Model techniques map
Techniquesinference & servingKV cache management

specific method · filed under inference & serving

Chunked prefill

Processes prefill in chunks; the evidence also mentions retaining chunked encoding for temporal audio and vision outputs.

Also called --chunked-prefill-size 16384, chunked prefilling, chunked-prefilling mechanism.

sources
6
models
4
labs adopt it
3
strongest
default

How sources treat it

One count per evidence span, weakest treatment to strongest.

optional 1used 3default 2

Documented in

Further reading

Picked by hand, not extracted: where to read more, not evidence for anything on this page.

Evidence

6 spans quoted from the sources, strongest treatment first.

All deployment snippets below default to port 8000, with chunked prefill, NVFP4 KV caching, and MTP (5 speculative tokens) enabled

defaultinference servingin Nemotron 3 UltraNVIDIA

enable-chunked-prefill

defaultinference servingin Nemotron 3 UltraNVIDIA

--enable-chunked-prefill

usedunclearin Qwen3.6-35B-A3BQwen

we retain the chunked-prefilling mechanism as implemented in Qwen3-Omni and Qwen2.5-Omni, whose audio and vision encoders are capable of outputting chunks along the temporal dimension.

usedinference servingin Qwen3.5-OmniQwen

--chunked-prefill-size 16384

usedinference servingin MiMo-V2.5Xiaomi

--chunked-prefill-size 16384

optionalinference servingin MiMo-V2.5Xiaomi

Filed alongside

Other methods under inference & serving :: KV cache management.