Skip to content

Commit efbba83

Browse files
zhulinJulia24zhulinJulia24
authored
[ci] add free disk before build test whl package and add session_len args in benchmark script (#4136)
1 parent 96e998a commit efbba83

File tree

9 files changed

+91
-1
lines changed

9 files changed

+91
-1
lines changed

‎.github/workflows/api_eval.yml‎

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -64,6 +64,18 @@ jobs:
6464
DOCKER_TAG: cuda12.8
6565
OUTPUT_FOLDER: cuda12.8_dist_${{ github.run_id }}
6666
steps:
67+
- name: Free disk space
68+
uses: jlumbroso/free-disk-space@main
69+
with:
70+
# This might remove tools that are actually needed, if set to "true" but frees about 6 GB
71+
tool-cache: false
72+
docker-images: false
73+
# All of these default to true, but feel free to set to "false" if necessary for your workflow
74+
android: true
75+
dotnet: true
76+
haskell: true
77+
large-packages: true
78+
swap-storage: false
6779
- name: Checkout repository
6880
uses: actions/checkout@v3
6981
with:

‎.github/workflows/benchmark.yml‎

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -46,6 +46,18 @@ jobs:
4646
PLAT_NAME: manylinux2014_x86_64
4747
DOCKER_TAG: cuda12.8
4848
steps:
49+
- name: Free disk space
50+
uses: jlumbroso/free-disk-space@main
51+
with:
52+
# This might remove tools that are actually needed, if set to "true" but frees about 6 GB
53+
tool-cache: false
54+
docker-images: false
55+
# All of these default to true, but feel free to set to "false" if necessary for your workflow
56+
android: true
57+
dotnet: true
58+
haskell: true
59+
large-packages: true
60+
swap-storage: false
4961
- name: Checkout repository
5062
uses: actions/checkout@v3
5163
with:

‎.github/workflows/daily_ete_test.yml‎

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -66,6 +66,18 @@ jobs:
6666
PLAT_NAME: manylinux2014_x86_64
6767
DOCKER_TAG: cuda12.8
6868
steps:
69+
- name: Free disk space
70+
uses: jlumbroso/free-disk-space@main
71+
with:
72+
# This might remove tools that are actually needed, if set to "true" but frees about 6 GB
73+
tool-cache: false
74+
docker-images: false
75+
# All of these default to true, but feel free to set to "false" if necessary for your workflow
76+
android: true
77+
dotnet: true
78+
haskell: true
79+
large-packages: true
80+
swap-storage: false
6981
- name: Checkout repository
7082
uses: actions/checkout@v3
7183
with:

‎.github/workflows/daily_ete_test_3090.yml‎

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -65,6 +65,18 @@ jobs:
6565
PLAT_NAME: manylinux2014_x86_64
6666
DOCKER_TAG: cuda12.4
6767
steps:
68+
- name: Free disk space
69+
uses: jlumbroso/free-disk-space@main
70+
with:
71+
# This might remove tools that are actually needed, if set to "true" but frees about 6 GB
72+
tool-cache: false
73+
docker-images: false
74+
# All of these default to true, but feel free to set to "false" if necessary for your workflow
75+
android: true
76+
dotnet: true
77+
haskell: true
78+
large-packages: true
79+
swap-storage: false
6880
- name: Checkout repository
6981
uses: actions/checkout@v3
7082
with:

‎.github/workflows/daily_ete_test_5080.yml‎

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -65,6 +65,18 @@ jobs:
6565
PLAT_NAME: manylinux2014_x86_64
6666
DOCKER_TAG: cuda12.8
6767
steps:
68+
- name: Free disk space
69+
uses: jlumbroso/free-disk-space@main
70+
with:
71+
# This might remove tools that are actually needed, if set to "true" but frees about 6 GB
72+
tool-cache: false
73+
docker-images: false
74+
# All of these default to true, but feel free to set to "false" if necessary for your workflow
75+
android: true
76+
dotnet: true
77+
haskell: true
78+
large-packages: true
79+
swap-storage: false
6880
- name: Checkout repository
6981
uses: actions/checkout@v3
7082
with:

‎.github/workflows/daily_ete_test_h800.yml‎

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -65,6 +65,18 @@ jobs:
6565
PLAT_NAME: manylinux2014_x86_64
6666
DOCKER_TAG: cuda12.8
6767
steps:
68+
- name: Free disk space
69+
uses: jlumbroso/free-disk-space@main
70+
with:
71+
# This might remove tools that are actually needed, if set to "true" but frees about 6 GB
72+
tool-cache: false
73+
docker-images: false
74+
# All of these default to true, but feel free to set to "false" if necessary for your workflow
75+
android: true
76+
dotnet: true
77+
haskell: true
78+
large-packages: true
79+
swap-storage: false
6880
- name: Checkout repository
6981
uses: actions/checkout@v3
7082
with:

‎.github/workflows/evaluate.yml‎

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -56,6 +56,18 @@ jobs:
5656
DOCKER_TAG: cuda12.8
5757
OUTPUT_FOLDER: cuda12.8_dist_${{ github.run_id }}
5858
steps:
59+
- name: Free disk space
60+
uses: jlumbroso/free-disk-space@main
61+
with:
62+
# This might remove tools that are actually needed, if set to "true" but frees about 6 GB
63+
tool-cache: false
64+
docker-images: false
65+
# All of these default to true, but feel free to set to "false" if necessary for your workflow
66+
android: true
67+
dotnet: true
68+
haskell: true
69+
large-packages: true
70+
swap-storage: false
5971
- name: Checkout repository
6072
uses: actions/checkout@v3
6173
with:

‎autotest/utils/benchmark_utils.py‎

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -105,12 +105,14 @@ def longtext_throughput_test(config,
105105
for input_len, out_len, num_prompts, case_name, concurrency in [(1, 32768, 20, '32k', None),
106106
(1, 65536, 10, '64k', None),
107107
(198000, 1024, 3, '198k', 1)]:
108+
session_len = input_len + out_len
108109
csv_path = f'{benchmark_path}/longtext_{case_name}_1th.csv'
109110
benchmark_log = os.path.join(
110111
log_path, f'benchmark_longtext_throughput_{case_name}' + model.split('/')[1] + worker_id + '.log')
111112
cmd = ' '.join([
112113
command, '--dataset-name random', f'--random-input-len {input_len}', f'--random-output-len {out_len}',
113-
f'--num-prompts {num_prompts}', '--stream-output', f'--csv {csv_path}'
114+
f'--num-prompts {num_prompts}', '--stream-output', f'--session-len {session_len}', '--random-range-ratio 1',
115+
f'--csv {csv_path}'
114116
])
115117
if concurrency:
116118
cmd += f' --concurrency {concurrency}'

‎benchmark/profile_pipeline_api.py‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -259,13 +259,15 @@ def parse_args():
259259

260260
tp_act = ArgumentHelper.tp(pt_group)
261261
cache_count_act = ArgumentHelper.cache_max_entry_count(pt_group)
262+
session_len_act = ArgumentHelper.session_len(pt_group)
262263
cache_block_seq_len_act = ArgumentHelper.cache_block_seq_len(pt_group)
263264
prefix_caching_act = ArgumentHelper.enable_prefix_caching(pt_group)
264265

265266
# turbomind engine args
266267
tb_group = parser.add_argument_group('TurboMind engine argument')
267268
tb_group._group_actions.append(tp_act)
268269
tb_group._group_actions.append(cache_count_act)
270+
tb_group._group_actions.append(session_len_act)
269271
tb_group._group_actions.append(cache_block_seq_len_act)
270272
tb_group._group_actions.append(prefix_caching_act)
271273
ArgumentHelper.model_format(tb_group, default='hf')
@@ -287,6 +289,7 @@ def main():
287289
max_batch_size=args.concurrency,
288290
tp=args.tp,
289291
cache_max_entry_count=args.cache_max_entry_count,
292+
session_len=args.session_len,
290293
cache_block_seq_len=args.cache_block_seq_len,
291294
model_format=args.model_format,
292295
quant_policy=args.quant_policy,
@@ -298,6 +301,7 @@ def main():
298301
elif args.backend == 'pytorch':
299302
engine_config = PytorchEngineConfig(
300303
cache_max_entry_count=args.cache_max_entry_count,
304+
session_len=args.session_len,
301305
block_size=args.cache_block_seq_len,
302306
max_batch_size=args.concurrency,
303307
tp=args.tp,

0 commit comments

Comments
 (0)