List your company

Multi-turn Dialogue Processing on BCP

17.1Throughput (k)Full-cache
7.7410.1712.615.03Jun 6, 2026

Evaluation results

MethodLinks
Full-cacheBackbone=Qwen2.5-14B, Capacity (C)=16k2026.0617.1203.825.227.1
H2OBackbone=Qwen2.5-14B, Capacity (C)=16k2026.0615.2521.380.958.7
IntentKV-Phase-2Backbone=Qwen2.5-14B, Capacity (C)=16k2026.0615.110528.824.1
SnapKVBackbone=Qwen2.5-14B, Capacity (C)=16k2026.0614.9447.570.752.6
StreamingLLMBackbone=Qwen2.5-14B, Capacity (C)=16k2026.0614.5292.464.551.8
SnapKVBackbone=Qwen2.5-14B, Capacity (C)=8k2026.0612.7574.210156.5
H2OBackbone=Qwen2.5-14B, Capacity (C)=8k2026.0612.6373.193.652.4
IntentKV-Phase-2Backbone=Qwen2.5-14B, Capacity (C)=8k2026.0611.810426.721
Full-cacheBackbone=Qwen3-8B, Capacity (C)=16k2026.0610.6131.432.245.1
StreamingLLMBackbone=Qwen2.5-14B, Capacity (C)=8k2026.0610.5275.368.344.5
H2OBackbone=Qwen3-8B, Capacity (C)=16k2026.0610.2244.765.563.6
StreamingLLMBackbone=Qwen3-8B, Capacity (C)=16k2026.069.9165.35051.5
SnapKVBackbone=Qwen3-8B, Capacity (C)=16k2026.069.9228.260.759.1
IntentKV-Phase-2Backbone=Qwen3-8B, Capacity (C)=16k2026.069.9139.23243.2
H2OBackbone=Qwen3-8B, Capacity (C)=8k2026.069.1281.477.259.7
SnapKVBackbone=Qwen3-8B, Capacity (C)=8k2026.069279.277.159.4
StreamingLLMBackbone=Qwen3-8B, Capacity (C)=8k2026.068.4180.654.345.2
IntentKV-Phase-2Backbone=Qwen3-8B, Capacity (C)=8k2026.068.1140.431.639.8