Performance degrades over time when running in Docker with Nvidia GPU

### What is the issue? I am working in a multi-GPU environment. I set up multiple docker containers to assign each GPU to it so I can process my workload in parallel. Here is the command I use to set up the container: `sudo docker run -d --gpus device=GPU-46b6fece-aec9-853f-0956-2d43359e28e3 -v ollama:/root/.ollama -p 11435:11434 --name ollama0 ollama/ollama` I change the port for each container and use a list of clients to split the workload. I noticed the performance of the Ollama Docker container degrades significantly over time. I am processing a workload with over 134,000 queries with llama3:instruct. In the beginning, the processing speed is about 1 to 2 items/s, after processing a few thousands of queries, it slows down to 10 to 12 items/s, and it gets worse over time. If I remove and reconfigure the container, The performance will return to normal. ### OS Linux ### GPU Nvidia ### CPU AMD ### Ollama version 0.1.38