Skip to content

[Bugfix] Massage MLA's usage of flash attn for RoCM #13310

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Feb 15, 2025
Merged

[Bugfix] Massage MLA's usage of flash attn for RoCM #13310

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
31bf997
Massage MLA's usage of flash attn for RoCM
tlrmchlsmth Feb 14, 2025
460783f
formatting
tlrmchlsmth Feb 14, 2025
645a911
Improve
tlrmchlsmth Feb 14, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 11 additions & 2 deletions vllm/attention/backends/mla/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
# SPDX-License-Identifier: Apache-2.0

import functools
from abc import abstractmethod
from dataclasses import dataclass
from typing import Any, Dict, Generic, List, Optional, Tuple
Expand Down Expand Up @@ -183,6 +184,15 @@ def __init__(
self.o_proj = o_proj
self.vllm_flash_attn_version = get_flash_attn_version()

# 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
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)

def _v_up_proj_and_o_proj(self, x):
if envs.VLLM_MLA_PERFORM_MATRIX_ABSORPTION:
if is_fp8(self.W_UV_O):
Expand Down Expand Up @@ -487,7 +497,7 @@ def _forward_prefill_flash(
v_padded = torch.nn.functional.pad(v, [0, q.shape[-1] - v.shape[-1]],
value=0)

attn_output = flash_attn_varlen_func(
attn_output = self.flash_attn_varlen_func(
q=q,
k=k,
v=v_padded,
Expand All @@ -497,7 +507,6 @@ def _forward_prefill_flash(
max_seqlen_k=max_prefill_seq_len,
softmax_scale=self.scale,
causal=True,
fa_version=self.vllm_flash_attn_version,
)
attn_output = attn_output\
.view(-1, self.num_heads, q.shape[-1])[..., :v.shape[-1]]\
Expand Down