[CI][PD] Add hybrid SSM P_TP>D_TP accuracy sweep entry (#49593)
Signed-off-by: NickLucche <nicolo.lucchesi@mistral.ai> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
Nicolò Lucchesi committed
Jul 23, 2026 at 23:49 UTC
26594674975cf68b4b33b329b9a9ad2d508bb243
1 file changed
+1
tests/v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
+1
@@ -25,6 +25,7 @@ dp_ep_configs=(
25
hybrid_ssm_configs=(
26
"VLLM_SSM_CONV_STATE_LAYOUT=DS GPU_MEMORY_UTILIZATION=0.8 MODEL_NAMES=ibm-granite/granite-4.0-h-tiny VLLM_SERVE_EXTRA_ARGS=--max-model-len,8192,--trust-remote-code"
27
"VLLM_SSM_CONV_STATE_LAYOUT=DS PREFILLER_TP_SIZE=2 DECODER_TP_SIZE=2 GPU_MEMORY_UTILIZATION=0.8 MODEL_NAMES=ibm-granite/granite-4.0-h-tiny VLLM_SERVE_EXTRA_ARGS=--max-model-len,8192,--trust-remote-code"
28
+ "VLLM_SSM_CONV_STATE_LAYOUT=DS PREFILLER_TP_SIZE=2 DECODER_TP_SIZE=1 GPU_MEMORY_UTILIZATION=0.8 MODEL_NAMES=ibm-granite/granite-4.0-h-tiny VLLM_SERVE_EXTRA_ARGS=--max-model-len,8192,--trust-remote-code"
29
# GDN (Qwen3.5)
30
"VLLM_SSM_CONV_STATE_LAYOUT=DS GPU_MEMORY_UTILIZATION=0.8 MODEL_NAMES=Qwen/Qwen3.5-0.8B"
31
"VLLM_SSM_CONV_STATE_LAYOUT=DS PREFILLER_TP_SIZE=1 DECODER_TP_SIZE=2 GPU_MEMORY_UTILIZATION=0.8 MODEL_NAMES=Qwen/Qwen3.5-0.8B"