From 922d66c5c11d0f1e006d1666309cf86b7311dfc6 Mon Sep 17 00:00:00 2001
From: Gahow Wang <gahow.wang@gmail.com>
Date: Sun, 19 Jul 2026 06:21:33 +0800
Subject: [PATCH] Align Qwen235 attention profile batch limit

---
 .../frontier-fidelity-envelope-v1/run_qwen235_v020_profiles.sh | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/runs/frontier-fidelity-envelope-v1/run_qwen235_v020_profiles.sh b/runs/frontier-fidelity-envelope-v1/run_qwen235_v020_profiles.sh
index 1dc0e42..dee2a83 100644
--- a/runs/frontier-fidelity-envelope-v1/run_qwen235_v020_profiles.sh
+++ b/runs/frontier-fidelity-envelope-v1/run_qwen235_v020_profiles.sh
@@ -36,7 +36,8 @@ common_profile() {
     "${VENV_ROOT}/bin/python" -m frontier.profiling.attention.main \
       --disable_ray --num_gpus 1 --device h20 --output_dir "${root}" --models "${MODEL}" \
       --num_tensor_parallel_workers 4 8 --max_model_len 40960 --max_seq_len 40960 \
-      --batch_size_list "${BATCHES[@]}" --decode_kv_cache_size_list "${KV[@]}" \
+      --max_batch_size 256 --batch_size_list "${BATCHES[@]}" \
+      --decode_kv_cache_size_list "${KV[@]}" \
       --fixed_chunked_prefill_size 8192 --attention_backend FLASHINFER \
       --profile_method "${method}" --yes
 }