llama.cpp
Feature: Integrate with unified SYCL backend for Intel GPUs
#2690

Merged

Feature: Integrate with unified SYCL backend for Intel GPUs #2690

ggerganov merged 92 commits into ggml-org:master from abhilash1910:sycl

abhilash1910 marked this pull request as draft 2 years ago

abhilash1910 marked this pull request as ready for review 2 years ago

AidanBeltonS commented on 2024-01-22

Green-Sky commented on 2024-01-22

AidanBeltonS commented on 2024-01-22

first update for migration

7a4343df

update init_cublas

23387693

add debug functio, commit all help code

0c00b4f6

step 1

ff837110

step 2

02dffb68

step3 add fp16, slower 31->28

43f2c358

add GGML_LIST_DEVICE function

da752eda

step 5 format device and print

6dd32789

step6, enhance error check, remove CUDA macro, enhance device id to f…

3a9d2c54

support main device is non-zero

65f895d4

step7 add debug for code path, rm log

3b1a743e

step 8, rename all macro & func from cuda by sycl

c2ef7a9c

fix error of select non-zero device, format device list

69d76c8b

ren ggml-sycl.hpp -> ggml-sycl.h

c709c3cb

clear CMAKE to rm unused lib and options

fa3a5860

correct queue: rm dtct:get_queue

3645f25d

add print tensor function to debug

bd38129a

fix error: wrong result in 658746bb26702e50f2c59c0e4ada8e9da6010481

5b538994

summary dpct definition in one header file to replace folder:dpct

a47f5ec4

refactor device log

c67c2ab2

mv dpct definition from folder dpct to ggml-sycl.h

c3c5b20a

update readme, refactor build script

ca2cb698

fix build with sycl

95daece9

set nthread=1 when sycl, increase performance

a8936f49

add run script, comment debug code

79d30d77

add ls-sycl-device tool

0d6e7219

add ls-sycl-device, rm unused files

7350fd48

rm rear space

09b5619d

dos2unix

d80dd65f

Update README_sycl.md

593ce001

fix return type

57e9fbad

remove sycl version from include path

d5f7d364

restore rm code to fix hang issue

35a0daaa

add syc and link for sycl readme

ae941b1b

rm original sycl code before refactor

e3481faa

fix code err

623d8031

add know issue for pvc hang issue

f396a3b6

enable SYCL_F16 support

f008cc7b

align pr4766

67e6b3cb

airMeng force pushed from 991e9cba to 67e6b3cb 1 year ago

check for sycl blas, better performance

533c647d

cleanup 1

dd7f1396

abhilash1910 changed the title ~~[Draft] Feature: Integrate with SYCL backend for Intel GPUs~~ Feature: Integrate with unified SYCL backend for Intel GPUs , Nvidia and AMD 1 year ago

remove extra endif

b4037842

add build&run script, clean CMakefile, update guide by review comments

a0a1304b

Merge branch 'sycl' of https://github.com/abhilash1910/llama.cpp into…

27c08c04

rename macro to intel hardware

97cbe18d

abhilash1910 changed the title ~~Feature: Integrate with unified SYCL backend for Intel GPUs , Nvidia and AMD~~ Feature: Integrate with unified SYCL backend for Intel GPUs 1 year ago

editor config format

1ddaf44c

format fixes

bd716b25

format fixes

be31379e

editor format fix

d097e2a4

Remove unused headers

88f64b7d

skip build sycl tool for other code path

756c4acc

replace tab by space

b42a32d3

fix blas matmul function

5f83a123

ggerganov commented on 2024-01-23

ggerganov added need feedback

fix mac build

d6fc1a03

restore hip dependency

c7e745e6

fix conflict

3bfb846d

ren as review comments

498121b1

mv internal function to .cpp file

91b14610

export funciton print_sycl_devices(), mv class dpct definition to sou…

816f480e

update CI/action for sycl code, fix CI error of repeat/dup

7a44a95b

fix action ID format issue

7babd769

rm unused strategy

04a46c46

enable llama_f16 in ci

799af056

airMeng force pushed 1 year ago

airMeng force pushed to 799af056 1 year ago

airMeng force pushed from 799af056 1 year ago

airMeng force pushed to 799af056 1 year ago

fix conflict

ec5c8bc0

fix build break on MacOS, due to CI of MacOS depend on external ggml,…

22e1b45c

koech-v approved these changes on 2024-01-24

Merge branch 'master' into sycl

238ec31a

slaren requested changes on 2024-01-24

fix ci cases for unsupported data type

67de350e

revert unrelated changed in cuda cmake

fb15de38

revert hip cmake changes

96186a74

fix indent

d07a88d4

add prefix in func name

8dd1b60a

revert no mmq

3aabd8a2

rm cpu blas duplicate

18742f7a

fix no_new_line

0e235fb8

fix src1->type==F16 bug.

56001182

pass batch offset for F16 src1

eef5faae

fix batch error

5bb93d41

fix wrong code

0635f844

abhilash1910 requested a review from

slaren 1 year ago

abhilash1910 requested a review from

ggerganov 1 year ago

slaren commented on 2024-01-25

revert sycl checking in test-sampling

f1bab501

pass void as arguments of ggml_backend_sycl_print_sycl_devices

66e24c24

slaren commented on 2024-01-25

remove extra blank line in test-sampling

b06dca67

revert setting n_threads in sycl

05b7f9be

slaren approved these changes on 2024-01-25

koech-v approved these changes on 2024-01-26

ggerganov added high priority

implement std::isinf for icpx with fast math.

d6a65050

ggerganov approved these changes on 2024-01-26

Update ci/run.sh

174c9a0e

Update examples/sycl/run-llama2.sh

c08fec2a

Update examples/sycl/run-llama2.sh

2cba564b

Update CMakeLists.txt

f7070510

Update CMakeLists.txt

45b06180

Update CMakeLists.txt

55317540

Update CMakeLists.txt

b9ffaab1

add copyright and MIT license declare

2ab97157

ngxson commented on 2024-01-27

update the cmd example

d394ca7f

ggerganov merged 0f648573 into master 1 year ago

Reviewers

ggerganov

slaren

koech-v

NeoZhangJianyu

ngxson

airMeng

AidanBeltonS

Green-Sky

Assignees

No one assigned

Labels

high priority need feedback

Milestone

No milestone

llama.cpp Feature: Integrate with unified SYCL backend for Intel GPUs #2690 Merged

Feature: Integrate with unified SYCL backend for Intel GPUs #2690

llama.cpp
Feature: Integrate with unified SYCL backend for Intel GPUs
#2690

Merged