Sponsored Search Retrieval
64.8Precision@100Qwen3-8B (CL)
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen3-8B (CL)Hidden dim=4096, FFN dim=12288, Query Heads=32, Key-Value Heads=8, Number of Parameters=7.57B, Latency (ms)=186.14, Throughput (queries/sec)=3382026.05 | 64.8 | |
| Qwen3-0.6B (Align + CR)Hidden dim=1024, FFN dim=3072, Query Heads=16, Key-Value Heads=8, Number of Parameters=0.60B, Latency (ms)=40.96, Throughput (queries/sec)=1,4612026.05 | 64.3 | |
| Pruned 4L layers (Align + CR)Hidden dim=1024, FFN dim=1024, Query Heads=16, Key-Value Heads=8, Number of Parameters=0.19B, Latency (ms)=6.80, Throughput (queries/sec)=6,8082026.05 | 63.1 | |
| Qwen3-4B (CL)Hidden dim=2560, FFN dim=9728, Query Heads=32, Key-Value Heads=8, Number of Parameters=4.02B, Latency (ms)=117.81, Throughput (queries/sec)=5302026.05 | 62.4 |