frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[CI Failure] Disable FlashInfer RoPE to unblock CI (#25299)

Browse Source 
Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Michael Goin
2025-09-20 04:16:56 -04:00
committed by GitHub
parent 3d9a1d2de5
commit e08a3a3fdb
1 changed files with 7 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
vllm/model_executor/layers/rotary_embedding/base.py
View File

@ -6,8 +6,6 @@ from typing import Optional
import torch
from vllm.model_executor.custom_op import CustomOp
from vllm.platforms import current_platform
from vllm.utils.flashinfer import has_flashinfer
from .common import apply_rotary_emb_torch
@ -32,13 +30,15 @@ class RotaryEmbedding(CustomOp):
self.base = base
self.is_neox_style = is_neox_style
self.dtype = dtype
# TODO(mgoin): disabled for now due to failures
# Flashinfer only supports head_size=64, 128, 256, 512.
# https://github.com/flashinfer-ai/flashinfer/blob/ebfd655efe830048dba5d582aaa61d61d1cf9a87/include/flashinfer/utils.cuh#L174-L202
self.use_flashinfer = (self.enabled()
and dtype in (torch.float16, torch.bfloat16)
and current_platform.is_cuda()
and has_flashinfer()
and self.head_size in [64, 128, 256, 512])
# self.use_flashinfer = (self.enabled()
# and dtype in (torch.float16, torch.bfloat16)
# and current_platform.is_cuda()
# and has_flashinfer()
# and self.head_size in [64, 128, 256, 512])
self.use_flashinfer = False
cache = self._compute_cos_sin_cache()
if not self.use_flashinfer: