Add Apertus 1.5 and WavTokenizer #47662
RaphaelKreft
wants to merge
36 commits into
huggingface:main
from
swiss-ai:add-apertus1p5
integration of WavTokenizer
c29d76ec
add parity check script for development
b385b2e0
Document float32 requirement for reproducible audio codes
d668c4b1
Fix WavTokenizer padding masks and docs
ead51303
Fix WavTokenizer single-code decoding
1e93e8f0
Support all WavTokenizer checkpoints
ce485c2b
Allow WavTokenizer padding with truncation
df0b72c5
update docs with uploaded checkpoints
cbfad1fd
Initial Integration of Inference Only Emu3.5 Image Tokenizer as part …
ec863ae3
Add Emu3.5 Image Tokenizer parity test
c162d4b7
fix(generation): support pruned output vocabularies
af822c31
feat(apertus1p5): finalize composite model and text backbone
27a7b3af
feat(apertus1p5): add unified image and audio processing
5980d1cd
feat(apertus1p5): add composite checkpoint conversion
4b79ce59
test(apertus1p5): add parity and validation utilities
60e27457
test(apertus1p5): add hands-on integration scripts
badfd2cf
docs(apertus1p5): document multimodal usage
3700416a
docs: update test cases and setup readme
6c1dacd7
refactor(apertus1p5): keep pruned logits model-local
aaeea2b5
feat(apertus1p5): hardening inference with pruned lm-head by padding …
3f207c0b
fix issues after upstream rebase
745e3625
update docs and set correct checkpoint-hf path
edd8b26f
updated testcases and add train loop and multidevice placement
dcc724b8
dev(apertus1p5): improve and restructure ap_testcase test scripts
b1692a16
docs(apertus1p5): improve documentation for apertus
6a2194f5
docs(wavtokenizer): improve documentation for wavtokenizer
180dcc4f
fix(apertus1p5): validate negative labels, harden converter, allowlis…
a2215926
docs(apertus1p5,wavtokenizer): tighten docstrings and document prepro…
ae03ebf9
chore(apertus1p5): regenerate modeling after rebase onto upstream main
386fdb69
fix(tests): repair wavtokenizer GPU batch tests and skip unsupported …
eb40b427
docs(apertus1p5,wavtokenizer): drop references to files stripped from…
683a94b5
feat(apertus1p5): add a standalone EMU3.5 vision tokenizer converter
4f853358
fix(apertus1p5): verify stored precision and reject unconverted token…
bbe11dd2
refactor(apertus1p5): reuse the shipped converter in the vision parit…
ea00e601
docs(apertus1p5,wavtokenizer): document the checkpoint conversion path
d6750e34
chore: drop development harnesses for upstream submission
a988895b
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub