = torch.cat([new_kv_c, cache_kv_c], dim=0) k_pe = torch.cat([new_k_pe, cache_k_pe], dim=0) // MQA with QK headdim = Lkv + R // V headdim = Lkv // spda_o shape [Sq, N, Lkv] // NOTE: this is less compute-friendly since Lkv > P // but is more data-movement friendly since its MQA vs MHA spda_o = scaled_dot_product_attention( torch.cat([ql_nope, q_pe], dim=-1), torch.cat([kv_c, k_pe], dim=-1), kv_c ) o = einsum("snl,lnv->snv", spda_o.reshape(-1, N, Lkv), W_UV) return o.view(-1, N * V) @ self.num_heads @ W_O ## Chunked Prefill For chunked prefill we want to use the compute friendly algorithm. We are assuming sufficiently large Sq / Skv ratio, in the future may want to switch to the data-movement friendly approach if the chunk (i.e. `Sq`) is small. However, the compute-friendly approach can potentially run out of memory if Skv is large due to: `k_nope = (kv_c @ W_UK).view(Skv, N, P)` To mitigate this, we chunk the computation of attention with respect to the current context (i.e. `cache_kv_c` and `cache_k_pe`) so that we can used a fixed workspace size. The chunked prefill approach is as follows: MCC Max chunk of context to process per iter, computed dynamically, used to bound the memory usage q_c = h_t @ W_DQ q_nope = (q_c @ W_UQ).view(Sq, N, P) q_pe = RoPE(q_c @ W_QR).view(Sq, N, R) new_kv_c = h_t @ W_DKV new_k_pe = RoPE(h_t @ W_KR) new_k_nope = (new_kv_c @ W_UK.view(Lkv, N * P)).view(Sq, N, P) new_v = (new_kv_c @ W_UV.view(Lkv, N * V)).view(Sq, N, V) // MHA between queries and new KV // with QK headdim = P + R // V headdim = V // curr_o shape [Sq, N, V] // curr_lse shape [N, Sq], this is just order FA returns curr_o, curr_lse = scaled_dot_product_attention( torch.cat([q_nope, q_pe], dim=-1), torch.cat([new_k_nope, new_k_pe.unsqueeze(1).expand(-1, N, -1)], dim=-1), new_v, casual=True, return_softmax_lse=True ) // Compute attention with the already existing context for chunk_idx in range(cdiv(C, MCC)): chunk_start = chunk_idx * MCC chunk_end = min(chunk_start + MCC, C) Sc = chunk_end - chunk_start cache_kv_c_chunk = cache_kv_c[chunk_start:chunk_end] cache_k_pe_chunk = cache_k_pe[chunk_start:chunk_end] cache_k_nope_chunk = (cache_kv_c_chunk @ W_UK).view(-1, N, P) cache_v_chunk = (cache_kv_c_chunk @ W_UV).view(-1, N, V) chunk_o, chunk_lse = scaled_dot_product_attention( torch.cat([q_nope, q_pe], dim=-1), torch.cat([cache_k_nope_chunk, cache_k_pe_chunk.unsqueeze(1).expand(-1, N, -1)], dim=-1), cache_v_chunk, casual=False, return_softmax_lse=True ) curr_o, curr_lse = merge_attn_states( suffix_output=curr_o, suffix_lse=curr_lse, prefix_output=chunk_o, prefix_lse=chunk_lse, ) return curr_o @ W_O """ import functools from abc import abstractmethod from dataclasses import dataclass, field from typing import Generic, Optional, TypeVar, Union import torch from tqdm import tqdm import vllm.envs as envs import vllm_kunlun.platforms.envs as vllm_kunlun_envs from vllm import _custom_ops as ops from vllm.attention.backends.abstract import (AttentionBackend, AttentionLayer, AttentionMetadata, MLAAttentionImpl) from vllm.attention.backends.utils import get_mla_dims from vllm.attention.ops.common import cp_lse_ag_out_rs from vllm.attention.ops.merge_attn_states import merge_attn_states from vllm.attention.utils.fa_utils import get_flash_attn_version from vllm.config import VllmConfig, get_current_vllm_config from vllm.distributed.parallel_state import get_dcp_group, is_global_first_rank from vllm.logger import init_logger from vllm.model_executor.layers.linear import (ColumnParallelLinear, LinearBase, UnquantizedLinearMethod) from vllm.platforms import current_platform from vllm.utils import cdiv, round_down from vllm.utils.flashinfer import has_nvidia_artifactory from vllm.v1.attention.backends.utils import (AttentionMetadataBuilder, CommonAttentionMetadata, get_per_layer_parameters, infer_global_hyperparameters, split_decodes_and_prefills) from vllm.v1.kv_cache_interface import AttentionSpec import kunlun_ops try: from vllm.vllm_flash_attn import flash_attn_varlen_func is_vllm_fa = True except ImportError: # For rocm use upstream flash attention if current_platform.is_rocm(): from flash_attn import flash_attn_varlen_func is_vllm_fa = False try: from flashinfer import BatchPrefillWithRaggedKVCacheWrapper from flashinfer.prefill import ( # noqa: F401 cudnn_batch_prefill_with_kv_cache) flashinfer_available = True except ImportError: flashinfer_available = False def is_rocm_aiter_fp8bmm_enabled() -> bool: return current_platform.is_rocm() \ and envs.VLLM_ROCM_USE_AITER_FP8BMM \ and envs.VLLM_ROCM_USE_AITER if is_rocm_aiter_fp8bmm_enabled(): from aiter.ops.triton.batched_gemm_a8w8_a_per_token_group_prequant_w_per_batched_tensor_quant import ( # noqa: E501 # isort: skip batched_gemm_a8w8_a_per_token_group_prequant_w_per_batched_tensor_quant as aiter_triton_fp8_bmm) def dynamic_per_batched_tensor_quant( x: torch.Tensor, dtype: torch.dtype = torch.float8_e4m3fn): DTYPE_MAX = torch.finfo(dtype).max min_val, max_val = x.aminmax() amax = torch.maximum(min_val.abs(), max_val.abs()).clamp(min=1e-10) scale = DTYPE_MAX / amax x_scl_sat = (x * scale).clamp(min=-DTYPE_MAX, max=DTYPE_MAX) return x_scl_sat.to(dtype).contiguous(), scale.float().reciprocal() logger = init_logger(__name__) CUDNN_WORKSPACE_SIZE = 12800 class MLACommonBackend(AttentionBackend): accept_output_buffer: bool = True @staticmethod def get_name() -> str: return "TRITON_MLA" @staticmethod def get_metadata_cls() -> type["AttentionMetadata"]: return MLACommonMetadata @staticmethod def get_builder_cls() -> type["MLACommonMetadataBuilder"]: return MLACommonMetadataBuilder @staticmethod def get_kv_cache_shape( num_blocks: int, block_size: int, num_kv_heads: int, # assumed to be 1 for MLA head_size: int, cache_dtype_str: str = "auto", ) -> tuple[int, ...]: return (num_blocks, block_size, head_size) @classmethod def get_supported_dtypes(cls) -> list[torch.dtype]: return [torch.float16, torch.bfloat16] @classmethod def get_supported_head_sizes(cls) -> list[int]: return [576] @classmethod def validate_head_size(cls, head_size: int) -> None: supported_head_sizes = cls.get_supported_head_sizes() if head_size not in supported_head_sizes: attn_type = cls.__name__.removesuffix("Backend") raise ValueError( f"Head size {head_size} is not supported by {attn_type}. " f"Supported head sizes are: {supported_head_sizes}. " "Set VLLM_ATTENTION_BACKEND=FLEX_ATTENTION to use " "FlexAttention backend which supports all head sizes.") @dataclass class MLACommonPrefillMetadata: """ Prefill Specific Metadata """ @dataclass class ChunkedContextMetadata: # New for MLA (compared to FlashAttention) # For handling chunked prefill cu_seq_lens: torch.Tensor cu_seq_lens_cpu: torch.Tensor starts: torch.Tensor seq_tot: list[int] max_seq_lens: list[int] seq_lens: torch.Tensor workspace: torch.Tensor # for mla DCP cp_chunk_seq_lens: Optional[list[list[int]]] = None origin_context_lens: Optional[list[int]] = None cp_cu_seq_lens: Optional[torch.Tensor] = None chunk_size: Optional[int] = None cu_seq_lens_lst: Optional[list[list[int]]] = None block_table: torch.Tensor query_start_loc: torch.Tensor query_start_loc_cpu: torch.Tensor max_query_len: int chunked_context: Optional[ChunkedContextMetadata] = None @dataclass class FlashInferPrefillMetadata(MLACommonPrefillMetadata): prefill_main: Optional['BatchPrefillWithRaggedKVCacheWrapper'] = None prefill_chunks: list['BatchPrefillWithRaggedKVCacheWrapper'] = field( default_factory=list) @dataclass class CudnnPrefillMetadata(MLACommonPrefillMetadata): class ChunkedContextMetadata( MLACommonPrefillMetadata.ChunkedContextMetadata): seq_lens: torch.Tensor query_seq_lens: Optional[torch.Tensor] = None cudnn_workspace: Optional[torch.Tensor] = None @dataclass class MLACommonDecodeMetadata: block_table: torch.Tensor seq_lens: torch.Tensor D = TypeVar("D", bound=MLACommonDecodeMetadata) @dataclass class MLACommonMetadata(Generic[D]): """Metadata for MLACommon. NOTE: Please read the comment at the top of the file before trying to understand this class """ # NOTE(sang): Definition of context_len, query_len, and seq_len. # |---------- N-1 iteration --------| # |---------------- N iteration ---------------------| # |- tokenA -|......................|-- newTokens ---| # |---------- context_len ----------| # |-------------------- seq_len ---------------------| # |-- query_len ---| num_reqs: int max_query_len: int max_seq_len: int num_actual_tokens: int # Number of tokens excluding padding. query_start_loc: torch.Tensor query_start_loc_cpu: torch.Tensor slot_mapping: torch.Tensor # New for MLA (compared to FlashAttention) # For handling prefill decode split num_decodes: int num_decode_tokens: int num_prefills: int # The dimension of the attention heads head_dim: Optional[int] = None decode: Optional[D] = None prefill: Optional[Union[MLACommonPrefillMetadata, FlashInferPrefillMetadata, CudnnPrefillMetadata]] = None def __post_init__(self): if self.head_dim is not None: MLACommonBackend.validate_head_size(self.head_dim) M = TypeVar("M", bound=MLACommonMetadata) A = TypeVar("A") def use_flashinfer_prefill() -> bool: # For blackwell default to flashinfer prefill if it's available since # it is faster than FA2. return (not envs.VLLM_DISABLE_FLASHINFER_PREFILL and flashinfer_available and not envs.VLLM_USE_CUDNN_PREFILL and current_platform.is_device_capability(100)) def use_cudnn_prefill() -> bool: return (flashinfer_available and envs.VLLM_USE_CUDNN_PREFILL and current_platform.is_device_capability(100) and has_nvidia_artifactory()) # Currently 394MB, this can be tuned based on GEMM sizes used. # Chosen to be the same as sglang: # https://github.com/sgl-project/sglang/blob/766392c6bda2558b61ce6d1c1bfd8081a549e1f1/python/sglang/global_config.py#L37 FLASHINFER_WORKSPACE_BUFFER_SIZE = 394 * 1024 * 1024 class MLACommonMetadataBuilder(AttentionMetadataBuilder[M]): """ NOTE: Please read the comment at the top of the file before trying to understand this class """ reorder_batch_threshold: int = 1 @staticmethod def determine_chunked_prefill_workspace_size( vllm_config: VllmConfig) -> int: scheduler_config = vllm_config.scheduler_config cache_config = vllm_config.cache_config model_config = vllm_config.model_config chunked_prefill_workspace_size = min( # Try for 8 full length request or at least 4 pages per-request max(8 * model_config.max_model_len, 4 * scheduler_config.max_num_seqs * cache_config.block_size), # For long-context models try not to over-allocate limiting # kv-cache space, limiting it to 64k tokens, # which would result in the workspace being: # 2*(576)*(64*1024) = 144mb # (assuming 576 MLA head dim, and fp16) # which would result in up-projected context being # 2*(192*128)*(64*1024) = 3gb # (assuming 192 QK head dim, 128 heads, and fp16) 64 * 1024) # Enforce that we enough for at least 1 page per request chunked_prefill_workspace_size = max( chunked_prefill_workspace_size, scheduler_config.max_num_seqs * cache_config.block_size) return chunked_prefill_workspace_size def __init__(self, kv_cache_spec: AttentionSpec, layer_names: list[str], vllm_config: VllmConfig, device: torch.device, metadata_cls: Optional[type[M]] = None): self.metadata_cls = metadata_cls \ if metadata_cls is not None else MLACommonMetadata self.kv_cache_spec = kv_cache_spec scheduler_config = vllm_config.scheduler_config self.model_config = vllm_config.model_config parallel_config = vllm_config.parallel_config self.compilation_config = vllm_config.compilation_config self.device = device self.num_heads = self.model_config.get_num_attention_heads( parallel_config) self.mla_dims = get_mla_dims(self.model_config) self.aot_schedule = current_platform.is_cuda() try: self.dcp_world_size = get_dcp_group().world_size self.dcp_rank = get_dcp_group().rank_in_group except AssertionError: # DCP might not be initialized in testing self.dcp_world_size = 1 self.dcp_rank = 0 # Don't try to access the runner on AMD if self.aot_schedule: self.page_size = self.kv_cache_spec.block_size self.chunked_prefill_workspace_size = \ self.determine_chunked_prefill_workspace_size(vllm_config) if self.dcp_world_size > 1: # Note(hc): The local kvcache is incomplete when DCP is triggered, # an additional kvcache allgather across the DCP group is therefore # required, so the workspace has to be enlarged by 1/DCP relative # to the original TP allocation. assert self.chunked_prefill_workspace_size % \ self.dcp_world_size == 0 self.chunked_prefill_workspace = torch.empty( (self.chunked_prefill_workspace_size + self.chunked_prefill_workspace_size // self.dcp_world_size, self.model_config.get_head_size()), dtype=self.model_config.dtype, device=device, ) else: self.chunked_prefill_workspace = torch.empty( (self.chunked_prefill_workspace_size, self.model_config.get_head_size()), dtype=self.model_config.dtype, device=device, ) self._use_cudnn_prefill = use_cudnn_prefill() self._use_fi_prefill = use_flashinfer_prefill() self.prefill_metadata_cls = ( FlashInferPrefillMetadata if self._use_fi_prefill else CudnnPrefillMetadata if self._use_cudnn_prefill else MLACommonPrefillMetadata) if self._use_fi_prefill: self._workspace_buffer = torch.empty( FLASHINFER_WORKSPACE_BUFFER_SIZE, dtype=torch.uint8, device=device) self._fi_prefill_main: Optional[ BatchPrefillWithRaggedKVCacheWrapper] = None self._fi_prefill_chunks: list[ BatchPrefillWithRaggedKVCacheWrapper] = [] self._global_hyperparameters = infer_global_hyperparameters( get_per_layer_parameters(vllm_config, layer_names, MLACommonImpl)) if self._use_cudnn_prefill: self.cudnn_workspace = torch.empty( CUDNN_WORKSPACE_SIZE * scheduler_config.max_num_seqs, dtype=torch.int8, device=device, ) def _build_fi_prefill_wrappers(self, prefill: FlashInferPrefillMetadata): qo_indptr = prefill.query_start_loc has_context = False if prefill.chunked_context is not None: chunked_context = prefill.chunked_context has_context = True if self._fi_prefill_main is None: self._fi_prefill_main = BatchPrefillWithRaggedKVCacheWrapper( self._workspace_buffer, "NHD", backend="cutlass") if has_context: num_chunks = chunked_context.cu_seq_lens.shape[0] # Allocate more prefill chunk wrappers if needed if len(self._fi_prefill_chunks) < num_chunks: for _ in range(len(self._fi_prefill_chunks), num_chunks): self._fi_prefill_chunks.append( BatchPrefillWithRaggedKVCacheWrapper( self._workspace_buffer, "NHD", backend="cutlass")) assert num_chunks <= len(self._fi_prefill_chunks) # In MLA, the non-latent num_qo_heads == num_kv_heads num_qo_heads = self.num_heads num_kv_heads = num_qo_heads # Sanity: Verify that num_kv_heads == 1 since it is latent space assert self.kv_cache_spec.num_kv_heads == 1 # Get non-latent head_dim_qk and head_dim_vo head_dim_qk = (self.mla_dims.qk_nope_head_dim + self.mla_dims.qk_rope_head_dim) head_dim_vo = self.mla_dims.v_head_dim # For main run, qo_indptr == kv_indptr kv_indptr = qo_indptr.clone() # Prepare main prefill self._fi_prefill_main.plan( qo_indptr=qo_indptr, kv_indptr=kv_indptr, num_qo_heads=num_qo_heads, num_kv_heads=num_kv_heads, head_dim_qk=head_dim_qk, head_dim_vo=head_dim_vo, causal=True, # This is main run sm_scale=self._global_hyperparameters.sm_scale, window_left=self._global_hyperparameters.window_left, logits_soft_cap=self._global_hyperparameters.logits_soft_cap, q_data_type=self.model_config.dtype, ) # Prepare context prefills if has_context: for i in range(num_chunks): kv_indptr_chunk = chunked_context.cu_seq_lens[i] self._fi_prefill_chunks[i].plan( qo_indptr=qo_indptr, kv_indptr=kv_indptr_chunk, num_qo_heads=num_qo_heads, num_kv_heads=num_kv_heads, head_dim_qk=head_dim_qk, head_dim_vo=head_dim_vo, causal=False, # This is context run sm_scale=self._global_hyperparameters.sm_scale, window_left=self._global_hyperparameters.window_left, logits_soft_cap=self._global_hyperparameters. logits_soft_cap, q_data_type=self.model_config.dtype, ) prefill.prefill_main = self._fi_prefill_main prefill.prefill_chunks = self._fi_prefill_chunks def _build_decode(self, block_table_tensor: torch.Tensor, seq_lens_cpu: torch.Tensor, seq_lens_device: torch.Tensor, query_start_loc_cpu: torch.Tensor, query_start_loc_device: torch.Tensor, num_decode_tokens: int) -> MLACommonDecodeMetadata: return MLACommonDecodeMetadata( block_table=block_table_tensor, seq_lens=seq_lens_device, ) def build_for_cudagraph_capture( self, common_attn_metadata: CommonAttentionMetadata) -> M: """ This method builds the metadata for full cudagraph capture. Currently, only decode is supported for full cudagraphs with MLA. """ m = common_attn_metadata assert m.num_reqs <= (m.num_actual_tokens * self.reorder_batch_threshold), \ "MLA only supports decode-only full CUDAGraph capture. " \ "Make sure all cudagraph capture sizes <= max_num_seq." assert m.max_query_len <= self.reorder_batch_threshold # decode only return self.build(0, m) def build(self, common_prefix_len: int, common_attn_metadata: CommonAttentionMetadata, fast_build: bool = False) -> M: num_reqs = common_attn_metadata.num_reqs num_tokens = common_attn_metadata.num_actual_tokens max_query_len = common_attn_metadata.max_query_len max_seq_len = common_attn_metadata.max_seq_len # Note(simon): be careful about the CPU <> GPU memory movement in this # function. We should avoid GPU -> CPU sync as much as possible because # it blocks on all previous kernels. device = self.device block_table_tensor = common_attn_metadata.block_table_tensor slot_mapping = common_attn_metadata.slot_mapping query_start_loc = common_attn_metadata.query_start_loc query_start_loc_cpu = common_attn_metadata.query_start_loc_cpu seq_lens = common_attn_metadata.seq_lens seq_lens_cpu = common_attn_metadata.seq_lens_cpu query_seq_lens_cpu = query_start_loc_cpu[1:] - query_start_loc_cpu[:-1] num_computed_tokens_cpu = (common_attn_metadata.seq_lens_cpu - query_seq_lens_cpu) num_decodes, num_prefills, num_decode_tokens, num_prefill_tokens = \ split_decodes_and_prefills(common_attn_metadata, decode_threshold=self.reorder_batch_threshold) # Note(hc): update seq_lens of decode reqs under DCP. if self.dcp_world_size > 1: seq_lens[:num_decodes] = seq_lens[:num_decodes] \ // self.dcp_world_size + (self.dcp_rank <= \ (seq_lens[:num_decodes] - 1) % self.dcp_world_size) assert num_decodes + num_prefills == num_reqs assert num_decode_tokens + num_prefill_tokens == num_tokens prefill_metadata = None if num_prefills > 0: reqs_start = num_decodes # prefill_start context_lens_cpu = num_computed_tokens_cpu[reqs_start:num_reqs] # Note(hc): The context lengths in the perspective of dcp rank0. cp_context_lens_cpu = torch.ceil(context_lens_cpu.float() / self.dcp_world_size).int() origin_context_lens = context_lens_cpu.tolist() max_context_len_cpu = context_lens_cpu.max().item() num_prefills_with_context_cpu = (context_lens_cpu > 0).sum().item() prefill_query_start_loc = query_start_loc[ reqs_start:] - query_start_loc[reqs_start] chunked_context_metadata = None if max_context_len_cpu > 0: # NOTE: it is recommend you read the `Chunked Prefill` section # in the comment at the top of the file before trying to # understand the following code # currently we allocate an equal amount of workspace for each # prefill in the batch, we could probably use a more advanced # algorithm here and allocate more workspace to prefills with # longer context lengths max_context_chunk = (self.chunked_prefill_workspace_size // num_prefills_with_context_cpu) if self.aot_schedule: # align max_context_chunk to page_size by rounding down, # currently the `gather_and_maybe_dequant_cache` kernel # cannot handle `context_chunk_starts` that are not aligned # to page_size max_context_chunk = round_down(max_context_chunk, self.page_size) assert max_context_chunk > 0 num_chunks = cdiv(max_context_len_cpu, max_context_chunk) # if `max_context_chunk = 256`, `num_chunks = 3`, and # `num_prefills_with_context = 4`, create a tensor that looks # like # [[0, 0, 0, 0], [256, 256, 256, 256], [512, 512, 512, 512]] # Note(simon): this is done in CPU because of downstream's # of `to_list`. chunk_starts = \ torch.arange(num_chunks, dtype=torch.int32) \ .unsqueeze(1).expand(-1, num_prefills) \ * max_context_chunk chunk_ends = torch.min(context_lens_cpu.unsqueeze(0), chunk_starts + max_context_chunk) chunk_seq_lens = (chunk_ends - chunk_starts).clamp(min=0) cu_seq_lens_cpu = torch.zeros(num_chunks, num_prefills + 1, dtype=torch.int32, pin_memory=True) torch.cumsum(chunk_seq_lens, dim=1, out=cu_seq_lens_cpu[:, 1:], dtype=torch.int32) if self.dcp_world_size > 1: # Note(hc): The above max_context_chunk already enforces # block_size alignment, DCP just need the block_size can # be divisible by dcp_world_size, because DCP use # cp_gather_cache which not require `cp_chunk_starts` # aligned to page_size. assert max_context_chunk % self.dcp_world_size == 0 cp_max_context_chunk = max_context_chunk // \ self.dcp_world_size cp_chunk_starts = \ torch.arange(num_chunks, dtype=torch.int32) \ .unsqueeze(1).expand(-1, num_prefills) \ * cp_max_context_chunk cp_chunk_ends = torch.min( cp_context_lens_cpu.unsqueeze(0), cp_chunk_starts + cp_max_context_chunk) cp_chunk_seq_lens = (cp_chunk_ends - cp_chunk_starts).clamp(min=0) cp_cu_seq_lens_cpu = torch.zeros(num_chunks, num_prefills + 1, dtype=torch.int32, pin_memory=True) torch.cumsum(cp_chunk_seq_lens, dim=1, out=cp_cu_seq_lens_cpu[:, 1:], dtype=torch.int32) chunked_context_metadata_cls = \ CudnnPrefillMetadata.ChunkedContextMetadata \ if self._use_cudnn_prefill else \ MLACommonPrefillMetadata.ChunkedContextMetadata if self.dcp_world_size > 1: chunked_context_metadata = \ chunked_context_metadata_cls( cu_seq_lens=cu_seq_lens_cpu \ .to(device, non_blocking=True), cu_seq_lens_cpu=cu_seq_lens_cpu, starts=cp_chunk_starts.to(device, non_blocking=True), seq_tot=cp_chunk_seq_lens.sum(dim=1).tolist(), max_seq_lens=chunk_seq_lens.max(dim=1).values.tolist(), seq_lens=chunk_seq_lens, workspace=self.chunked_prefill_workspace, cp_chunk_seq_lens=cp_chunk_seq_lens.tolist(), origin_context_lens=origin_context_lens, cp_cu_seq_lens=cp_cu_seq_lens_cpu \ .to(device, non_blocking=True), chunk_size=max_context_chunk, cu_seq_lens_lst=cu_seq_lens_cpu.tolist(), ) else: chunked_context_metadata = \ chunked_context_metadata_cls( cu_seq_lens=cu_seq_lens_cpu \ .to(device, non_blocking=True), cu_seq_lens_cpu=cu_seq_lens_cpu, starts=chunk_starts.to(device, non_blocking=True), seq_tot=chunk_seq_lens.sum(dim=1).tolist(), max_seq_lens=chunk_seq_lens.max(dim=1).values.tolist(), seq_lens=chunk_seq_lens, workspace=self.chunked_prefill_workspace, ) if self._use_cudnn_prefill: chunked_context_metadata.seq_lens = chunk_seq_lens assert max(chunked_context_metadata.max_seq_lens) <= \ self.chunked_prefill_workspace_size prefill_metadata = self.prefill_metadata_cls( block_table=block_table_tensor[reqs_start:, ...], query_start_loc=prefill_query_start_loc, query_start_loc_cpu=prefill_query_start_loc.cpu(), max_query_len=max_query_len, chunked_context=chunked_context_metadata, ) if self._use_cudnn_prefill: assert isinstance(prefill_metadata, CudnnPrefillMetadata) prefill_metadata.query_seq_lens = prefill_query_start_loc[1:] \ - prefill_query_start_loc[:-1] prefill_metadata.cudnn_workspace = self.cudnn_workspace decode_metadata = None if num_decodes > 0: decode_metadata = self._build_decode( block_table_tensor=block_table_tensor[:num_decodes, ...], seq_lens_cpu=seq_lens_cpu[:num_decodes], seq_lens_device=seq_lens[:num_decodes], query_start_loc_cpu=query_start_loc_cpu[:num_decodes + 1], query_start_loc_device=query_start_loc[:num_decodes + 1], num_decode_tokens=num_decode_tokens, ) attn_metadata = self.metadata_cls( num_reqs=common_attn_metadata.num_reqs, max_query_len=common_attn_metadata.max_query_len, max_seq_len=max_seq_len, num_actual_tokens=num_tokens, query_start_loc=query_start_loc, query_start_loc_cpu=query_start_loc_cpu, slot_mapping=slot_mapping, head_dim=self.model_config.get_head_size(), # MLACommonMetadata Chunk prefill specific num_decodes=num_decodes, num_decode_tokens=num_decode_tokens, num_prefills=num_prefills, prefill=prefill_metadata, decode=decode_metadata, ) if self._use_fi_prefill and num_prefills > 0: assert isinstance(attn_metadata.prefill, FlashInferPrefillMetadata) self._build_fi_prefill_wrappers(attn_metadata.prefill) return attn_metadata def reorg_kvcache( allgatered_kv_c_normed: torch.Tensor, allgatered_k_pe: torch.Tensor, cp_chunk_seq_lens_lst: list[int], origin_context_lens: list[int], cp_world_size: int, sum_seq_len: int, max_seq_len: int, chunk_size: int, chunk_idx: int, toks: int, ) -> tuple[torch.Tensor, torch.Tensor]: """ reorg kvcache after cp local gather to tp layout for attn kernel. Args: cp_chunk_seq_lens_lst: chunk context lengths under CP. origin_context_lens: origin full context lengths under CP. cp_world_size: CP size. sum_seq_len: the sum of cp_chunk_seq_lens_lst. max_seq_len: the max value of cp_chunk_seq_lens_lst. chunk_size: equals to max_context_chunk from chunked_context_metadata building. chunk_idx: chunk idx of chunked_prefill. toks: the number of tokens for local gather cache. """ kv_c_segments = [] k_pe_segments = [] src_token_idx = 0 max_seq_len_check = 0 for cp_chunk_seq_len, origin_context_len in zip(cp_chunk_seq_lens_lst, origin_context_lens): chunk_context_len = chunk_size if cp_chunk_seq_len != 0: chunk_context_len = min( chunk_context_len, origin_context_len - chunk_size * chunk_idx) cp_target_rank = (chunk_context_len - 1) % cp_world_size cur_seq_len = 0 for rank in range(cp_world_size): if rank > cp_target_rank and cp_chunk_seq_len: real_cp_chunk_seq_len = cp_chunk_seq_len - 1 else: real_cp_chunk_seq_len = cp_chunk_seq_len if real_cp_chunk_seq_len: kv_c_segment = allgatered_kv_c_normed[rank * toks + src_token_idx:rank * toks + src_token_idx + real_cp_chunk_seq_len] k_pe_segment = allgatered_k_pe[rank * toks + src_token_idx:rank * toks + src_token_idx + real_cp_chunk_seq_len] kv_c_segments.append(kv_c_segment) k_pe_segments.append(k_pe_segment) cur_seq_len += real_cp_chunk_seq_len max_seq_len_check = max(max_seq_len_check, cur_seq_len) src_token_idx += cp_chunk_seq_len reorganized_kv_c_normed = torch.cat(kv_c_segments, dim=0) reorganized_k_pe = torch.cat(k_pe_segments, dim=0) assert reorganized_kv_c_normed.shape[0] == sum_seq_len assert reorganized_k_pe.shape[0] == sum_seq_len assert max_seq_len_check == max_seq_len return reorganized_kv_c_normed, reorganized_k_pe # TODO(Lucas): rename MLACommonBaseImpl -> MLACommonImpl, # and MLACommonImpl -> MLACommonDenseImpl or somthing like that class MLACommonBaseImpl(MLAAttentionImpl[A], Generic[A]): """ NOTE: Please read the comment at the top of the file before trying to understand this class """ def __init__( self, num_heads: int, head_size: int, scale: float, num_kv_heads: int, alibi_slopes: Optional[list[float]], sliding_window: Optional[int], kv_cache_dtype: str, logits_soft_cap: Optional[float], attn_type: str, kv_sharing_target_layer_name: Optional[str], # MLA Specific Arguments q_lora_rank: Optional[int], kv_lora_rank: int, qk_nope_head_dim: int, qk_rope_head_dim: int, qk_head_dim: int, v_head_dim: int, kv_b_proj: ColumnParallelLinear, indexer=None, q_pad_num_heads: Optional[int] = None, ) -> None: if kv_sharing_target_layer_name is not None: raise NotImplementedError("KV sharing is not supported for MLA") self.num_heads = num_heads self.head_size = head_size self.scale = float(scale) self.num_kv_heads = num_kv_heads self.kv_cache_dtype = kv_cache_dtype self.q_lora_rank = q_lora_rank self.kv_lora_rank = kv_lora_rank self.qk_nope_head_dim = qk_nope_head_dim self.qk_rope_head_dim = qk_rope_head_dim self.qk_head_dim = qk_head_dim self.v_head_dim = v_head_dim self.kv_b_proj = kv_b_proj self.indexer = indexer self.q_pad_num_heads = q_pad_num_heads def process_weights_after_loading(self, act_dtype: torch.dtype): def get_layer_weight(layer): WEIGHT_NAMES = ("weight", "qweight", "weight_packed") for attr in WEIGHT_NAMES: if hasattr(layer, attr): return getattr(layer, attr) raise AttributeError( f"Layer '{layer}' has no recognized weight attribute:" f" {WEIGHT_NAMES}.") def get_and_maybe_dequant_weights(layer: LinearBase): if not isinstance(layer.quant_method, UnquantizedLinearMethod): # NOTE: This should only be used offline, since it's O(N^3) eye = torch.eye(layer.input_size_per_partition, dtype=act_dtype, device=get_layer_weight(layer).device) dequant_weights = layer.quant_method.apply(layer, eye, bias=None) del eye # standardize to (output, input) return dequant_weights.T return layer.weight # we currently do not have quantized bmm's which are needed for # `W_UV` and `W_UK_T`, we just store fp16/bf16 copies and perform # the bmm's in 16-bit, the extra memory overhead of this is fairly low kv_b_proj_weight = get_and_maybe_dequant_weights(self.kv_b_proj).T assert kv_b_proj_weight.shape == ( self.kv_lora_rank, self.num_heads * (self.qk_nope_head_dim + self.v_head_dim)), ( f"{kv_b_proj_weight.shape=}, " f"{self.kv_lora_rank=}, " f"{self.num_heads=}, " f"{self.qk_nope_head_dim=}, " f"{self.v_head_dim=}") kv_b_proj_weight = kv_b_proj_weight.view( self.kv_lora_rank, self.num_heads, self.qk_nope_head_dim + self.v_head_dim, ) W_UK, W_UV = kv_b_proj_weight.split( [self.qk_nope_head_dim, self.v_head_dim], dim=-1) if is_rocm_aiter_fp8bmm_enabled(): W_K = W_UK.transpose(0, 1) # 16 512 128 W_V = W_UV.permute(1, 2, 0) # 16 128 512 self.W_K, self.W_K_scale = dynamic_per_batched_tensor_quant( W_K, dtype=current_platform.fp8_dtype()) self.W_V, self.W_V_scale = dynamic_per_batched_tensor_quant( W_V, dtype=current_platform.fp8_dtype()) # The kernel operates on non-padded inputs. Hence, pre-compiling # triton kernel to avoid runtime compilation for unseen batch sizes # Pre-compile for batch sizes 1 to 1024 to cover most use-cases. # On DS-R1, this step adds roughly 50s to the model loading time. max_batch_size = 1024 # [ToDo] Find the optimal upper limit pre_compilation_list = list(range(1, max_batch_size + 1)) if is_global_first_rank(): pre_compilation_list = tqdm( pre_compilation_list, desc="[Aiter Triton] Pre-compiling fp8 BMM kernel", total=max_batch_size, ) for m in pre_compilation_list: x = torch.empty((self.W_K.shape[0], m, self.W_K.shape[2]), dtype=torch.bfloat16, device=self.W_K.device) aiter_triton_fp8_bmm(x, self.W_K, self.W_K_scale, group_size=128, transpose_bm=True) x = torch.empty((self.W_V.shape[0], m, self.W_V.shape[2]), dtype=torch.bfloat16, device=self.W_V.device) aiter_triton_fp8_bmm(x, self.W_V, self.W_V_scale, group_size=128, transpose_bm=True) else: # Convert from (L, N, V) to (N, L, V) self.W_UV = W_UV.transpose(0, 1) # Convert from (L, N, P) to (N, P, L) self.W_UK_T = W_UK.permute(1, 2, 0) def _v_up_proj(self, x: torch.Tensor, out: torch.Tensor): # Convert from (B, N, L) to (N, B, L) if is_rocm_aiter_fp8bmm_enabled(): # Multiply + Transpose (N, B, L) x (N, L, V)->(N, B, V)->(B, N, V) x = aiter_triton_fp8_bmm(x, self.W_V, self.W_V_scale, group_size=128, transpose_bm=True) # Convert from (B, N, V) to (B, N * V) x = x.reshape(-1, self.num_heads * self.v_head_dim) # Copy result out.copy_(x) else: if vllm_kunlun_envs.VLLM_KUNLUN_ENABLE_INT8_BMM: x = x.view(-1, self.num_heads, self.kv_lora_rank) out = out.view(-1, self.num_heads, self.v_head_dim) q_len = x.shape[0] extra_params = {"trans": False} sorted_tokens_num_lod = torch.arange( self.num_heads + 1, dtype=torch.int, device="cuda" ) * q_len sorted_tokens_idx = torch.arange( self.num_heads * q_len, dtype=torch.int, device="cuda") kunlun_ops.mla_bmm_I8( x.contiguous(), # [1, 16, 512] torch.float16 self.W_UV, # [16, 128, 512] torch.int8 self.W_UV_SCALE, # [2048, 1] torch.float32 out, # [1, 16, 128] torch.float16 sorted_tokens_num_lod, # [17] sorted_tokens_idx, # [16] **extra_params ) # out_new = out.reshape(-1, self.num_heads * self.v_head_dim) # out.resize_(origin_out_shape) # out.copy_(out_new) else: x = x.view(-1, self.num_heads, self.kv_lora_rank).transpose(0, 1) # Convert from (B, N * V) to (N, B, V) out = out.view(-1, self.num_heads, self.v_head_dim).transpose(0, 1) # Multiply (N, B, L) x (N, L, V) -> (N, B, V) torch.bmm(x, self.W_UV, out=out) # Reuse "out" to make it "hot" # Convert from (N, B, V) to (B, N * V) out_new = out.transpose(0, 1).reshape( -1, self.num_heads * self.v_head_dim) # Adjust output buffer shape back to the original (B, N * V) N, B, V = out.shape out.resize_((B, N * V)) out.copy_(out_new) # Copy result class MLACommonImpl(MLACommonBaseImpl[M], Generic[M]): """ NOTE: Please read the comment at the top of the file before trying to understand this class """ def __init__(self, *args, **kwargs) -> None: super().__init__(*args, **kwargs) if use_flashinfer_prefill(): logger.debug_once("Using FlashInfer prefill for MLA") self._run_prefill_context_chunk = self._run_prefill_context_chunk_fi self._run_prefill_new_tokens = self._run_prefill_new_tokens_fi self._pad_v = False elif use_cudnn_prefill(): logger.debug_once("Using CUDNN prefill for MLA") self._run_prefill_context_chunk = \ self._run_prefill_context_chunk_cudnn self._run_prefill_new_tokens = self._run_prefill_new_tokens_cudnn self._pad_v = False else: # Use FlashAttention logger.debug_once("Using FlashAttention prefill for MLA") self._run_prefill_context_chunk = self._run_prefill_context_chunk_fa self._run_prefill_new_tokens = self._run_prefill_new_tokens_fa # Handle the differences between the flash_attn_varlen from # flash_attn and the one from vllm_flash_attn. The former is used on # RoCM and the latter has an additional parameter to control # FA2 vs FA3 self.flash_attn_varlen_func = flash_attn_varlen_func self.vllm_flash_attn_version = get_flash_attn_version() if self.vllm_flash_attn_version is not None: self.flash_attn_varlen_func = \ functools.partial(flash_attn_varlen_func, fa_version=self.vllm_flash_attn_version) # For MLA the v head dim is smaller than qk head dim so we pad out # v with 0s to match the qk head dim for attention backends that do # not support different headdims # We don't need to pad V if we are on a hopper system with FA3 self._pad_v = self.vllm_flash_attn_version is None or not ( self.vllm_flash_attn_version == 3 and current_platform.get_device_capability()[0] == 9) self.dcp_world_size: Optional[int] = None self.chunked_prefill_workspace_size = \ MLACommonMetadataBuilder.determine_chunked_prefill_workspace_size( get_current_vllm_config()) def _flash_attn_varlen_diff_headdims(self, q, k, v, context_seq_lod_xpu=None, context_seq_lod_cpu=None, return_softmax_lse=False, causal=True, softmax_scale=None, **kwargs): maybe_padded_v = v if self._pad_v: maybe_padded_v = torch.nn.functional.pad( v, [0, q.shape[-1] - v.shape[-1]], value=0) if is_vllm_fa: kwargs["return_softmax_lse"] = return_softmax_lse else: # ROCm leverages the upstream flash_attn, which takes a parameter # called "return_attn_probs" instead of return_softmax_lse kwargs["return_attn_probs"] = return_softmax_lse # attn_out = self.flash_attn_varlen_func( # q=q, # k=k, # v=maybe_padded_v, # softmax_scale=softmax_scale, # **kwargs, # ) attn_out = torch.empty_like(q) ds_alpha = 1.8738542070926265 tp_q_head_num=q.size(1) softmax_lse = torch.zeros(tp_q_head_num, q.size(0), dtype=torch.float32, device=q.device) softmax_lse.fill_(float('-inf')) kunlun_ops.attention( q=q, k_cache=k, v_cache=maybe_padded_v, out=attn_out, is_causal=causal, is_prefill=True, prefill_len=0, k_perchannel_scale=None, v_perchannel_scale=None, smooth=None, context_seq_lod_cpu=context_seq_lod_cpu, context_seq_lod_xpu=context_seq_lod_xpu, slot_mapping_cpu=None, slot_mapping_xpu=None, v_trans=False, v_trans_threshold=0, alpha=ds_alpha, softmax_lse=softmax_lse ) # Unpack the output if there is multiple results lse = None if isinstance(attn_out, tuple): attn_out, lse = attn_out[0], attn_out[1] # Remain consistent with old `flash_attn_varlen_func` where there # is only one output tensor if `return_softmax_lse` is False. if return_softmax_lse: return attn_out, softmax_lse return attn_out def _run_prefill_new_tokens_fa(self, prefill: MLACommonPrefillMetadata, q, k, v, return_softmax_lse): return self._flash_attn_varlen_diff_headdims( q=q, k=k, v=v, context_seq_lod_xpu=prefill.query_start_loc, context_seq_lod_cpu=prefill.query_start_loc_cpu, softmax_scale=self.scale, causal=True, return_softmax_lse=return_softmax_lse, ) def _run_prefill_new_tokens_fi(self, prefill: MLACommonPrefillMetadata, q, k, v, return_softmax_lse): assert isinstance(prefill, FlashInferPrefillMetadata) assert prefill.prefill_main is not None ret = prefill.prefill_main.run( q=q, k=k, v=v, return_lse=return_softmax_lse, ) if isinstance(ret, tuple): # Convert from (q_len, num_heads) to (num_heads, q_len) return ret[0], ret[1].transpose(0, 1).contiguous() return ret def _run_prefill_new_tokens_cudnn(self, prefill: MLACommonPrefillMetadata, q, k, v, return_softmax_lse): assert isinstance(prefill, CudnnPrefillMetadata) assert prefill.query_seq_lens is not None output, lse = cudnn_batch_prefill_with_kv_cache( q=q, k_cache=k, v_cache=v, scale=self.scale, workspace_buffer=prefill.cudnn_workspace, max_token_per_sequence=prefill.max_query_len, max_sequence_kv=prefill.max_query_len, actual_seq_lens_q=prefill.query_seq_lens.view(-1, 1, 1, 1), actual_seq_lens_kv=prefill.query_seq_lens.view(-1, 1, 1, 1), causal=True, return_lse=True, # do not support False for now is_cuda_graph_compatible= True, #Indicates actual_seq_lens are on GPU or CPU. ) if return_softmax_lse: return output, lse return output def _run_prefill_context_chunk_fa(self, prefill: MLACommonPrefillMetadata, chunk_idx: int, q, k, v): assert prefill.chunked_context is not None return self._flash_attn_varlen_diff_headdims( q=q, k=k, v=v, context_seq_lod_xpu=prefill.query_start_loc, context_seq_lod_cpu=prefill.chunked_context.cu_seq_lens_cpu[chunk_idx], softmax_scale=self.scale, causal=False, return_softmax_lse=True, ) def _run_prefill_context_chunk_fi(self, prefill: MLACommonPrefillMetadata, chunk_idx: int, q, k, v): assert isinstance(prefill, FlashInferPrefillMetadata) attn_out, lse = prefill.prefill_chunks[chunk_idx].run( q=q, k=k, v=v, return_lse=True, ) # Convert from (q_len, num_heads) to (num_heads, q_len) return attn_out, lse.transpose(0, 1).contiguous() def _run_prefill_context_chunk_cudnn(self, prefill: MLACommonPrefillMetadata, chunk_idx: int, q, k, v): assert isinstance(prefill, CudnnPrefillMetadata) assert prefill.chunked_context is not None assert prefill.chunked_context.seq_lens[chunk_idx] is not None assert prefill.query_seq_lens is not None return cudnn_batch_prefill_with_kv_cache( q=q, k_cache=k, v_cache=v, scale=self.scale, workspace_buffer=prefill.cudnn_workspace, max_token_per_sequence=prefill.max_query_len, max_sequence_kv=prefill.chunked_context.max_seq_lens[chunk_idx], actual_seq_lens_q=prefill.query_seq_lens.view(-1, 1, 1, 1), actual_seq_lens_kv=prefill.chunked_context.seq_lens[chunk_idx]. view(-1, 1, 1, 1), causal=False, return_lse=True, is_cuda_graph_compatible= True, #Indicates actual_seq_lens are on GPU or CPU. ) def process_weights_after_loading(self, act_dtype: torch.dtype): def get_layer_weight(layer): WEIGHT_NAMES = ("weight", "qweight", "weight_packed") for attr in WEIGHT_NAMES: if hasattr(layer, attr): return getattr(layer, attr) raise AttributeError( f"Layer '{layer}' has no recognized weight attribute:" f" {WEIGHT_NAMES}.") def get_layer_weight_scale(layer): WEIGHT_SCALE_NAMES = ("weight_scale",) for attr in WEIGHT_SCALE_NAMES: if hasattr(layer, attr): return getattr(layer, attr) raise AttributeError( f"Layer '{layer}' has no recognized weight scale attribute:" f" {WEIGHT_SCALE_NAMES}.") def get_and_maybe_dequant_weights(layer: LinearBase): if not isinstance(layer.quant_method, UnquantizedLinearMethod): # NOTE: This should only be used offline, since it's O(N^3) eye = torch.eye(layer.input_size_per_partition, dtype=act_dtype, device=get_layer_weight(layer).device) dequant_weights = layer.quant_method.apply(layer, eye, bias=None) del eye # standardize to (output, input) return dequant_weights.T return layer.weight if vllm_kunlun_envs.VLLM_KUNLUN_ENABLE_INT8_BMM: kv_b_proj_weight = get_layer_weight(self.kv_b_proj).T kv_b_proj_weight_scale = get_layer_weight_scale(self.kv_b_proj) assert kv_b_proj_weight.dtype == torch.int8, \ f"weight type {kv_b_proj_weight.dtype} not support for int8 MLA BMM" W_UK, W_UV = kv_b_proj_weight.unflatten( 0, (-1, self.qk_nope_head_dim + self.v_head_dim) ).split([self.qk_nope_head_dim, self.v_head_dim], dim=1) W_UK_SCALE, W_UV_SCALE = kv_b_proj_weight_scale.unflatten( 0, (-1, self.qk_nope_head_dim + self.v_head_dim) ).split([self.qk_nope_head_dim, self.v_head_dim], dim=1) W_UK_SCALE = W_UK_SCALE / 127.0 w_uk_dq = W_UK.contiguous().cpu().to(torch.bfloat16).to(kv_b_proj_weight.device) \ * W_UK_SCALE.contiguous().to(torch.bfloat16) w_uk_dq_trans = w_uk_dq.transpose(1, 2).contiguous() self.W_UK_T = W_UK.transpose(1, 2).contiguous() self.W_UK_SCALE = torch.empty([W_UK.shape[0] * W_UK.shape[2], 1], dtype=torch.float, device=kv_b_proj_weight.device) kunlun_ops.quant2d(w_uk_dq_trans, self.W_UK_T, self.W_UK_SCALE) self.W_UV = W_UV.contiguous() self.W_UV_SCALE = W_UV_SCALE.contiguous().reshape(-1, 1) else: # we currently do not have quantized bmm's which are needed for # `W_UV` and `W_UK_T`, we just store fp16/bf16 copies and perform # the bmm's in 16-bit, the extra memory overhead of this is fairly low kv_b_proj_weight = get_and_maybe_dequant_weights(self.kv_b_proj).T assert kv_b_proj_weight.shape == ( self.kv_lora_rank, self.num_heads * (self.qk_nope_head_dim + self.v_head_dim)), ( f"{kv_b_proj_weight.shape=}, " f"{self.kv_lora_rank=}, " f"{self.num_heads=}, " f"{self.qk_nope_head_dim=}, " f"{self.v_head_dim=}") kv_b_proj_weight = kv_b_proj_weight.view( self.kv_lora_rank, self.num_heads, self.qk_nope_head_dim + self.v_head_dim, ) W_UK, W_UV = kv_b_proj_weight.split( [self.qk_nope_head_dim, self.v_head_dim], dim=-1) if is_rocm_aiter_fp8bmm_enabled(): W_K = W_UK.transpose(0, 1) # 16 512 128 W_V = W_UV.permute(1, 2, 0) # 16 128 512 self.W_K, self.W_K_scale = dynamic_per_batched_tensor_quant( W_K, dtype=current_platform.fp8_dtype()) self.W_V, self.W_V_scale = dynamic_per_batched_tensor_quant( W_V, dtype=current_platform.fp8_dtype()) # The kernel operates on non-padded inputs. Hence, pre-compiling # triton kernel to avoid runtime compilation for unseen batch sizes # Pre-compile for batch sizes 1 to 1024 to cover most use-cases. # On DS-R1, this step adds roughly 50s to the model loading time. max_batch_size = 1024 # [ToDo] Find the optimal upper limit pre_compilation_list = list(range(1, max_batch_size + 1)) if is_global_first_rank(): pre_compilation_list = tqdm( pre_compilation_list, desc="[Aiter Triton] Pre-compiling fp8 BMM kernel", total=max_batch_size, ) for m in pre_compilation_list: x = torch.empty((self.W_K.shape[0], m, self.W_K.shape[2]), dtype=torch.bfloat16, device=self.W_K.device) aiter_triton_fp8_bmm(x, self.W_K, self.W_K_scale, group_size=128, transpose_bm=True) x = torch.empty((self.W_V.shape[0], m, self.W_V.shape[2]), dtype=torch.bfloat16, device=self.W_V.device) aiter_triton_fp8_bmm(x, self.W_V, self.W_V_scale, group_size=128, transpose_bm=True) else: # Convert from (L, N, V) to (N, L, V) self.W_UV = W_UV.transpose(0, 1) # Convert from (L, N, P) to (N, P, L) self.W_UK_T = W_UK.permute(1, 2, 0) def gather_and_maybe_dequant_cache_py_optimized( self, src_cache: torch.Tensor, dst: torch.Tensor, block_table: torch.Tensor, cu_seq_lens: torch.Tensor, batch_size: int, kv_cache_dtype: str, scale: torch.Tensor, seq_starts: Optional[torch.Tensor] = None ) -> None: device = src_cache.device num_blocks, block_size, head_dim = src_cache.shape tot_tokens = dst.shape[0] src_cache_2d = src_cache.reshape(-1, head_dim) token_ids = torch.arange(tot_tokens, device=device, dtype=torch.int32) batch_ids = torch.zeros(tot_tokens, dtype=torch.int32, device=device) for i in range(batch_size): start = cu_seq_lens[i] end = cu_seq_lens[i + 1] mask = (token_ids >= start) & (token_ids < end) batch_ids[mask] = i batch_starts = cu_seq_lens[batch_ids] batch_offsets = token_ids - batch_starts if seq_starts is not None: seq_start_offsets = seq_starts[batch_ids] batch_offsets = batch_offsets + seq_start_offsets block_table_ids = batch_offsets // block_size slot_ids = batch_offsets % block_size block_table_flat = block_table.view(-1) block_table_stride = block_table.shape[1] block_table_indices = batch_ids * block_table_stride + block_table_ids block_ids = block_table_flat[block_table_indices] valid_mask = block_ids >= 0 linear_indices = torch.where( valid_mask, block_ids * block_size + slot_ids, torch.zeros_like(block_ids, device=device) ) if kv_cache_dtype == 'auto': gathered = src_cache_2d[linear_indices.long()] dst[valid_mask] = gathered[valid_mask].to(dst.dtype) else: gathered = src_cache_2d[linear_indices.long()].float() if kv_cache_dtype in ['fp8', 'fp8_e4m3', 'fp8_e5m2']: gathered = gathered * scale.item() dst[valid_mask] = gathered[valid_mask].to(dst.dtype) def _compute_prefill_context( self, q: torch.Tensor, kv_c_and_k_pe_cache: torch.Tensor, attn_metadata: MLACommonMetadata, k_scale: torch.Tensor, ): assert attn_metadata.prefill is not None prefill_metadata = attn_metadata.prefill assert prefill_metadata.chunked_context is not None output = None iters = len(prefill_metadata.chunked_context.seq_tot) workspace = prefill_metadata.chunked_context.workspace for i in range(iters): toks = prefill_metadata.chunked_context.seq_tot[i] torch.ops.xspeedgate_ops.gather_and_maybe_dequant_cache( src_cache=kv_c_and_k_pe_cache, dst=workspace, block_table=prefill_metadata.block_table, cu_seq_lens=prefill_metadata.chunked_context.cu_seq_lens[i], batch_size=attn_metadata.num_prefills, kv_cache_dtype=self.kv_cache_dtype, scale=k_scale, seq_starts=prefill_metadata.chunked_context.starts[i], ) kv_c_normed = workspace[:toks]\ [..., :self.kv_lora_rank] k_pe = workspace[:toks]\ [..., self.kv_lora_rank:].unsqueeze(1) kv_nope = self.kv_b_proj(kv_c_normed)[0].view( \ -1, self.num_heads, self.qk_nope_head_dim + self.v_head_dim) k_nope, v = kv_nope\ .split([self.qk_nope_head_dim, self.v_head_dim], dim=-1) k = torch.cat((k_nope, k_pe.expand((*k_nope.shape[:-1], -1))), dim=-1) attn_output, attn_softmax_lse = self._run_prefill_context_chunk( prefill=prefill_metadata, chunk_idx=i, q=q, k=k, v=v, ) if output is None: output = attn_output output_lse = attn_softmax_lse else: output_tmp = torch.empty_like(output) output_lse_tmp = torch.empty_like(output_lse) merge_attn_states( output=output_tmp, output_lse=output_lse_tmp, prefix_output=output, prefix_lse=output_lse, suffix_output=attn_output, suffix_lse=attn_softmax_lse, ) output = output_tmp output_lse = output_lse_tmp return output, output_lse def _context_parallel_compute_prefill_context( self, q: torch.Tensor, kv_c_and_k_pe_cache: torch.Tensor, attn_metadata: MLACommonMetadata, k_scale: torch.Tensor, dcp_world_size: int, ): assert k_scale is None, "DCP not support scaled kvcache now." assert attn_metadata.prefill is not None prefill_metadata = attn_metadata.prefill assert prefill_metadata.chunked_context is not None assert prefill_metadata.chunked_context.cp_chunk_seq_lens is not None assert prefill_metadata.chunked_context.origin_context_lens is not None assert prefill_metadata.chunked_context.cp_cu_seq_lens is not None assert prefill_metadata.chunked_context.chunk_size is not None assert prefill_metadata.chunked_context.cu_seq_lens_lst is not None output = None iters = len(prefill_metadata.chunked_context.seq_tot) workspace = prefill_metadata.chunked_context.workspace for i in range(iters): toks = prefill_metadata.chunked_context.seq_tot[i] ops.cp_gather_cache( src_cache=kv_c_and_k_pe_cache, dst=workspace, block_table=prefill_metadata.block_table, cu_seq_lens=prefill_metadata.chunked_context.cp_cu_seq_lens[i], batch_size=attn_metadata.num_prefills, seq_starts=prefill_metadata.chunked_context.starts[i], ) # workspace # |------- N tokens --------|--------- N*dcp_size tokens ----------| # |<- use for loca_gather ->|<--------- use for allgather -------->| allgather_offset = workspace.shape[0] // (dcp_world_size + 1) assert allgather_offset * (dcp_world_size + 1) == workspace.shape[0] assert toks <= allgather_offset local_gathered_kvcache = workspace[:toks] cur_allgather_workspace = workspace[ allgather_offset:allgather_offset * (1 + dcp_world_size)] assert toks * dcp_world_size <= cur_allgather_workspace.shape[0] cur_allgather_kvcache = cur_allgather_workspace[:toks * dcp_world_size] cur_allgather_kvcache.copy_(get_dcp_group().all_gather( local_gathered_kvcache, dim=0)) assert cur_allgather_kvcache.shape[ -1] == self.kv_lora_rank + self.qk_rope_head_dim allgatered_kv_c_normed, allgatered_k_pe = \ cur_allgather_kvcache.unsqueeze( 1).split([self.kv_lora_rank, self.qk_rope_head_dim], dim=-1) kv_c_normed, k_pe = reorg_kvcache( allgatered_kv_c_normed, allgatered_k_pe, cp_chunk_seq_lens_lst=prefill_metadata.chunked_context. cp_chunk_seq_lens[i], origin_context_lens=prefill_metadata.chunked_context. origin_context_lens, cp_world_size=dcp_world_size, sum_seq_len=prefill_metadata.chunked_context.cu_seq_lens_lst[i] [-1], max_seq_len=prefill_metadata.chunked_context.max_seq_lens[i], chunk_size=prefill_metadata.chunked_context.chunk_size, chunk_idx=i, toks=toks) kv_nope = self.kv_b_proj(kv_c_normed)[0].view( \ -1, self.num_heads, self.qk_nope_head_dim + self.v_head_dim) k_nope, v = kv_nope\ .split([self.qk_nope_head_dim, self.v_head_dim], dim=-1) k = torch.cat((k_nope, k_pe.expand((*k_nope.shape[:-1], -1))), dim=-1) attn_output, attn_softmax_lse = self._run_prefill_context_chunk( prefill=prefill_metadata, chunk_idx=i, q=q, k=k, v=v, ) if output is None: output = attn_output output_lse = attn_softmax_lse else: output_tmp = torch.empty_like(output) output_lse_tmp = torch.empty_like(output_lse) merge_attn_states( output=output_tmp, output_lse=output_lse_tmp, prefix_output=output, prefix_lse=output_lse, suffix_output=attn_output, suffix_lse=attn_softmax_lse, ) output = output_tmp output_lse = output_lse_tmp return output, output_lse def _forward_prefill( self, q: torch.Tensor, kv_c_normed: torch.Tensor, k_pe: torch.Tensor, kv_c_and_k_pe_cache: torch.Tensor, attn_metadata: MLACommonMetadata, k_scale: torch.Tensor, ) -> torch.Tensor: # TODO (zyongye): Prefill function here assert attn_metadata.prefill is not None assert self.dcp_world_size is not None has_context = attn_metadata.prefill.chunked_context is not None kv_nope = self.kv_b_proj(kv_c_normed)[0].view(\ -1, self.num_heads, self.qk_nope_head_dim + self.v_head_dim) k_nope, v = kv_nope\ .split([self.qk_nope_head_dim, self.v_head_dim], dim=-1) k = torch.cat((k_nope, k_pe.expand((*k_nope.shape[:-1], -1))), dim=-1) output = self._run_prefill_new_tokens( prefill=attn_metadata.prefill, q=q, k=k, v=v, return_softmax_lse=has_context, ) if has_context: suffix_output, suffix_lse = output if self.dcp_world_size > 1: context_output, context_lse = \ self._context_parallel_compute_prefill_context( q, kv_c_and_k_pe_cache, attn_metadata, k_scale=None, dcp_world_size=self.dcp_world_size) else: context_output, context_lse = \ self._compute_prefill_context( q, kv_c_and_k_pe_cache, attn_metadata, k_scale) output = torch.empty_like(suffix_output) output_lse = torch.empty_like(output) merge_attn_states( output=output, prefix_output=context_output, prefix_lse=context_lse, suffix_output=suffix_output, suffix_lse=suffix_lse, output_lse=output_lse, ) # unpad if necessary if self._pad_v: output = output[..., :v.shape[-1]] return output.flatten(start_dim=-2) @abstractmethod def _forward_decode( self, q: Union[torch.Tensor, tuple[torch.Tensor, torch.Tensor]], kv_c_and_k_pe_cache: torch.Tensor, attn_metadata: M, layer: AttentionLayer, ) -> tuple[torch.Tensor, Optional[torch.Tensor]]: raise NotImplementedError def forward( self, layer: AttentionLayer, q: torch.Tensor, k_c_normed: torch.Tensor, # key in unified attn k_pe: torch.Tensor, # value in unified attn kv_cache: torch.Tensor, attn_metadata: M, output: Optional[torch.Tensor] = None, output_scale: Optional[torch.Tensor] = None, output_block_scale: Optional[torch.Tensor] = None, ) -> torch.Tensor: assert output is not None, "Output tensor must be provided." if output_scale is not None or output_block_scale is not None: raise NotImplementedError( "fused output quantization is not yet supported" " for MLACommonImpl") if attn_metadata is None: # During the profile run try to simulate to worse case output size # for `self.kv_b_proj(kv_c_normed)` in `_compute_prefill_context` # since this can be large _ = torch.empty( (self.chunked_prefill_workspace_size, self.num_heads, self.qk_nope_head_dim + self.v_head_dim), device=k_c_normed.device, dtype=k_c_normed.dtype, ) # The zero fill is required when used with DP + EP # to ensure all ranks within a DP group compute the # same expert outputs. return output.fill_(0) if self.dcp_world_size is None: self.dcp_world_size = get_dcp_group().world_size fp8_attention = self.kv_cache_dtype.startswith("fp8") num_actual_toks = attn_metadata.num_actual_tokens # Inputs and outputs may be padded for CUDA graphs output_padded = output output = output[:num_actual_toks, ...] q = q[:num_actual_toks, ...] k_c_normed = k_c_normed[:num_actual_toks, ...] k_pe = k_pe[:num_actual_toks, ...] assert attn_metadata.num_decodes is not None and \ attn_metadata.num_prefills is not None and \ attn_metadata.num_decode_tokens is not None has_decode = attn_metadata.num_decodes > 0 has_prefill = attn_metadata.num_prefills > 0 num_decode_tokens = attn_metadata.num_decode_tokens decode_q = q[:num_decode_tokens] prefill_q = q[num_decode_tokens:] prefill_k_pe = k_pe[num_decode_tokens:] prefill_k_c_normed = k_c_normed[num_decode_tokens:] # write the latent and rope to kv cache if kv_cache.numel() > 0: kunlun_ops.concat_and_cache_mla( k_c_normed, k_pe.squeeze(1), attn_metadata.slot_mapping.flatten(), kv_cache, ) if fp8_attention: kv_cache = kv_cache.view(current_platform.fp8_dtype()) if has_prefill: output[num_decode_tokens:] = self._forward_prefill( prefill_q, prefill_k_c_normed, prefill_k_pe, kv_cache, attn_metadata, layer._k_scale) if has_decode: assert attn_metadata.decode is not None decode_q_nope, decode_q_pe = decode_q.split( [self.qk_nope_head_dim, self.qk_rope_head_dim], dim=-1) # Pads the head_dim if necessary (for the underlying kernel) if self.q_pad_num_heads is not None: B, N, L = decode_q_pe.shape decode_pe_padded = decode_q_pe.new_empty( (B, self.q_pad_num_heads, L)) decode_pe_padded.resize_((B, N, L)) decode_pe_padded.copy_(decode_q_pe) decode_q_pe = decode_pe_padded if is_rocm_aiter_fp8bmm_enabled(): # Multiply+Transpose (N, B, P)x(N, P, L)->(N, B, L)->(B, N, L) decode_ql_nope = aiter_triton_fp8_bmm(decode_q_nope, self.W_K, self.W_K_scale, group_size=128, transpose_bm=True) else: if vllm_kunlun_envs.VLLM_KUNLUN_ENABLE_INT8_BMM: q_len = decode_q_nope.shape[0] decode_ql_nope = decode_q_nope.new_empty( q_len, self.num_heads, self.kv_lora_rank, dtype=torch.float16, ) sorted_tokens_num_lod = torch.arange( self.num_heads + 1, dtype=torch.int, device="cuda" ) * q_len sorted_tokens_idx = torch.arange( self.num_heads * q_len, dtype=torch.int, device="cuda") extra_params = {"trans": False} kunlun_ops.mla_bmm_I8( decode_q_nope.contiguous(), self.W_UK_T, self.W_UK_SCALE, decode_ql_nope, sorted_tokens_num_lod, sorted_tokens_idx, **extra_params ) else: # Convert from (B, N, P) to (N, B, P) decode_q_nope = decode_q_nope.transpose(0, 1) N, B, P = decode_q_nope.shape _, _, L = self.W_UK_T.shape if self.q_pad_num_heads is not None: decode_ql_nope = decode_q_nope.new_empty( (self.q_pad_num_heads, B, L)) decode_ql_nope.resize_((N, B, L)) else: decode_ql_nope = decode_q_nope.new_empty((N, B, L)) # Multiply (N, B, P) x (N, P, L) -> (N, B, L) torch.bmm(decode_q_nope, self.W_UK_T, out=decode_ql_nope) # Convert from (N, B, L) to (B, N, L) decode_ql_nope = decode_ql_nope.transpose(0, 1) if fp8_attention: ql_nope_shape = decode_ql_nope.shape decode_ql_nope, _ = ops.scaled_fp8_quant( decode_ql_nope.reshape([ ql_nope_shape[0], ql_nope_shape[1] * ql_nope_shape[2] ]), layer._q_scale) decode_ql_nope = decode_ql_nope.reshape(ql_nope_shape) q_pe_shape = decode_q_pe.shape decode_q_pe, _ = ops.scaled_fp8_quant( decode_q_pe.reshape( [q_pe_shape[0], q_pe_shape[1] * q_pe_shape[2]]), layer._q_scale) decode_q_pe = decode_q_pe.reshape(q_pe_shape) decode_q = (decode_ql_nope, decode_q_pe) if self.dcp_world_size > 1: assert not fp8_attention, "DCP not support fp8 kvcache now." # concatenate decode_ql_nope and decode_q_pe -> (B, N, L + P) decode_q = torch.cat(decode_q, dim=-1) # decode_q do allgather in head dim. decode_q = get_dcp_group().all_gather(decode_q, dim=1) # call decode attn attn_out, lse = self._forward_decode(decode_q, kv_cache, attn_metadata, layer) # recorect dcp attn_out with lse. if self.dcp_world_size > 1: attn_out = cp_lse_ag_out_rs(attn_out, lse, get_dcp_group()) # v_up projection self._v_up_proj(attn_out, out=output[:num_decode_tokens]) return output_padded