# Base-2 vs. base-e log-sum-exp mismatch silently corrupts attention in SGLang

> Source: <https://twitter.com/advprop/status/2089046843690729620>
> Published: 2026-08-16 17:52:11+00:00

If you serve models with mla on sglang with --attention-backend flashinfer and you get radix cache hits above 8k your outputs are wrong right now.
DeepSeek V2/V3/R1/V3.1/V3.2, bailing, glm4moe_lite, Sarvam affected.
Fix:

# Konstantin on X: "If you serve models with mla on sglang with --attention-backend flashinfer and you get radix cache hits above 8k your outputs are wrong right now. DeepSeek V2/V3/R1/V3.1/V3.2, bailing, glm4moe_lite, Sarvam affected. Fix: https://t.co/IRxPRx9Ojf"

- flashinfer api proposal for clarity:
[github.com/flashinfer-ai/…](https://github.com/flashinfer-ai/flashinfer/pull/4547)some context: there is merge v2 api in sglang and it does: S = log(exp(s_a) + exp(s_b)) v = v_a·exp(s_a − S) + v_b·exp(s_b − S) which is true if and only if s_a and s_b are natural logs. but this is not true for