fix bench script + align concurrencies

Signed-off-by: jthomson04 <[email protected]>

Author: jthomson04

components/backends/trtllm/performance_sweeps/scripts/bench.sh

@@ -178,8 +178,6 @@ fi
 mkdir -p ${log_path}/results
 echo "Starting benchmark..."
 for concurrency in ${concurrency_list}; do
-    original_concurrency=${concurrency}
-    concurrency=$((concurrency * num_gen_servers))
     num_prompts=$((concurrency * multi_round))
     echo "Benchmarking with concurrency ${concurrency} ... ${num_prompts} prompts"
     mkdir -p ${log_path}/concurrency_${concurrency}
@@ -204,7 +202,7 @@ for concurrency in ${concurrency_list}; do
             --port ${port} \
             --save-result \
             --result-dir "${log_path}/results" \
-            --result-filename "results_concurrency_${original_concurrency}_gpus_${total_gpus}_ctx_${prefill_gpus}_gen_${decode_gpus}.json"
+            --result-filename "results_concurrency_${concurrency}_gpus_${total_gpus}_ctx_${prefill_gpus}_gen_${decode_gpus}.json"
     else
         aiperf profile \
     	    --model ${model} \
@@ -226,32 +224,12 @@ for concurrency in ${concurrency_list}; do
     	    --warmup-request-count $(($concurrency*2)) \
 	    --num-dataset-entries ${num_prompts} \
     	    --random-seed 100 \
-    	    --artifact-dir "${log_path}/results/concurrency_${original_concurrency}" \
+    	    --artifact-dir "${log_path}/results/concurrency_${concurrency}" \
     	    --ui simple \
 	    -v \
     	    -H 'Authorization: Bearer NOT USED' \
     	    -H 'Accept: text/event-stream'    
     fi
-    python3 ${SCRIPTS_DIR}/scripts/bench/benchmark_serving.py \
-        --served-model-name ${model} \
-        --model ${model_path} \
-        --dataset-name random \
-        --num-prompts "$num_prompts" \
-        --random-input-len ${isl} \
-        --random-output-len ${osl} \
-        --random-range-ratio 0.8 \
-        --use-chat-template \
-        --ignore-eos \
-        --use-chat-template \
-        --backend "dynamo" \
-        --endpoint "/v1/completions" \
-        --percentile-metrics ttft,tpot,itl,e2el \
-        --max-concurrency "$concurrency" \
-        --host ${hostname} \
-        --port ${port} \
-        --save-result \
-        --result-dir "${log_path}/results" \
-        --result-filename "results_concurrency_${original_concurrency}_gpus_${total_gpus}_ctx_${prefill_gpus}_gen_${decode_gpus}.json"
 
     echo "Benchmark with concurrency ${concurrency} done"
 done