revert: drop MAX_MODEL_LEN=131072 default from Kimi MI355X/B200 launchers

Per agentic benchmark design: must not cap context. Removes the
MAX_MODEL_LEN=131072 default from kimik2.5_fp4_mi355x,
kimik2.5_fp4_b200, and kimik2.5_int4_b200 launchers so vLLM uses the
model's native context window (matches the H100/H200/B300 launchers
that already have no cap).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: cquil11

benchmarks/single_node/agentic/kimik2.5_fp4_b200.sh

@@ -16,11 +16,6 @@ DURATION=${DURATION:-1800}
 MAX_DELAY=${MAX_DELAY:-60}
 ADVANCE_MIN=${ADVANCE_MIN:-0.0}
 ADVANCE_MAX=${ADVANCE_MAX:-0.7}
-# Agentic matrix entries don't set max-model-len, so the workflow passes 0.
-# ${:-DEFAULT} only fires on unset/empty, so handle 0 explicitly.
-if [ -z "${MAX_MODEL_LEN:-}" ] || [ "$MAX_MODEL_LEN" = "0" ]; then
-    MAX_MODEL_LEN=131072
-fi
 
 if [[ -n "${SLURM_JOB_ID:-}" ]]; then
     echo "JOB $SLURM_JOB_ID running on ${SLURMD_NODENAME:-unknown}"
@@ -64,7 +59,6 @@ vllm serve $MODEL \
 --port $PORT \
 --tensor-parallel-size=$TP \
 --gpu-memory-utilization 0.90 \
---max-model-len $MAX_MODEL_LEN \
 --max-num-seqs $CONC \
 --reasoning-parser kimi_k2 \
 --tool-call-parser kimi_k2 \

benchmarks/single_node/agentic/kimik2.5_fp4_mi355x.sh

@@ -17,9 +17,6 @@ MAX_DELAY=${MAX_DELAY:-60}
 ADVANCE_MIN=${ADVANCE_MIN:-0.0}
 ADVANCE_MAX=${ADVANCE_MAX:-0.7}
 EP_SIZE=${EP_SIZE:-1}
-if [ -z "${MAX_MODEL_LEN:-}" ] || [ "$MAX_MODEL_LEN" = "0" ]; then
-    MAX_MODEL_LEN=131072
-fi
 
 if [[ -n "${SLURM_JOB_ID:-}" ]]; then
     echo "JOB $SLURM_JOB_ID running on ${SLURMD_NODENAME:-unknown}"
@@ -84,7 +81,6 @@ vllm serve $MODEL \
 --tensor-parallel-size=$TP \
 $EP \
 --gpu-memory-utilization 0.90 \
---max-model-len $MAX_MODEL_LEN \
 --block-size=1 \
 --trust-remote-code \
 --max-num-seqs $CONC \

benchmarks/single_node/agentic/kimik2.5_int4_b200.sh

@@ -16,9 +16,6 @@ DURATION=${DURATION:-1800}
 MAX_DELAY=${MAX_DELAY:-60}
 ADVANCE_MIN=${ADVANCE_MIN:-0.0}
 ADVANCE_MAX=${ADVANCE_MAX:-0.7}
-if [ -z "${MAX_MODEL_LEN:-}" ] || [ "$MAX_MODEL_LEN" = "0" ]; then
-    MAX_MODEL_LEN=131072
-fi
 
 if [[ -n "${SLURM_JOB_ID:-}" ]]; then
     echo "JOB $SLURM_JOB_ID running on ${SLURMD_NODENAME:-unknown}"
@@ -58,7 +55,6 @@ vllm serve $MODEL \
 --port $PORT \
 --gpu-memory-utilization 0.95 \
 --tensor-parallel-size $TP \
---max-model-len $MAX_MODEL_LEN \
 --max-num-seqs $CONC \
 --reasoning-parser kimi_k2 \
 --tool-call-parser kimi_k2 \