Qeravio
Canonical AI event

Qwen4exp adds recurrent state rollback for faster decoding.

The recurrent state rollback feature improves decoding speed to 183 tok/s on code and 144 tok/s on prose.

2 Sept 20261 verified claims1 sources1 observations
What happened

The official source reports this update: b10731: qwen4exp: support recurrent state rollback. qwen4exp: support recurrent state rollback ( #28123 ) MTP speculative decoding needs the target state to move back by the number of rejected draft tokens. Without rollback support the context is classified as SEQ_RM_TYPE_FULL and the server serializes the whole recurrent state to host memory on every round, which costs more than the drafting saves. The recurrent cache already holds n_rs_seq + 1 snapshot planes and the delta net writes its SSM state into them, but build_conv_state_at wrote a single plane, so a rollback restored a convolution history that was never captured.

Why it matters

This official update documents a development concerning b10731: qwen4exp: support recurrent state rollback. Its practical significance depends on the scope and evidence stated by the source.

What to watch next

Read the official source update and verify its stated scope, evidence and timing before acting on it.

Connected knowledge

Entities affected by this event

Evidence trail

Sources behind the event