fix: fix benchmark and plotting script path in run_bench.sh

Author: yuezhu1

bench/run_bench.sh

@@ -10,7 +10,7 @@ export ROUTER_MODELS="auto"
 export VLLM_MODELS="openai/gpt-oss-20b"
 
 # Run the benchmark
-python src/training/model_eval/router_reason_bench.py \
+python router_reason_bench.py \
   --run-router \
   --router-endpoint "$ROUTER_ENDPOINT" \
   --router-api-key "$ROUTER_API_KEY" \
@@ -19,7 +19,7 @@ python src/training/model_eval/router_reason_bench.py \
   --vllm-endpoint "$VLLM_ENDPOINT" \
   --vllm-api-key "$VLLM_API_KEY" \
   --vllm-models "$VLLM_MODELS" \
-  --samples-per-category 20 \
+  --samples-per-category 5 \
   --vllm-exec-modes NR XC \
   --concurrent-requests 4 \
   --output-dir results/reasonbench
@@ -29,6 +29,6 @@ VLLM_MODEL_FIRST="${VLLM_MODELS%% *}"
 ROUTER_MODEL_FIRST="${ROUTER_MODELS%% *}"
 VLLM_MODELS_SAFE="${VLLM_MODEL_FIRST//\//_}"
 ROUTER_MODELS_SAFE="${ROUTER_MODEL_FIRST//\//_}"
-python src/training/model_eval/bench_plot.py \
+python bench_plot.py \
   --summary "results/reasonbench/vllm::${VLLM_MODELS_SAFE}/summary.json" \
   --router-summary "results/reasonbench/router::${ROUTER_MODELS_SAFE}/summary.json"