Align Qwen235 attention profile batch limit

This commit is contained in:
2026-07-19 06:21:33 +08:00
parent ac5061fa5d
commit 922d66c5c1

View File

@@ -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
}