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 13, 00:45 UTC
[CI][NIXL] Change default `block_size` for tests (#28927)
Signed-off-by: NickLucche <nlucches@redhat.com>
Nicolò Lucchesi committed
Nov 18, 2025 at 18:22 UTC
f226a3f0c11aed72f585ebd2942d4a6832adbfb9
1 file changed
+2
-2
tests/v1/kv_connector/nixl_integration/run_accuracy_test.sh
+2
-2
@@ -49,8 +49,8 @@ NUM_DECODE_INSTANCES=${NUM_DECODE_INSTANCES:-1} # Default to 1
49
PREFILLER_TP_SIZE=${PREFILLER_TP_SIZE:-1}
50
DECODER_TP_SIZE=${DECODER_TP_SIZE:-1}
51
GPU_MEMORY_UTILIZATION=${GPU_MEMORY_UTILIZATION:-0.2}
52
-PREFILL_BLOCK_SIZE=${PREFILL_BLOCK_SIZE:-16}
53
-DECODE_BLOCK_SIZE=${DECODE_BLOCK_SIZE:-16}
52
+PREFILL_BLOCK_SIZE=${PREFILL_BLOCK_SIZE:-128}
53
+DECODE_BLOCK_SIZE=${DECODE_BLOCK_SIZE:-128}
54
55
# Find the git repository root directory
56
GIT_ROOT=$(git rev-parse --show-toplevel)