Release calibrated Image 2.1 Turbo FP8 transformer, accelerated SM120 runtime, quality evidence and real-time demo

ffc54ec verified 24 days ago

501 Bytes

{
  "torch": "2.14.0+cu130",
  "cuda": "13.0",
  "diffusers": "0.41.0",
  "transformers": "5.17.0",
  "triton": "3.8.0",
  "flashinfer": "0.7.1",
  "sageattention": "2.2.0",
  "gpu": "NVIDIA RTX PRO 6000 Blackwell Workstation Edition",
  "capability": [
    12,
    0
  ],
  "driver": "615.71.09",
  "python": "3.12.3",
  "os": "Linux-7.0.0-34-generic-x86_64-with-glibc2.43",
  "sageattention_source": "github.com/thu-ml/SageAttention commit d1a57a546c3d395b1ffcbeecc66d81db76f3b4b5 (built locally)"
}

	{
	"torch": "2.14.0+cu130",
	"cuda": "13.0",
	"diffusers": "0.41.0",
	"transformers": "5.17.0",
	"triton": "3.8.0",
	"flashinfer": "0.7.1",
	"sageattention": "2.2.0",
	"gpu": "NVIDIA RTX PRO 6000 Blackwell Workstation Edition",
	"capability": [
	12,
	0
	],
	"driver": "615.71.09",
	"python": "3.12.3",
	"os": "Linux-7.0.0-34-generic-x86_64-with-glibc2.43",
	"sageattention_source": "github.com/thu-ml/SageAttention commit d1a57a546c3d395b1ffcbeecc66d81db76f3b4b5 (built locally)"
	}