[KV Cache] GLM-5.3-Flash: use the generic packed KV layout; CircularBufferSpec tail with speculative ring slots (#57169)
Signed-off-by: Yifan Qiao <yifanqiao@inferact.ai> Signed-off-by: Dakai An <dakaian108@gmail.com> Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com> Co-authored-by: Yifan Qiao <yifanqiao@inferact.ai> Co-authored-by: OpenAI Codex <noreply@openai.com> Co-authored-by: OpenAI Codex <codex@openai.com> Co-authored-by: Lucas Wilkinson <lwilkins@redhat.com> Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Co-authored-by: Lucas Wilkinson <LucasWilkinson@users.noreply.github.com>
This commit is contained in:
co-authored by
Yifan Qiao
OpenAI Codex
OpenAI Codex
Lucas Wilkinson
Claude Opus 5.5
Lucas Wilkinson
parent
3766a67a32
commit
02e55d2af3
@@ -2,7 +2,7 @@
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||
"""Utility functions for sparse MLA backends."""
|
||||
|
||||
from dataclasses import dataclass
|
||||
from dataclasses import dataclass, replace
|
||||
from typing import Any
|
||||
|
||||
import numpy as np
|
||||
@@ -22,6 +22,7 @@ from vllm.model_executor.warmup.jit_warmup_triton_helper import (
|
||||
)
|
||||
from vllm.triton_utils import tl, triton
|
||||
from vllm.utils.math_utils import cdiv
|
||||
from vllm.v1.kv_cache_interface import AttentionSpec
|
||||
from vllm.v1.worker.block_table import get_block_table_width
|
||||
|
||||
|
||||
@@ -40,6 +41,12 @@ def request_row_bounds(req_idx: np.ndarray) -> np.ndarray:
|
||||
return bounds
|
||||
|
||||
|
||||
def align_blocks_to_rows(spec: AttentionSpec, rows: int = 1) -> AttentionSpec:
|
||||
"""Sparse kernels read the cache as flat rows, viewed as ``rows``-row pages,
|
||||
so blocks must be whole pages apart."""
|
||||
return replace(spec, block_stride_alignment=rows * spec.state_content_size_bytes)
|
||||
|
||||
|
||||
def flat_kv_row_view(
|
||||
kv_cache: torch.Tensor, # [num_blocks, block_size, head_dim]
|
||||
block_size: int,
|
||||
|
||||
Reference in New Issue
Block a user