462 lines
20 KiB
Python
462 lines
20 KiB
Python
import numpy as np
|
|
import torch
|
|
from vllm.distributed import get_dcp_group, get_pcp_group
|
|
from vllm.utils.math_utils import cdiv
|
|
from vllm.v1.attention.backends.utils import PAD_SLOT_ID
|
|
from vllm.v1.kv_cache_interface import KVCacheGroupSpec, MambaSpec, UniformTypeKVCacheSpecs
|
|
from vllm.v1.utils import CpuGpuBuffer
|
|
from vllm.v1.worker.block_table import _compute_slot_mapping_kernel
|
|
from vllm.v1.worker.cp_utils import get_total_cp_world_size
|
|
|
|
|
|
class BlockTable:
|
|
def __init__(
|
|
self,
|
|
block_size: int,
|
|
max_num_reqs: int,
|
|
max_num_blocks_per_req: int,
|
|
max_num_batched_tokens: int,
|
|
pin_memory: bool,
|
|
device: torch.device,
|
|
kernel_sizes: list[int] | None = None,
|
|
cp_kv_cache_interleave_size: int = 1,
|
|
num_speculative_tokens: int = 0,
|
|
kv_cache_group: KVCacheGroupSpec = None,
|
|
):
|
|
self.max_num_reqs = max_num_reqs
|
|
self.pcp_world_size = get_pcp_group().world_size
|
|
self.pcp_rank = get_pcp_group().rank_in_group if self.pcp_world_size > 1 else 0
|
|
self.dcp_world_size = get_dcp_group().world_size
|
|
self.dcp_rank = get_dcp_group().rank_in_group
|
|
compress_ratio = 1
|
|
if (
|
|
kv_cache_group is not None
|
|
and hasattr(kv_cache_group, "kv_cache_spec")
|
|
and isinstance(kv_cache_group.kv_cache_spec, UniformTypeKVCacheSpecs)
|
|
):
|
|
kv_cache_spec = next(iter(kv_cache_group.kv_cache_spec.kv_cache_specs.values()), None)
|
|
if kv_cache_spec is not None and hasattr(kv_cache_spec, "compress_ratio"):
|
|
compress_ratio = kv_cache_spec.compress_ratio
|
|
if (
|
|
kv_cache_group is not None
|
|
and hasattr(kv_cache_group, "kv_cache_spec")
|
|
and (self.pcp_world_size * self.dcp_world_size > 1)
|
|
and isinstance(kv_cache_group.kv_cache_spec, MambaSpec)
|
|
):
|
|
max_num_blocks_per_req = max_num_blocks_per_req * self.pcp_world_size * self.dcp_world_size
|
|
max_num_blocks_per_req = max(cdiv(max_num_blocks_per_req, compress_ratio), 1)
|
|
self.max_num_blocks_per_req = max_num_blocks_per_req
|
|
self.max_num_batched_tokens = max_num_batched_tokens
|
|
self.pin_memory = pin_memory
|
|
self.device = device
|
|
self.physical_block_size = block_size
|
|
self.is_mamba_group = (
|
|
kv_cache_group is not None
|
|
and hasattr(kv_cache_group, "kv_cache_spec")
|
|
and isinstance(kv_cache_group.kv_cache_spec, MambaSpec)
|
|
)
|
|
|
|
# If kernel_sizes is None or [0], use physical block size (no splitting)
|
|
if kernel_sizes is None or kernel_sizes == [0]:
|
|
self.block_size = block_size
|
|
self.logical_block_size = block_size
|
|
self.blocks_per_phys_block = 1
|
|
self.use_hybrid_blocks = False
|
|
else:
|
|
# Find the first kernel size that divides physical_block_size evenly
|
|
selected_kernel_size = None
|
|
for kernel_size in kernel_sizes:
|
|
if kernel_size > 0 and self.physical_block_size % kernel_size == 0:
|
|
selected_kernel_size = kernel_size
|
|
break
|
|
|
|
if selected_kernel_size is None:
|
|
raise ValueError(
|
|
f"None of the kernel sizes {kernel_sizes} can divide "
|
|
f"physical block size {self.physical_block_size} evenly"
|
|
)
|
|
|
|
self.block_size = selected_kernel_size
|
|
self.logical_block_size = selected_kernel_size
|
|
self.blocks_per_phys_block = self.physical_block_size // self.logical_block_size
|
|
if self.blocks_per_phys_block > 1:
|
|
self.use_hybrid_blocks = True
|
|
else:
|
|
self.use_hybrid_blocks = False
|
|
|
|
if self.use_hybrid_blocks:
|
|
logical_table_size = max_num_blocks_per_req * self.blocks_per_phys_block
|
|
else:
|
|
logical_table_size = max_num_blocks_per_req
|
|
|
|
duplicate_size = 1
|
|
if self.pcp_world_size * self.dcp_world_size > 1:
|
|
duplicate_size += num_speculative_tokens
|
|
self.block_table = self._make_buffer(max_num_reqs * duplicate_size, logical_table_size, dtype=torch.int32)
|
|
self.num_blocks_per_row = np.zeros(max_num_reqs, dtype=np.int32)
|
|
# MTP slot preparation appends up to num_speculative_tokens - 1
|
|
# draft positions for every request in addition to graph padding.
|
|
num_mtp_draft_slots = max(num_speculative_tokens - 1, 0) * self.max_num_reqs
|
|
self.slot_mapping = self._make_buffer(
|
|
self.max_num_batched_tokens + 2 * self.pcp_world_size * self.max_num_reqs + num_mtp_draft_slots,
|
|
dtype=torch.int32,
|
|
)
|
|
|
|
self.kernel_sizes = kernel_sizes
|
|
self.cp_kv_cache_interleave_size = cp_kv_cache_interleave_size
|
|
|
|
def append_row(
|
|
self,
|
|
block_ids,
|
|
row_idx: int,
|
|
) -> None:
|
|
if not block_ids:
|
|
return
|
|
block_ids = np.array(block_ids)
|
|
if self.use_hybrid_blocks:
|
|
block_ids = self._convert_physical_to_logical_blocks(block_ids)
|
|
|
|
num_blocks = len(block_ids)
|
|
start = self.num_blocks_per_row[row_idx]
|
|
|
|
self.block_table.np[row_idx, start : start + num_blocks] = block_ids
|
|
self.num_blocks_per_row[row_idx] += num_blocks
|
|
|
|
def add_row(self, block_ids: list[int], row_idx: int) -> None:
|
|
self.num_blocks_per_row[row_idx] = 0
|
|
self.append_row(block_ids, row_idx)
|
|
|
|
def clear_row(self, row_idx: int) -> None:
|
|
num_blocks = self.num_blocks_per_row[row_idx]
|
|
if num_blocks > 0:
|
|
self.block_table.np[row_idx, :num_blocks] = 0
|
|
self.num_blocks_per_row[row_idx] = 0
|
|
|
|
def move_row(self, src: int, tgt: int) -> None:
|
|
num_blocks = self.num_blocks_per_row[src]
|
|
self.block_table.np[tgt, :num_blocks] = self.block_table.np[src, :num_blocks]
|
|
self.num_blocks_per_row[tgt] = num_blocks
|
|
|
|
def swap_row(self, src: int, tgt: int) -> None:
|
|
num_blocks_src = self.num_blocks_per_row[src]
|
|
num_blocks_tgt = self.num_blocks_per_row[tgt]
|
|
self.num_blocks_per_row[src] = num_blocks_tgt
|
|
self.num_blocks_per_row[tgt] = num_blocks_src
|
|
|
|
self.block_table.np[[src, tgt]] = self.block_table.np[[tgt, src]]
|
|
|
|
def compute_slot_mapping(
|
|
self,
|
|
num_reqs: int,
|
|
query_start_loc: torch.Tensor,
|
|
positions: torch.Tensor,
|
|
) -> None:
|
|
num_tokens = positions.shape[0]
|
|
total_cp_world_size = self.pcp_world_size * self.dcp_world_size
|
|
total_cp_rank = self.pcp_rank * self.dcp_world_size + self.dcp_rank
|
|
if self.dcp_world_size * self.pcp_world_size > 1:
|
|
req_indices = torch.repeat_interleave(
|
|
torch.arange(num_reqs, dtype=torch.int32, device=query_start_loc.device),
|
|
query_start_loc[1:] - query_start_loc[:-1],
|
|
output_size=num_tokens,
|
|
)
|
|
self._compute_pcp_dcp_slot_mapping(req_indices, positions)
|
|
else:
|
|
_compute_slot_mapping_kernel[(num_reqs + 1,)](
|
|
num_tokens,
|
|
self.max_num_batched_tokens,
|
|
query_start_loc,
|
|
positions,
|
|
self.block_table.gpu,
|
|
self.block_table.gpu.stride(0),
|
|
self.block_size,
|
|
self.slot_mapping.gpu,
|
|
TOTAL_CP_WORLD_SIZE=total_cp_world_size,
|
|
TOTAL_CP_RANK=total_cp_rank,
|
|
CP_KV_CACHE_INTERLEAVE_SIZE=self.cp_kv_cache_interleave_size,
|
|
PAD_ID=PAD_SLOT_ID,
|
|
BLOCK_SIZE=1024,
|
|
)
|
|
|
|
def compute_slot_mapping_draft(
|
|
self,
|
|
req_indices: np.ndarray | torch.Tensor,
|
|
positions: np.ndarray | torch.Tensor,
|
|
) -> None:
|
|
# E.g., [0, 1, 0, 1, 2, 3, 4, 0, 1, 2]
|
|
# -> [0, 0, K, K, K + 1, K + 1, K + 2, 2 * K, 2 * K, 2 * K + 1]
|
|
# where K is the max_num_blocks_per_req and the block size is 2.
|
|
# NOTE(woosuk): We can't simply use `token_indices // block_size`
|
|
# here because M (max_model_len) is not necessarily divisible by
|
|
# block_size.
|
|
|
|
if self.dcp_world_size * self.pcp_world_size > 1:
|
|
if not isinstance(req_indices, torch.Tensor):
|
|
req_indices = torch.from_numpy(req_indices)
|
|
if not isinstance(positions, torch.Tensor):
|
|
positions = torch.from_numpy(positions)
|
|
self._compute_pcp_dcp_slot_mapping(req_indices, positions)
|
|
else:
|
|
if isinstance(req_indices, torch.Tensor):
|
|
if req_indices.device.type != "cpu":
|
|
raise ValueError("Device tensor inputs are only supported for CP draft slot mapping.")
|
|
req_indices = req_indices.numpy()
|
|
if isinstance(positions, torch.Tensor):
|
|
if positions.device.type != "cpu":
|
|
raise ValueError("Device tensor inputs are only supported for CP draft slot mapping.")
|
|
positions = positions.numpy()
|
|
assert self.kernel_sizes is not None
|
|
assert self.block_size == self.kernel_sizes[0]
|
|
# IMPORTANT: In hybrid mode, positions are in logical block space,
|
|
# but we need to map them to the correct logical block table indices
|
|
logical_block_idx = positions // self.block_size
|
|
|
|
# Account for the expanded logical table
|
|
# (always needed with unified tensor)
|
|
# Each physical block is split into multiple logical blocks
|
|
# The logical table has been expanded to accommodate this
|
|
block_table_indices = (
|
|
req_indices * self.max_num_blocks_per_req * self.blocks_per_phys_block + logical_block_idx
|
|
)
|
|
|
|
block_offsets = positions % self.block_size
|
|
block_numbers = self.block_table.np.ravel()[block_table_indices]
|
|
np.add(
|
|
block_numbers * self.block_size,
|
|
block_offsets,
|
|
out=self.slot_mapping.np[: req_indices.shape[0]],
|
|
)
|
|
self.slot_mapping.copy_to_gpu(req_indices.shape[0])
|
|
|
|
def _compute_pcp_dcp_slot_mapping(
|
|
self,
|
|
req_indices: torch.Tensor,
|
|
positions: torch.Tensor,
|
|
) -> None:
|
|
# Note(hc): The DCP implement store kvcache with an interleave
|
|
# style, the kvcache for the token whose token_idx is i is
|
|
# always stored on the GPU whose dcp_rank equals i % pcp_world_size:
|
|
|
|
# Use a "virtual block" which equals to world_size * block_size
|
|
# for block_table_indices calculation.
|
|
# virtual_block_size = self.block_size * self.dcp_world_size * self.pcp_world_size
|
|
|
|
# IMPORTANT: In hybrid mode, positions are in logical block space,
|
|
# but we need to map them to the correct logical block table indices
|
|
# logical_block_idx = positions // virtual_block_size
|
|
|
|
total_cp_world_size = self.dcp_world_size * self.pcp_world_size
|
|
virtual_physical_block_size = self.physical_block_size * total_cp_world_size
|
|
physical_block_idx = positions // virtual_physical_block_size
|
|
virtual_block_offsets = positions % virtual_physical_block_size
|
|
|
|
self.current_rank = self.dcp_world_size * self.pcp_rank + self.dcp_rank
|
|
mask = virtual_block_offsets // self.cp_kv_cache_interleave_size % total_cp_world_size == self.current_rank
|
|
local_physical_offsets = (
|
|
virtual_block_offsets
|
|
// (total_cp_world_size * self.cp_kv_cache_interleave_size)
|
|
* self.cp_kv_cache_interleave_size
|
|
+ virtual_block_offsets % self.cp_kv_cache_interleave_size
|
|
)
|
|
logical_block_idx = physical_block_idx * self.blocks_per_phys_block + (
|
|
local_physical_offsets // self.block_size
|
|
)
|
|
|
|
block_table_indices = req_indices * self.max_num_blocks_per_req * self.blocks_per_phys_block + logical_block_idx
|
|
|
|
block_offsets = local_physical_offsets % self.block_size
|
|
|
|
if block_table_indices.device.type != "cpu":
|
|
block_numbers = self.block_table.gpu.flatten()[block_table_indices]
|
|
slot_mapping = block_numbers * self.block_size + block_offsets
|
|
self.slot_mapping.gpu[: req_indices.shape[0]] = torch.where(mask, slot_mapping, -1)
|
|
else:
|
|
block_numbers = self.block_table.cpu.flatten()[block_table_indices]
|
|
slot_mapping = block_numbers * self.block_size + block_offsets
|
|
self.slot_mapping.cpu[: req_indices.shape[0]] = torch.where(mask, slot_mapping, -1)
|
|
|
|
def commit_block_table(self, num_reqs: int) -> None:
|
|
self.block_table.gpu[:num_reqs].copy_(
|
|
self.block_table.cpu[:num_reqs].clone().pin_memory(),
|
|
non_blocking=True,
|
|
)
|
|
|
|
def clear(self) -> None:
|
|
self.block_table.fill_(0)
|
|
self.block_table.cpu.fill_(0)
|
|
|
|
def _convert_physical_to_logical_blocks(self, physical_blocks: np.ndarray) -> np.ndarray:
|
|
"""Convert physical block IDs to logical block IDs."""
|
|
if not self.use_hybrid_blocks:
|
|
return physical_blocks
|
|
|
|
# Create logical block IDs by splitting each physical block
|
|
logical_blocks: list[int] = []
|
|
for phys_block in physical_blocks:
|
|
# Convert physical block to multiple logical blocks
|
|
# Physical block 1 becomes logical blocks
|
|
# [1*split_ratio, 1*split_ratio+1, ...]
|
|
# But we need to account for the fact that block 0 is special
|
|
base_logical = phys_block * self.blocks_per_phys_block
|
|
logical_blocks.extend(range(base_logical, base_logical + self.blocks_per_phys_block))
|
|
|
|
return np.array(logical_blocks, dtype=np.int32)
|
|
|
|
def get_device_tensor(self, num_reqs: int | None = None) -> torch.Tensor:
|
|
"""Returns the device tensor of the block table."""
|
|
if num_reqs is not None:
|
|
return self.block_table.gpu[:num_reqs]
|
|
return self.block_table.gpu
|
|
|
|
def get_cpu_tensor(self) -> torch.Tensor:
|
|
"""Returns the CPU tensor of the block table."""
|
|
return self.block_table.cpu
|
|
|
|
def get_numpy_array(self) -> np.ndarray:
|
|
"""Returns the numpy array of the block table."""
|
|
return self.block_table.np
|
|
|
|
def _make_buffer(self, *size: int | torch.SymInt, dtype: torch.dtype) -> CpuGpuBuffer:
|
|
return CpuGpuBuffer(*size, dtype=dtype, device=self.device, pin_memory=self.pin_memory)
|
|
|
|
|
|
class MultiGroupBlockTable:
|
|
"""The BlockTables for each KV cache group."""
|
|
|
|
def __init__(
|
|
self,
|
|
max_num_reqs: int,
|
|
max_model_len: int,
|
|
max_num_batched_tokens: int,
|
|
pin_memory: bool,
|
|
device: torch.device,
|
|
block_sizes: list[int],
|
|
num_speculative_tokens: int = 0,
|
|
max_num_blocks: list[int] | None = None,
|
|
kernel_sizes: list[list[int]] | None = None,
|
|
cp_kv_cache_interleave_size: int = 1,
|
|
kv_cache_groups: KVCacheGroupSpec = None,
|
|
) -> None:
|
|
if kernel_sizes is None:
|
|
kernel_sizes = [[0]] * len(block_sizes)
|
|
# Ensure kernel_sizes matches block_sizes length
|
|
elif len(kernel_sizes) == 1 and len(block_sizes) > 1:
|
|
kernel_sizes = kernel_sizes * len(block_sizes)
|
|
elif len(kernel_sizes) != len(block_sizes):
|
|
raise ValueError(
|
|
f"kernel_sizes length ({len(kernel_sizes)}) must match block_sizes length ({len(block_sizes)})"
|
|
)
|
|
|
|
if max_num_blocks is None:
|
|
# Note(hc): each dcp rank only store
|
|
# (max_model_len//dcp_world_size) tokens in kvcache,
|
|
# so the block_size which used for calc max_num_blocks_per_req
|
|
# must be multiplied by dcp_world_size.
|
|
total_cp_world_size = get_total_cp_world_size()
|
|
max_num_blocks = [cdiv(max_model_len, block_size * total_cp_world_size) for block_size in block_sizes]
|
|
|
|
if len(max_num_blocks) != len(block_sizes):
|
|
raise ValueError(
|
|
f"max_num_blocks length ({len(max_num_blocks)}) must match block_sizes length ({len(block_sizes)})"
|
|
)
|
|
|
|
# Use zip to pair block_sizes with kernel_sizes one-to-one
|
|
if kv_cache_groups is not None:
|
|
self.block_tables = [
|
|
BlockTable(
|
|
block_size,
|
|
max_num_reqs,
|
|
max_num_blocks_per_req,
|
|
max_num_batched_tokens,
|
|
pin_memory,
|
|
device,
|
|
kernel_size_list,
|
|
cp_kv_cache_interleave_size,
|
|
num_speculative_tokens,
|
|
kv_cache_group,
|
|
)
|
|
for block_size, kernel_size_list, max_num_blocks_per_req, kv_cache_group in zip(
|
|
block_sizes, kernel_sizes, max_num_blocks, kv_cache_groups
|
|
)
|
|
]
|
|
else:
|
|
self.block_tables = [
|
|
BlockTable(
|
|
block_size,
|
|
max_num_reqs,
|
|
max_num_blocks_per_req,
|
|
max_num_batched_tokens,
|
|
pin_memory,
|
|
device,
|
|
kernel_size_list,
|
|
cp_kv_cache_interleave_size,
|
|
num_speculative_tokens,
|
|
)
|
|
for block_size, kernel_size_list, max_num_blocks_per_req in zip(
|
|
block_sizes, kernel_sizes, max_num_blocks
|
|
)
|
|
]
|
|
|
|
def append_row(self, block_ids: tuple[list[int], ...], row_idx: int) -> None:
|
|
for i, block_table in enumerate(self.block_tables):
|
|
block_table.append_row(block_ids[i], row_idx)
|
|
|
|
def add_row(self, block_ids: tuple[list[int], ...], row_idx: int) -> None:
|
|
for i, block_table in enumerate(self.block_tables):
|
|
block_table.add_row(block_ids[i], row_idx)
|
|
|
|
def clear_row(self, row_idx: int) -> None:
|
|
for block_table in self.block_tables:
|
|
block_table.clear_row(row_idx)
|
|
|
|
def move_row(self, src: int, tgt: int) -> None:
|
|
for block_table in self.block_tables:
|
|
block_table.move_row(src, tgt)
|
|
|
|
def swap_row(self, src: int, tgt: int) -> None:
|
|
for block_table in self.block_tables:
|
|
block_table.swap_row(src, tgt)
|
|
|
|
def compute_slot_mapping(
|
|
self,
|
|
num_reqs: int,
|
|
query_start_loc: torch.Tensor,
|
|
positions: torch.Tensor,
|
|
positions_compressed_list: list[np.ndarray] | None = None,
|
|
req_indices_compressed_list: list[np.ndarray] | None = None,
|
|
) -> None:
|
|
for i, block_table in enumerate(self.block_tables):
|
|
if block_table.is_mamba_group:
|
|
continue
|
|
if positions_compressed_list and req_indices_compressed_list:
|
|
block_table.compute_slot_mapping_draft(req_indices_compressed_list[i], positions_compressed_list[i])
|
|
else:
|
|
block_table.compute_slot_mapping(num_reqs, query_start_loc, positions)
|
|
|
|
def compute_slot_mapping_draft(
|
|
self,
|
|
req_indices: np.ndarray | torch.Tensor,
|
|
positions: np.ndarray | torch.Tensor,
|
|
positions_compressed_list: list[np.ndarray] | None = None,
|
|
req_indices_compressed_list: list[np.ndarray] | None = None,
|
|
) -> None:
|
|
for i, block_table in enumerate(self.block_tables):
|
|
if block_table.is_mamba_group:
|
|
continue
|
|
if positions_compressed_list and req_indices_compressed_list:
|
|
block_table.compute_slot_mapping_draft(req_indices_compressed_list[i], positions_compressed_list[i])
|
|
else:
|
|
block_table.compute_slot_mapping_draft(req_indices, positions)
|
|
|
|
def commit_block_table(self, num_reqs: int) -> None:
|
|
for block_table in self.block_tables:
|
|
block_table.commit_block_table(num_reqs)
|
|
|
|
def clear(self) -> None:
|
|
for block_table in self.block_tables:
|
|
block_table.clear()
|
|
|
|
def __getitem__(self, idx: int) -> "BlockTable":
|
|
"""Returns the BlockTable for the i-th KV cache group."""
|
|
return self.block_tables[idx]
|