A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 2, 12:15 UTC
[CI] Enable prefix caching in BFCL benchmark (#43925)
Signed-off-by: Yifan Zong <yzong@redhat.com>
yzong-rh committed
May 28, 2026 at 19:36 UTC
325a1ec4fbc007b7860238fff5fe4ba9e023b8cd
1 file changed
+2
-2
.buildkite/scripts/tool_call/run-bfcl-eval.sh
+2
-2
@@ -70,7 +70,7 @@ echo "============================================"
70
# ---- Install bfcl-eval if missing ----
71
if ! python3 -c "import bfcl_eval" 2>/dev/null; then
72
echo "Installing bfcl-eval..."
73
- pip install "bfcl-eval>=2025.10.20.1,<2026"
73
+ uv pip install "bfcl-eval>=2025.10.20.1,<2026"
74
fi
75
76
# ---- Cleanup handler ----
@@ -100,7 +100,7 @@ SERVE_ARGS=(
100
--tensor-parallel-size "$TP_SIZE"
101
--max-model-len "$MAX_MODEL_LEN"
102
--enforce-eager
103
- --no-enable-prefix-caching
103
+ --enable-prefix-caching
104
)
105
106
# Append reasoning parser if specified