Production-ready offline FP8 checkpoint for vLLM — 47% less VRAM, 80% more concurrency vs BF16.
Browse all Text Generation models →