Skip to content

Commit 55172b0

Browse files
authored
Fix model selection (#43)
* Add ROCm benchmarks Signed-off-by: Huy Do <huydhn@gmail.com> * Fix typo Signed-off-by: Huy Do <huydhn@gmail.com> * Add missing argument Signed-off-by: Huy Do <huydhn@gmail.com> * More tweaks Signed-off-by: Huy Do <huydhn@gmail.com> * Use rocm-smi Signed-off-by: Huy Do <huydhn@gmail.com> * Authenticate with AWS on ROCm Signed-off-by: Huy Do <huydhn@gmail.com> * Ready for review Signed-off-by: Huy Do <huydhn@gmail.com> * Beautifier Signed-off-by: Huy Do <huydhn@gmail.com> * Add id-token Signed-off-by: Huy Do <huydhn@gmail.com> * Use DEVICE_NAME and DEVICE_TYPE consistently Signed-off-by: Huy Do <huydhn@gmail.com> * Fix model selection Signed-off-by: Huy Do <huydhn@gmail.com> --------- Signed-off-by: Huy Do <huydhn@gmail.com>
1 parent 319ad22 commit 55172b0

File tree

1 file changed

+6
-5
lines changed

1 file changed

+6
-5
lines changed

.github/scripts/generate_vllm_benchmark_matrix.py

Lines changed: 6 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -113,13 +113,12 @@ def generate_benchmark_matrix(
113113
Parse all the JSON files in vLLM benchmark configs directory to get the
114114
model name and tensor parallel size (aka number of GPUs)
115115
"""
116-
get_all_models = True if not models else False
117116
use_all_gpus = True if not gpus else False
118-
119117
benchmark_matrix: Dict[str, Any] = {
120118
"include": [],
121119
}
122120

121+
selected_models = []
123122
for file in glob.glob(f"{benchmark_configs_dir}/*.json"):
124123
with open(file) as f:
125124
try:
@@ -139,10 +138,12 @@ def generate_benchmark_matrix(
139138
model = benchmark_config["model"].lower()
140139

141140
# Dedup
142-
if model in models:
141+
if model in selected_models:
142+
continue
143+
# and only choose the selected model:
144+
if models and model not in models:
143145
continue
144-
if get_all_models:
145-
models.append(model)
146+
selected_models.append(model)
146147

147148
if "tensor_parallel_size" in benchmark_config:
148149
tp = benchmark_config["tensor_parallel_size"]

0 commit comments

Comments
 (0)