Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

5 sequences

Benchmarks

Task NameDataset NameSOTA ResultTrend
KV Cache Compression5 sequences Qwen2.5-1.5B-Instruct
KL Divergence0.0358
15
KV cache compression5 sequences Mean
KL Divergence0.0575
2
KV cache compression5 sequences Seq 0 easy
KL Divergence0.0081
2
Showing 3 of 3 rows