dylanyunlong/project_6
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

[fix] baseline docker build from ex_engine.python.xllm_ops import _get error

Browse Source
This commit is contained in:
root
2026-08-19 02:46:40 +00:00
parent 53daf03d57
commit 7d74ad4e8e
1 changed files with 13 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
qwen3_6_scripts/paged_attn.py
View File

@@ -2294,8 +2294,18 @@ class PagedAttention:
# Use xllm_cache.block_copy kernel — single launch for all layers. # Use xllm_cache.block_copy kernel — single launch for all layers.
if src_to_dists.numel() == 0: if src_to_dists.numel() == 0:
return return
from ex_engine.python.xllm_ops import _get as _xllm_get
_xllm_block_copy = _xllm_get("xllm_cache").block_copy import importlib.util as _ilu
import os as _os
# paged_attn.py lives at VLLM_ROOT/attention/ops/paged_attn.py
# xllm_cache.so lives at VLLM_ROOT/xllm_cache.so
_vllm_root = _os.path.dirname(_os.path.dirname(_os.path.dirname(
_os.path.abspath(__file__))))
_so = _os.path.join(_vllm_root, "xllm_cache.so")
_spec = _ilu.spec_from_file_location("xllm_cache", _so)
_mod = _ilu.module_from_spec(_spec)
_spec.loader.exec_module(_mod)
_xllm_block_copy = _mod.block_copy
device = src_to_dists.device device = src_to_dists.device
n = src_to_dists.size(0) n = src_to_dists.size(0)
@@ -2319,3 +2329,4 @@ class PagedAttention:
_xllm_block_copy(key_ptrs, val_ptrs, _xllm_block_copy(key_ptrs, val_ptrs,
src_indices, dst_indices, cum_sum, src_indices, dst_indices, cum_sum,
numel_per_block, cache_dtype) numel_per_block, cache_dtype)