Use Ollama parallelism for startup benchmark

workers/openai/worker.py

@@ -12,6 +12,7 @@ MODEL_SERVER_URL           = 'http://127.0.0.1'
 MODEL_SERVER_PORT          = 11434
 MODEL_LOG_FILE             = os.environ.get('MODEL_LOG', '/var/log/ollama.log')
 MODEL_HEALTHCHECK_ENDPOINT = "/"
+BENCHMARK_CONCURRENCY      = max(1, int(os.environ.get("OLLAMA_NUM_PARALLEL", "1")))
 
 # Ollama-specific log messages
 def request_parser(request):
@@ -66,7 +67,7 @@ worker_config = WorkerConfig(
             max_queue_time=600.0,
             benchmark_config=BenchmarkConfig(
                 generator=completions_benchmark_generator,
-                concurrency=10,
+                concurrency=BENCHMARK_CONCURRENCY,
                 runs=3
             )
         ),
@@ -85,4 +86,4 @@ worker_config = WorkerConfig(
     )
 )
 
-Worker(worker_config).run()
+Worker(worker_config).run()