Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

OPT

Benchmarks

Task NameDataset NameSOTA ResultTrend
Inference LatencyOPT model family
Latency (ms)6.2
79
QuantizationOPT
Processing Time (s)4.8
46
QuantizationOPT v1 (train)
Processing Time (min)0.08
23
INT2 QuantizationOPT-1.3B
Memory Footprint (GB)4.49
12
INT2 QuantizationOPT-125M
Memory Cost (GB)1.39
12
Block-wise quantization complexity analysisOPT Models
GFLOPS0.24
12
INT2 QuantizationOPT-6.7B
Memory Footprint (GB)11.68
10
Open domain dialogueOPT-66B ZS
RSR74.7
9
Energy Consumption EstimationOPT-66B
Energy (J)1,786.62
8
Energy Consumption EstimationOPT-30B
Energy (J)846.54
8
Energy Consumption EstimationOPT-13B
Energy (J)385.31
8
Energy Consumption EstimationOPT 6.7B
Energy (J)215.27
8
Energy Consumption EstimationOPT-2.7B
Energy (J)95.67
8
Energy Consumption EstimationOPT 1.3B
Energy (J)55.8
8
Energy Consumption EstimationOPT-350M
Energy (J)20.17
8
Energy Consumption EstimationOPT-125M
Energy Consumption (J)11.41
8
Text-to-Image GenerationOPT-66B ZS user input pool v1.4 (test)
RSR71.34
5
Inference LatencyOPT-30B
Latency (ms)15.7
5
Inference LatencyOPT-175B first FFN layer
Latency (ms)0.225
5
Machine UnlearningOPT-125 Personal Set
Accuracy77.4
4
Machine UnlearningOPT-125 Forget Set
Accuracy65.3
4
Text GenerationOPT 512 prompt + 512 generation tokens 6.7B
Throughput (tokens/sec)72.5
4
Text GenerationOPT 512 prompt + 32 generation tokens 6.7B
Throughput (token/s)50.5
4
Text GenerationOPT 512 prompt + 1024 generation tokens 6.7B
Throughput (token/s)62.3
3
Showing 24 of 24 rows