Moshi integration #33624
ylacombe
merged
88 commits into
huggingface:main
from
ylacombe:moshi-integration
clean mimi commit
c0862316
some nits suggestions from Arthur
a544d27f
make fixup
502865ff
first moshi WIP
c8583216
converting weights working + configuration + generation configuration
2eaadcaf
finalize converting script - still missing tokenizer and FE and proce…
016d5386
fix saving model w/o default config
34b6e243
working generation
50f9eb80
Merge branch 'main' into moshi-integration
81432c0b
ylacombe
commented
on 2024-09-20
ylacombe
commented
on 2024-09-20
gante
commented
on 2024-09-20
Merge branch 'huggingface:main' into moshi-integration
1e051943
use GenerationMixin instead of inheriting
199e1334
add delay pattern mask
c6fe08eb
fix right order: moshi codes then user codes
b0fdc482
unconditional inputs + generation config
5553ade7
get rid of MoshiGenerationConfig
56818541
blank user inputs
852f1e1f
update convert script:fix conversion, add tokenizer, feature extract…
d1fe2fe3
add and correct Auto classes
76d6e471
update modeling code, configuration and tests
d1fcc4c8
make fixup
123b1f2d
fix some copies
25e88c29
WIP: add integration tests
1db75ac1
add dummy objects
8fe84b7b
Merge branch 'main' into moshi-integration
c2b1e60a
propose better readiblity and code organisation
24994199
update tokenization tests
58d72243
update docstrigns, eval and modeling
faca73b8
add .md
420040e8
make fixup
22f54ee0
add MoshiForConditionalGeneration to ignore Auto
1c060d90
Merge branch 'huggingface:main' into moshi-integration
9e5889d4
revert mimi changes
b16e6e1a
re
1d70d989
further fix
3eff2113
Update moshi.md
298c4a3d
correct md formating
219a9761
move prepare causal mask to class
fa3e6a98
fix copies
9e650fde
Merge branch 'huggingface:main' into moshi-integration
0a147ffc
fix depth decoder causal
67370733
fix and correct some tests
2e159617
make style and update .md
9f89deea
correct config checkpoitn
e2bc51f4
Update tests/models/moshi/test_tokenization_moshi.py
2aa26836
Update tests/models/moshi/test_tokenization_moshi.py
3fc6496a
Merge branch 'huggingface:main' into moshi-integration
b2ec0c3a
make style
3a1ed423
Merge branch 'huggingface:main' into moshi-integration
898e150b
Update src/transformers/models/moshi/__init__.py
d1975868
fixup
96bae02f
change firm in copyrights
a2bbba10
udpate config with nested dict
590e5c1f
replace einsum
c70c8901
make style
3b490d2f
Merge branch 'huggingface:main' into moshi-integration
335a6c61
Merge branch 'huggingface:main' into moshi-integration
963818d6
change split to True
85bcf7a4
add back splt=False
e8659387
remove tests in convert
2f856f50
Update tests/models/moshi/test_modeling_moshi.py
9711b004
add default config repo + add model to FA2 docstrings
22968d25
remove logits float
786b1767
fix some tokenization tests and ignore some others
29e121f8
make style tokenization tests
a8d7e510
update modeling with sliding window + update modeling tests
3241a34e
Merge branch 'huggingface:main' into moshi-integration
473cafc7
[run-slow] moshi
137985d0
Merge branch 'huggingface:main' into moshi-integration
9288bc75
Merge branch 'huggingface:main' into moshi-integration
1a6c1c20
Merge branch 'huggingface:main' into moshi-integration
52c599c8
remove prepare for generation frol CausalLM
6e9b4bd9
isort
d25a5a7f
remove copied from
9e8fd4ed
ignore offload tests
ab8f356b
Merge branch 'huggingface:main' into moshi-integration
9650b2bd
update causal mask and prepare 4D mask aligned with recent changes
7b1ca203
further test refine + add back prepare_inputs_for_generation for dept…
896ee4a5
correct conditional use of prepare mask
a9c3d6f4
Merge branch 'huggingface:main' into moshi-integration
38b5fee8
update slow integration tests
5784642c
fix multi-device forward
4f8f877b
Merge branch 'huggingface:main' into moshi-integration
1f912c7d
remove previous solution to device_map
a2d55190
save_load is flaky
bf4375fb
Merge remote-tracking branch 'upstream/main' into moshi-integration
dda8da25
fix generate multi-devices
f06e230c
fix device
a54d218f
ArthurZucker
approved these changes
on 2024-10-16
move tensor to int
f46e0d84
Login to write a write a comment.
Login via GitHub