Pascal 0725d2e53b codec: adaptive chunk width on the streaming decode ...

The stream graphs come in width classes T in {1, 2, 4, 8} sharing the
state and KV ring tensors, built lazily per class. The decoder ramps
1 -> 2 -> 4 -> 8 so the first frame keeps its latency while the steady
state interleaves the codec 8x less often and the batch amortizes the
kernel count; drain flushes the tail at EOS and the reference priming
runs in max width chunks.

2026-07-06 22:43:52 +02:00

..

audio-io.h

server: cloned voice registry over the OpenAI surface

2026-07-05 12:28:55 +02:00

audio-mel.h

nits

2026-05-14 16:47:45 +02:00

audio-resample.h

Initial release

2026-05-10 15:57:15 +02:00

backend.h

align symbol naming on omnivoice convention

2026-05-14 18:09:07 +02:00

bpe.h

nits

2026-05-14 16:47:45 +02:00

causal-trans-conv.h

codec: stateful frame by frame streaming decode on a static graph

2026-07-05 10:25:21 +02:00

code-predictor-forward.h

predictor: static graphs replayed with a single id upload per step

2026-07-06 21:46:08 +02:00

code-predictor-graph.h

predictor: static graphs replayed with a single id upload per step

2026-07-06 21:46:08 +02:00

code-predictor-weights.h

cleaning

2026-06-21 17:30:21 +02:00

codec-chunked-decode.h

codec: adaptive chunk width on the streaming decode

2026-07-06 22:43:52 +02:00

convnext-block.h

codec: stateful frame by frame streaming decode on a static graph

2026-07-05 10:25:21 +02:00

dac-decoder-v2.h

codec: stateful frame by frame streaming decode on a static graph

2026-07-05 10:25:21 +02:00

debug.h

cleanup: remove dead code and DRY pre_conv release via wctx_free

2026-05-14 23:27:09 +02:00

encoder-downsample.h

align symbol naming on omnivoice convention

2026-05-14 18:09:07 +02:00

encoder-transformer.h

align symbol naming on omnivoice convention

2026-05-14 18:09:07 +02:00

gguf-weights.h

cleanup: remove dead code and DRY pre_conv release via wctx_free

2026-05-14 23:27:09 +02:00

graph-arena.h

tts: persistent graph arenas and padded attention windows

2026-07-04 14:14:59 +02:00

kv-cache.h

nits

2026-05-14 16:47:45 +02:00

philox.h

cleanup: remove dead code and DRY pre_conv release via wctx_free

2026-05-14 23:27:09 +02:00

pipeline-codec.cpp

codec: adaptive chunk width on the streaming decode

2026-07-06 22:43:52 +02:00

pipeline-codec.h

codec: adaptive chunk width on the streaming decode

2026-07-06 22:43:52 +02:00

pipeline-tts.cpp

codec: adaptive chunk width on the streaming decode

2026-07-06 22:43:52 +02:00

pipeline-tts.h

talker: static decode graphs per attention window class

2026-07-06 22:26:19 +02:00

prompt-builder.h

graph: reset scheduler before remaining shared-sched graph allocs

2026-06-09 19:53:56 +02:00

qt-error.h

align symbol naming on omnivoice convention

2026-05-14 18:09:07 +02:00

quantizer-decode.h

codec: stateful frame by frame streaming decode on a static graph

2026-07-05 10:25:21 +02:00

quantizer-encode.h

align symbol naming on omnivoice convention

2026-05-14 18:09:07 +02:00

qwen.cpp

qwen: lazy speaker encoder load in the voice ref extraction path

2026-07-05 13:45:38 +02:00

qwen.h

Add voice reference extraction ABI

2026-06-24 17:18:32 +02:00

rvq-file.h

server: cloned voice registry over the OpenAI surface

2026-07-05 12:28:55 +02:00

sampling.h

nits

2026-05-14 16:47:45 +02:00

seanet-encoder.h

align symbol naming on omnivoice convention

2026-05-14 18:09:07 +02:00

speaker-encoder-extract.h

nits

2026-05-14 16:47:45 +02:00

speaker-encoder-forward.h

nits

2026-05-14 16:47:45 +02:00

speaker-encoder-weights.h

nits

2026-05-14 16:47:45 +02:00

talker-decode-graph.h

talker: static decode graphs per attention window class

2026-07-06 22:26:19 +02:00

talker-forward.h

talker: static decode graphs per attention window class

2026-07-06 22:26:19 +02:00

talker-weights.h

cleaning

2026-06-21 17:30:21 +02:00

timer.h

perf: add per-stage timer instrumentation to qwentts and omnivoice

2026-05-31 15:05:48 +02:00

tokenizer-transformer.h

codec: stateful frame by frame streaming decode on a static graph

2026-07-05 10:25:21 +02:00

tts-server.h

server: sampling overrides on the speech endpoint

2026-07-05 15:01:37 +02:00

utf8.h

utf8: normalize inbound text bytes to UTF-8

2026-07-04 01:14:00 +02:00

wav.h

Initial release

2026-05-10 15:57:15 +02:00

weight-ctx.h

Initial release

2026-05-10 15:57:15 +02:00