Spaces:

devon7y
/

rat-qwen

Running on Zero

Persist Triton autotune; size GPU durations from measured speed; one padded batch per validation dose

e75de23 verified 20 days ago

296 Bytes

{
 "gpu": "NVIDIA RTX PRO 6000 Blackwell Server Edition MIG 2g.48gb",
 "torch": "2.11.0+cu130",
 "transformers": "5.20.2",
 "fla": true,
 "causal_conv1d": false,
 "fp8_param_frac": 0.8892,
 "norm_scale": 142.2164764404297,
 "tok_per_s": {
  "warmup": 0.71,
  "timed": 7.62
 },
 "mem_gb": 30.88
}

	{
	"gpu": "NVIDIA RTX PRO 6000 Blackwell Server Edition MIG 2g.48gb",
	"torch": "2.11.0+cu130",
	"transformers": "5.20.2",
	"fla": true,
	"causal_conv1d": false,
	"fp8_param_frac": 0.8892,
	"norm_scale": 142.2164764404297,
	"tok_per_s": {
	"warmup": 0.71,
	"timed": 7.62
	},
	"mem_gb": 30.88
	}