Skip to content

refactor: skip flash_attn build #388

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
guocuimi merged 1 commit into from
Jan 25, 2025
Merged

refactor: skip flash_attn build #388

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
2 changes: 0 additions & 2 deletions src/kernels/attention/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -79,6 +79,4 @@ cc_binary(
-lineinfo
)

add_subdirectory(flash_attn)
# add_subdirectory(flash_infer)
add_subdirectory(tools)
1 change: 0 additions & 1 deletion src/layers/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -58,7 +58,6 @@ cc_library(
:pos_embedding
:attention
:kernels
:flash_attn.kernels
glog::glog
gflags::gflags
torch
Expand Down
3 changes: 0 additions & 3 deletions src/layers/attention/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,21 +8,18 @@ cc_library(
HDRS
handler.h
ref_handler.h
flash_attn_handler.h
scale_attn_handler.h
attention.h
SRCS
handler.cpp
ref_handler.cpp
flash_attn_handler.cpp
scale_attn_handler.cpp
attention.cpp
DEPS
:state_dict
:memory
:pos_embedding
:kernels
:flash_attn.kernels
:attention.kernels
glog::glog
gflags::gflags
Expand Down
84 changes: 0 additions & 84 deletions src/layers/attention/flash_attn_handler.cpp
View file
Open in desktop

This file was deleted.

71 changes: 0 additions & 71 deletions src/layers/attention/flash_attn_handler.h
View file
Open in desktop

This file was deleted.

19 changes: 9 additions & 10 deletions src/layers/attention/handler.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,15 +7,14 @@
#include <boost/algorithm/string.hpp>
#include <memory>

#include "flash_attn_handler.h"
#include "scale_attn_handler.h"
#include "layers/pos_embedding.h"
#include "ref_handler.h"

// decide which attention implementation to use
DEFINE_string(attention_handler,
"auto",
"attention handler, e.g. auto, pytorch, flash_attn");
"attention handler, e.g. auto, pytorch, scale_attn");

namespace llm {

Expand Down Expand Up @@ -64,15 +63,15 @@ std::unique_ptr<AttentionHandler> AttentionHandler::create_handler_with_alibi(
}

const bool is_cuda = options.device().is_cuda();
if (boost::iequals(FLAGS_attention_handler, "flash_attn")) {
CHECK(is_cuda) << "flash_attn only supports cuda device";
return std::make_unique<FlashAttnHandler>(
if (boost::iequals(FLAGS_attention_handler, "scale_attn")) {
CHECK(is_cuda) << "scale_attn only supports cuda device";
return std::make_unique<ScaleAttnHandler>(
sm_scale, args.attn_logit_soft_cap(), alibi_slopes);
}

// choose the best handler based on device type
if (is_cuda) {
// use flash_attn for cuda device
// use scale_attn for cuda device
return std::make_unique<ScaleAttnHandler>(
sm_scale, args.attn_logit_soft_cap(), alibi_slopes);
}
Expand Down Expand Up @@ -111,9 +110,9 @@ std::unique_ptr<AttentionHandler> AttentionHandler::create_handler_with_rope(
}

const bool is_cuda = options.device().is_cuda();
if (boost::iequals(FLAGS_attention_handler, "flash_attn")) {
CHECK(is_cuda) << "flash_attn only supports cuda device";
return std::make_unique<FlashAttnHandler>(sm_scale,
if (boost::iequals(FLAGS_attention_handler, "scale_attn")) {
CHECK(is_cuda) << "scale_attn only supports cuda device";
return std::make_unique<ScaleAttnHandler>(sm_scale,
args.attn_logit_soft_cap(),
rotary_dim,
args.max_position_embeddings(),
Expand All @@ -124,7 +123,7 @@ std::unique_ptr<AttentionHandler> AttentionHandler::create_handler_with_rope(

// choose the best handler based on device type
if (is_cuda) {
// use flash_attn for cuda device
// use scale_attn for cuda device
return std::make_unique<ScaleAttnHandler>(sm_scale,
args.attn_logit_soft_cap(),
rotary_dim,
Expand Down
Loading