Code

tts : add OuteTTS support#10784

Code

Merged

ggerganov merged 45 commits intomasterggml-org/llama.cpp:masterfrom gg/tts-add-outettsggml-org/llama.cpp:gg/tts-add-outettsCopy head branch name to clipboardDec 18, 2024

+2,507-530

Commits

Commits on Dec 18, 2024

• server : add "tokens" output

  Show description for 89eaf50
  ggerganov

  committed
  89eaf50

  Copy full SHA for 89eaf50

• server : output embeddings for all tokens when pooling = none

  Show description for 06e8540
  ggerganov

  committed
  06e8540

  Copy full SHA for 06e8540

• server : be explicit about the pooling type in the tests

  Show description for 1b18b2d
  ggerganov

  committed
  1b18b2d

  Copy full SHA for 1b18b2d

• server : do not normalize embeddings when there is no pooling

  Show description for e65556f
  ggerganov

  committed
  e65556f

  Copy full SHA for e65556f

• llama : add OuteTTS support (wip)

  ggerganov

  committed
  f169965

  Copy full SHA for f169965

• wip

  ggerganov

  committed
  ff2ea75

  Copy full SHA for ff2ea75

• extract features

  ggerganov

  committed
  aac7e04

  Copy full SHA for aac7e04

• first conv

  ggerganov

  committed
  6ef1409

  Copy full SHA for 6ef1409

• group norm

  ggerganov

  committed
  5296c96

  Copy full SHA for 5296c96

• resnet conv

  ggerganov

  committed
  3d08d62

  Copy full SHA for 3d08d62

• resnet

  ggerganov

  committed
  13dd894

  Copy full SHA for 13dd894

• attn

  ggerganov

  committed
  3046fde

  Copy full SHA for 3046fde

• pos net

  ggerganov

  committed
  435cfd7

  Copy full SHA for 435cfd7

• layer norm

  ggerganov

  committed
  b3ba05e

  Copy full SHA for b3ba05e

• convnext

  ggerganov

  committed
  fe6dd5a

  Copy full SHA for fe6dd5a

• head

  ggerganov

  committed
  839035d

  Copy full SHA for 839035d

• hann window

  ggerganov

  committed
  eb1b70f

  Copy full SHA for eb1b70f

• fix n_embd + remove llama.cpp hacks

  ggerganov

  committed
  a1f08ad

  Copy full SHA for a1f08ad

• compute hann window

  ggerganov

  committed
  e728cfd

  Copy full SHA for e728cfd

• fft

  ggerganov

  committed
  5a1c98e

  Copy full SHA for 5a1c98e

• spectrum processing

  ggerganov

  committed
  e527971

  Copy full SHA for e527971

• clean-up

  ggerganov

  committed
  191da33

  Copy full SHA for 191da33

• tts : receive input text and generate codes

  ggerganov

  committed
  b9a011e

  Copy full SHA for b9a011e

• clip : fix new conv name

  ggerganov

  committed
  db61391

  Copy full SHA for db61391

• tts : minor fix

  ggerganov

  committed
  8329e85

  Copy full SHA for 8329e85

• tts : add header + minor fixes

  Show description for d4fa34b
  ggerganov

  committed
  d4fa34b

  Copy full SHA for d4fa34b

• tts : add matchematical constant

  Show description for 2221e54
  ggerganov

  committed
  2221e54

  Copy full SHA for 2221e54

• tts : fix sampling + cut initial noise

  ggerganov

  committed
  906a0ed

  Copy full SHA for 906a0ed

• tts : fixes

  ggerganov

  committed
  1d7c27c

  Copy full SHA for 1d7c27c

• tts : update default samplers

  Show description for 3d54be4
  ggerganov

  committed
  3d54be4

  Copy full SHA for 3d54be4

• tts : text pre-processing

  ggerganov

  committed
  befdcd2

  Copy full SHA for befdcd2

• tts : outetts-voc -> wavtokenizer-dec

  ggerganov

  committed
  e70f140

  Copy full SHA for e70f140

• tts : remove hardcoded constants

  Show description for c096bbd
  ggerganov

  committed
  c096bbd

  Copy full SHA for c096bbd

• tts : fix tensor shapes

  ggerganov

  committed
  d1ef627

  Copy full SHA for d1ef627

• llama : refactor wavtokenizer tensors

  Show description for 980d631
  ggerganov

  committed
  980d631

  Copy full SHA for 980d631

• cont

  Show description for 35259e5
  ggerganov

  committed
  35259e5

  Copy full SHA for 35259e5

• cont [no ci]

  ggerganov

  committed
  2033fb7

  Copy full SHA for 2033fb7

• llama : update WavTokenizer to non-causal attn

  ggerganov

  committed
  824fa75

  Copy full SHA for 824fa75

• llama : handle no-vocab detokenization

  ggerganov

  committed
  d291c74

  Copy full SHA for d291c74

• tts : add Python example for OuteTTS (wip)

  ggerganov

  committed
  5038abe

  Copy full SHA for 5038abe

• tts : extend python example to generate spectrogram

  Show description for edb7896
  ggerganov

  committed
  edb7896

  Copy full SHA for edb7896

• server : fix rebase artifacts

  ggerganov

  committed
  2a1a6f6

  Copy full SHA for 2a1a6f6

• tts : enable "return_tokens" in Python example

  Show description for 29df666
  ggerganov

  committed
  29df666

  Copy full SHA for 29df666

• tts : minor fixes

  ggerganov

  committed
  a95191c

  Copy full SHA for a95191c

• common : support HF download for vocoder

  ggerganov

  committed
  c0df192

  Copy full SHA for c0df192