[LLM] Test load_low_bit in iGPU perf test on Windows (#10313)
This commit is contained in:
Yuwen Hu
GitHub
parent
fd81d66047
commit
d45e577d8c
4 changed files with 27 additions and 28 deletions
43
.github/workflows/llm_performance_tests.yml
vendored
43
.github/workflows/llm_performance_tests.yml
vendored
|
|
@ -679,16 +679,15 @@ jobs:
|
|||
|
||||
call conda deactivate
|
||||
|
||||
# 32-512
|
||||
- name: Prepare igpu perf test (32-512)
|
||||
# load_low_bit 1024-128
|
||||
- name: Prepare igpu perf test (load_low_bit 1024-128)
|
||||
shell: bash
|
||||
run: |
|
||||
# hide time info
|
||||
sed -i 's/2048-256/32-512/g' python/llm/dev/benchmark/all-in-one/run.py
|
||||
sed -i 's/2048-256/1024-128/g' python/llm/dev/benchmark/all-in-one/run.py
|
||||
sed -i 's/{today}_test3/{today}_test1/g' python/llm/dev/benchmark/all-in-one/run.py
|
||||
sed -i "s/path to your local model hub/$MODEL_HUB_DIR/g" python/llm/test/benchmark/igpu-perf/32-512.yaml
|
||||
sed -i "s/path to your local model hub/$MODEL_HUB_DIR/g" python/llm/test/benchmark/igpu-perf/1024-128_loadlowbit.yaml
|
||||
|
||||
- name: Test on igpu (32-512)
|
||||
- name: Test on igpu (load_low_bit 1024-128)
|
||||
shell: cmd
|
||||
run: |
|
||||
call conda activate igpu-perf
|
||||
|
|
@ -701,20 +700,20 @@ jobs:
|
|||
set TRANSFORMERS_OFFLINE=1
|
||||
|
||||
cd python\llm\dev\benchmark\all-in-one
|
||||
move ..\..\..\test\benchmark\igpu-perf\32-512.yaml config.yaml
|
||||
move ..\..\..\test\benchmark\igpu-perf\1024-128_loadlowbit.yaml config.yaml
|
||||
set PYTHONIOENCODING=utf-8
|
||||
python run.py >> %CSV_SAVE_PATH%\32-512\log\%LOG_FILE% 2>&1
|
||||
python run.py >> %CSV_SAVE_PATH%\1024-128_loadlowbit\log\%LOG_FILE% 2>&1
|
||||
if %ERRORLEVEL% neq 0 (exit /b 1)
|
||||
|
||||
call conda deactivate
|
||||
|
||||
- name: Prepare igpu perf test for Mistral (32-512)
|
||||
- name: Prepare igpu perf test for Mistral (load_low_bit 1024-128)
|
||||
shell: bash
|
||||
run: |
|
||||
sed -i 's/{today}_test1/{today}_test2/g' python/llm/dev/benchmark/all-in-one/run.py
|
||||
sed -i "s/path to your local model hub/$MODEL_HUB_DIR/g" python/llm/test/benchmark/igpu-perf/32-512_434.yaml
|
||||
sed -i "s/path to your local model hub/$MODEL_HUB_DIR/g" python/llm/test/benchmark/igpu-perf/1024-128_loadlowbit_434.yaml
|
||||
|
||||
- name: Test on igpu for Mistral (32-512)
|
||||
- name: Test on igpu for Mistral (load_low_bit 1024-128)
|
||||
shell: cmd
|
||||
run: |
|
||||
call conda activate igpu-perf
|
||||
|
|
@ -725,20 +724,20 @@ jobs:
|
|||
set BIGDL_LLM_XMX_DISABLED=1
|
||||
|
||||
cd python\llm\dev\benchmark\all-in-one
|
||||
move ..\..\..\test\benchmark\igpu-perf\32-512_434.yaml config.yaml
|
||||
move ..\..\..\test\benchmark\igpu-perf\1024-128_loadlowbit_434.yaml config.yaml
|
||||
set PYTHONIOENCODING=utf-8
|
||||
python run.py >> %CSV_SAVE_PATH%\32-512\log\%LOG_FILE% 2>&1
|
||||
python run.py >> %CSV_SAVE_PATH%\1024-128_loadlowbit\log\%LOG_FILE% 2>&1
|
||||
if %ERRORLEVEL% neq 0 (exit /b 1)
|
||||
|
||||
call conda deactivate
|
||||
|
||||
- name: Prepare igpu perf test for Qwen 1.5 (32-512)
|
||||
- name: Prepare igpu perf test for Qwen 1.5 (load_low_bit 1024-128)
|
||||
shell: bash
|
||||
run: |
|
||||
sed -i 's/{today}_test2/{today}_test3/g' python/llm/dev/benchmark/all-in-one/run.py
|
||||
sed -i "s/path to your local model hub/$MODEL_HUB_DIR/g" python/llm/test/benchmark/igpu-perf/32-512_437.yaml
|
||||
sed -i "s/path to your local model hub/$MODEL_HUB_DIR/g" python/llm/test/benchmark/igpu-perf/1024-128_loadlowbit_437.yaml
|
||||
|
||||
- name: Test on igpu for Qwen 1.5 (32-512)
|
||||
- name: Test on igpu for Qwen 1.5 (load_low_bit 1024-128)
|
||||
shell: cmd
|
||||
run: |
|
||||
call conda activate igpu-perf
|
||||
|
|
@ -749,14 +748,14 @@ jobs:
|
|||
set BIGDL_LLM_XMX_DISABLED=1
|
||||
|
||||
cd python\llm\dev\benchmark\all-in-one
|
||||
move ..\..\..\test\benchmark\igpu-perf\32-512_437.yaml config.yaml
|
||||
move ..\..\..\test\benchmark\igpu-perf\1024-128_loadlowbit_437.yaml config.yaml
|
||||
set PYTHONIOENCODING=utf-8
|
||||
python run.py >> %CSV_SAVE_PATH%\32-512\log\%LOG_FILE% 2>&1
|
||||
python run.py >> %CSV_SAVE_PATH%\1024-128_loadlowbit\log\%LOG_FILE% 2>&1
|
||||
if %ERRORLEVEL% neq 0 (exit /b 1)
|
||||
|
||||
call conda deactivate
|
||||
|
||||
- name: Concat csv and generate html (32-512)
|
||||
- name: Concat csv and generate html (load_low_bit 1024-128)
|
||||
shell: cmd
|
||||
run: |
|
||||
call conda activate html-gen
|
||||
|
|
@ -765,11 +764,11 @@ jobs:
|
|||
python ..\..\..\test\benchmark\concat_csv.py
|
||||
if %ERRORLEVEL% neq 0 (exit /b 1)
|
||||
del /q *test*.csv
|
||||
move *.csv %CSV_SAVE_PATH%\32-512\
|
||||
move *.csv %CSV_SAVE_PATH%\1024-128_loadlowbit\
|
||||
cd ..\..\..\test\benchmark
|
||||
python csv_to_html.py -f %CSV_SAVE_PATH%\32-512\
|
||||
python csv_to_html.py -f %CSV_SAVE_PATH%\1024-128_loadlowbit\
|
||||
if %ERRORLEVEL% neq 0 (exit /b 1)
|
||||
move %CSV_SAVE_PATH%\32-512\*.html %CSV_SAVE_PATH%
|
||||
move %CSV_SAVE_PATH%\1024-128_loadlowbit\*.html %CSV_SAVE_PATH%
|
||||
|
||||
call conda deactivate
|
||||
|
||||
|
|
|
|||
|
|
@ -23,7 +23,7 @@ num_beams: 1 # default to greedy search
|
|||
low_bit: 'sym_int4' # default to use 'sym_int4' (i.e. symmetric int4)
|
||||
batch_size: 1 # default to 1
|
||||
in_out_pairs:
|
||||
- '32-512'
|
||||
- '1024-128'
|
||||
test_api:
|
||||
- "transformer_int4_gpu_win" # on Intel GPU for Windows (catch GPU peak memory)
|
||||
- "transformer_int4_loadlowbit_gpu_win" # on Intel GPU for Windows (catch GPU peak memory)
|
||||
cpu_embedding: True # whether put embedding to CPU (only avaiable now for gpu win related test_api)
|
||||
|
|
@ -7,7 +7,7 @@ num_beams: 1 # default to greedy search
|
|||
low_bit: 'sym_int4' # default to use 'sym_int4' (i.e. symmetric int4)
|
||||
batch_size: 1 # default to 1
|
||||
in_out_pairs:
|
||||
- '32-512'
|
||||
- '1024-128'
|
||||
test_api:
|
||||
- "transformer_int4_gpu_win" # on Intel GPU for Windows (catch GPU peak memory)
|
||||
- "transformer_int4_loadlowbit_gpu_win" # on Intel GPU for Windows (catch GPU peak memory)
|
||||
cpu_embedding: True # whether put embedding to CPU (only avaiable now for gpu win related test_api)
|
||||
|
|
@ -7,7 +7,7 @@ num_beams: 1 # default to greedy search
|
|||
low_bit: 'sym_int4' # default to use 'sym_int4' (i.e. symmetric int4)
|
||||
batch_size: 1 # default to 1
|
||||
in_out_pairs:
|
||||
- '32-512'
|
||||
- '1024-128'
|
||||
test_api:
|
||||
- "transformer_int4_gpu_win" # on Intel GPU for Windows (catch GPU peak memory)
|
||||
- "transformer_int4_loadlowbit_gpu_win" # on Intel GPU for Windows (catch GPU peak memory)
|
||||
cpu_embedding: True # whether put embedding to CPU (only avaiable now for gpu win related test_api)
|
||||
Loading…
Reference in a new issue