fix: tp oom

Author: ivanium

benchmarks/multi_node/srt-slurm-recipes/vllm/deepseek-v4/agentic/agg-gb300-tp4-agentic.yaml

@@ -109,7 +109,7 @@ backend:
       attention-config: '{"backend":"FLASHINFER_MLA_SPARSE_DSV4","use_prefill_query_quantization":true,"use_fp4_indexer_cache":true}'
       max-model-len: 1048576
       max-num-seqs: 32
-      max-num-batched-tokens: 32768
+      max-num-batched-tokens: 16384
       trust-remote-code: true
       no-enable-flashinfer-autotune: true
       block-size: 256