text-generation-inference

fe80f536 - feat(server): auto max_batch_total_tokens for flash att models (#630)