Qwen2.5
Benchmarks
TaskDatasetSOTA resultTrendResults
Qwen2.5-3B responses (test)
82.7Avg Preference Score
30
Qwen2.5-7B Adaptive AutoDAN-T attack
30ASR
6
Qwen2.5 72B (q=64, k=8) (1k)
103.61PrefixLM Attention Throughput (TFLOPS)
4
Qwen2.5-7B 8K context length
7.1Decode Latency (ms/token)
4