From 2234a2da49fc5b3980d7a73001ab2135e3c1e4a7 Mon Sep 17 00:00:00 2001
From: functionstackx <47992694+functionstackx@users.noreply.github.com>
Date: Fri, 3 Jul 2026 21:34:20 -0400
Subject: [PATCH] Test: add hf-overrides indexer cache to MiniMax M3 H200 vLLM
 (do not merge)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 benchmarks/single_node/fixed_seq_len/minimaxm3_fp8_h200.sh | 1 +
 1 file changed, 1 insertion(+)

diff --git a/benchmarks/single_node/fixed_seq_len/minimaxm3_fp8_h200.sh b/benchmarks/single_node/fixed_seq_len/minimaxm3_fp8_h200.sh
index 057c0c230b..e745541db8 100755
--- a/benchmarks/single_node/fixed_seq_len/minimaxm3_fp8_h200.sh
+++ b/benchmarks/single_node/fixed_seq_len/minimaxm3_fp8_h200.sh
@@ -85,6 +85,7 @@ $PARALLEL_ARGS \
 --max-cudagraph-capture-size $CAPTURE_SIZE \
 --max-num-batched-tokens "$((ISL * 2 ))" \
 --stream-interval 20 --no-enable-prefix-caching \
+--hf-overrides '{"use_index_cache": true, "index_topk_freq": 4}' \
 --trust-remote-code > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!