Add Music Flamingo #43538
ebezzam
merged
92 commits into
huggingface:main
from
lashahub:modular-mf
Music flamingo
caf33f1a
Fix pos embeddings
b69a9d1f
Merge branch 'huggingface:main' into main
b4acc17c
Method arg docstrings
cf1e9bc8
Add tests & docs
44e801be
Merge branch 'huggingface:main' into main
f0956e30
Fix AF3 dtype bug
c9734457
Fix the MF performance issue
e3a17fb5
Fix pos embeddings
627dee87
Merge branch 'main' of https://github.com/lashahub/transformers
e9df30da
Fix embeddings & format
4c48132d
Remove external deps
d67c1143
Update processor token names
aedd3411
Cleanup
87d55a9c
Simplify RotaryEmbedding to lang-only
be227468
Reuse AF3 config classes
e5b4677e
Trim+rename rotary embedding
d7e0bcb2
Call parent _init_weights first and drop rotary einsum
74af4fa9
Precompute rotary cache at init
d622368e
Use modular processor pattern for MusicFlamingo
cab39379
Remove audio-only inference example
9dd94a0e
Refactor Audio Feature Casting Path
767a1d54
Clarify private source repo
91196605
Clean up modular
fc8ab3aa
Move config to modular
5159745d
Formatting
4aae4ff3
Remove dummy
ffef7dbe
Derive musicflamingo timing and rotary config
90148ed4
Llama style rotary embeddings
87cb03fb
Added reproducer comments
5abac953
Expose _init_weights for modular.
f4a1b681
Satisfy repo checks
3b9290f7
Align MusicFlamingo rotary with Llama style
82ab1ef2
Move MusicFlamingo _init_weights to encoder
e993dea0
Keep old behavior
2d88557f
Move MusicFlamingo rotary settings into encoder rope_parameters
27024f15
Use AutoConfig in AF3/MF
10797b97
Align MusicFlamingo RoTE with Llama RoPE conventions
f49014f8
Update outdated fixtures
f465d3d2
init_weights without changing others
4224af51
FIx import
2cbf219e
ebezzam
commented
on 2026-02-27
Remove backward compat
a25c8a6b
Merge branch 'main' into modular-mf
3ed73a80
Regenerate modeling for MF
f2e7d3f0
Fix AF3 batch inference bug
445cd245
Simplify config and nit.
fc25c48a
Conform more to transformers convention, e.g. removing unused code pa…
664e7b12
Add another possible AF3 prefix.
bbca79a4
Merge branch 'main' into modular-mf
69c9bee2
Use auto_docstring and update docstrings.
17846d93
ebezzam
commented
on 2026-03-11
Nits
24fbaa71
Nit for review
e25d377b
Shift RoTE to main model so that encoder can be directly used from AF3.
c99fc0ad
ebezzam
commented
on 2026-03-11
Refactoring nit.
9bb0c53d
Fix init
e9528a6b
Fix some failing tests
13af64d4
Fix AF3 & MF and add batching tests
a34e981a
Fix audio embedding masking (bad post length)
702eb272
Nits and remove since same as GLM was bug in post length computation
a32520ce
ebezzam
commented
on 2026-03-17
Simplify MF as AF3, and style checks.
7750bb5a
Merge branch 'main' into modular-mf
74417f6c
New config after merge and modular update.
54bfeccf
Merge branch 'main' into modular-mf
2f2b552e
Address music flamingo tests, and some cleanup.
612840f6
style check
2c3e5839
Regenerate config.
a20e163b
Update fixtures.
fdb849a2
Nits
057ad886
Merge branch 'modular-mf' of github.com:lashahub/transformers into mo…
0442a67a
Nit
7f04d2d6
Merge branch 'huggingface:main' into modular-mf
66ce2012
Improve RoTE config
7e06d278
Refine MusicFlamingo rotary time handling
d423510c
Simplification, and update AF3 processor for better modular
e79475b7
Fix torch export
6a1add49
Simplify modular, including upstreaming input_ids input to get_audio_…
e05b767f
eustlb
commented
on 2026-03-24
Remove upstreaming of input_ids to get_audio_features, and remove aud…
edcffefa
Switch to MoonshineRotaryEmbedding, and cleanup.
e986996b
ebezzam
commented
on 2026-03-24
eustlb
approved these changes
on 2026-03-24
Remove hardcoded MusicFlamingo partial_rotary_factor
ad76e285
Update fixtures
8e0dc3b2
Compile re.sub
a46aa1c9
Merge branch 'modular-mf' of github.com:lashahub/transformers into mo…
836da949
Update src/transformers/models/musicflamingo/modular_musicflamingo.py
10894776
Update src/transformers/models/musicflamingo/modular_musicflamingo.py
2ea55748
Style
999737e5
Merge branch 'main' into modular-mf
1533bec8
Update fixtures.
6857af95
Merge branch 'main' into modular-mf
cd693fc5
Conditional torch import for processor.
35fc31fd
Merge branch 'modular-mf' of github.com:lashahub/transformers into mo…
a62e7d9e
Merge branch 'main' into modular-mf
d0d18f2b
Merge branch 'main' into modular-mf
47fa9381
Reviewers
Assignees
Labels
New model
Audio
Milestone
No milestone
Login to write a write a comment.
Login via GitHub