{"id":"2606.03458","title":"KVarN: Variance-Normalized KV-Cache Quantization Mitigates Error Accumulation in Reasoning Tasks","url":"https://codeofpaper.com/paper/2606.03458","arxiv_url":"https://arxiv.org/abs/2606.03458","published":"2026-06-02","code":"official","best":{"full_name":"huawei-csl/KVarN","url":"https://github.com/huawei-csl/KVarN","tier":"official","stars":507,"license":"Apache-2.0","framework":null,"language":"Python","last_commit":"2026-06-03T13:46:53"},"most_starred_community":null,"implementations":[{"full_name":"huawei-csl/KVarN","url":"https://github.com/huawei-csl/KVarN","tier":"official","stars":507,"license":"Apache-2.0","framework":null,"language":"Python","last_commit":"2026-06-03T13:46:53"}],"possible_matches":0}