I’m working on a project to generate text from a 1.2B parameter full precision LLM (5gb)
Unfortunately I’m limited in the infrastructure I can use to deploy this model. There is no batch inference supported. The infrastructure I have allows me to deploy a copy of the model on a single A100, 1 per process with up to 9 processes supported (these are called “replicas”). I understand that this makes little sense given my model is memory bound, and each process will fight for memory bandwidth to read in the same weights, but I can’t change that for now.
My average input and output tokens are roughly 1000 each. I estimate the kv cache per token is roughly 400kB using full precision.
I have benchmarks of the latency of the model using various “replicas” as described above. I wanted to compare this to the theoretical performance of the A100. For my use case time to first token is negligible (<200ms), and generation is memory bound.
I find that with 5 or more replicas, the math works out and my model is roughly as fast as I expect. For example, with 1000 output tokens, 6 replicas, it’s like I’m generating using a batch of 6 requests from a 30gb model + 5gb for the kv cache. At a memory bandwidth around 1-1.3tbps that translates to ~30s per request, which is not far from what I see. The same goes for other replica numbers, 5, 7, 8 and 9.
However, when I run with a single replica, I expect generation to hover around the 5-6s mark on average. Instead, I see > 20s. I need to add 4 more replicas before the number starts to make sense. It almost seems like the model takes up too little memory to be allocated the entire memory bandwidth.
Does anyone know where this extra latency could be coming from? Do models have to reach a certain amount of used memory for A100 memory bandwidth to hit their available memory bandwidth?
Replicas shouldn't affect anything except memory pressure on the GPU and maybe a little overhead if you're sending inference requests one at a time.
The replicas allow you to run multiple inference requests simultaneously.
If you're using python, they are threads not "processes" though in unixspeak they're separate processes.
So once you have your inference container up and running, the latency per request should be relatively constant and shouldn't vary with the number of replicas.
With sufficient memory bandwidth, more replicas (up to what the gpu memory can support) should increase throughout proportional to the number of replicas.
If you're memory bandwidth limited, you should see a sublinear increase in throughput.