 mavlyutov
|
c4e2953141
Update training recipees for asr and M4T large; Fix GPU ooms with M4T large training
|
2 years ago |
|
Ruslan Mavlyutov
|
9ac6068700
Merge pull request #45 from fairinternal/ruslan_m4t_training
|
2 years ago |
 Kaushik Ram Sadagopan
|
f45bd3a4b9
Switching out seamlessM4T_v2_large to the multitask model with the text encoder. (#55)
|
2 years ago |
|
mavlyutov
|
d8c45cc9e7
Adjust for latest changes in fairseq2n
|
2 years ago |
|
mavlyutov
|
efe88afa2e
M4T training scripts and recipes
|
2 years ago |
|
Kaushik Ram Sadagopan
|
d8521ec060
Padding mask bug fix, it is boolean and not float, UnitY generator. (#54)
|
2 years ago |
 Can Balioglu
|
1bb19a6aa1
Model dim (#53)
|
2 years ago |
|
Can Balioglu
|
d314c9c4da
Statickv (#52)
|
2 years ago |
 Pierre Andrews
|
8a1808944c
Merge pull request #51 from fairinternal/commercial_vocoder_asset
|
2 years ago |
 Kaushik Ram Sadagopan
|
2f9eb12a13
Add new asset for the vocoder for commercial license.
|
2 years ago |
|
Kaushik Ram Sadagopan
|
8db669e967
Add support for `seamlessM4T_v2_large`. (#48)
|
2 years ago |
|
Kaushik Ram Sadagopan
|
cb9096fed3
Adds asset for `m4t_v2_x2t` model and makes translator compatible with x2t models, . (#47)
|
2 years ago |
|
Can Balioglu
|
71d26b65ec
Start using the new PaddingMask type (#49)
|
2 years ago |
|
Can Balioglu
|
ae29a800c8
Rename layer_norm_fn to layer_norm_factory (#46)
|
2 years ago |
|
Can Balioglu
|
9f142ac30e
Use embed.init_fn instead of scaled=True (#44)
|
2 years ago |
|
Kaushik Ram Sadagopan
|
c7f576a749
Introduce `wav2vec2_chunk` offline streaming-compatible speech encoder. (#42)
|
2 years ago |
|
Kaushik Ram Sadagopan
|
6bac442c00
Merge pull request #41 from fairinternal/static_kv
|
2 years ago |
|
Can Balioglu
|
d59b86aca9
Use static_kv for encoder-decoder attention
|
2 years ago |
|
Kaushik Ram Sadagopan
|
fbeabde759
Merge pull request #40 from fairinternal/embedding
|
2 years ago |
|
Can Balioglu
|
4034968f63
Rename Embedding to StandardEmbedding
|
2 years ago |
|
Kaushik Ram Sadagopan
|
e42276c223
Merge pull request #37 from fairinternal/revert_interleave_change
|
2 years ago |
|
Kaushik Ram Sadagopan
|
622a0b2f40
Revert back interleave change since this isn't a single dimension repeat.
|
2 years ago |
|
Can Balioglu
|
785cf2add8
Fixes and perf improvements from latest fairseq2 HEAD (#36)
|
2 years ago |
|
Kaushik Ram Sadagopan
|
4d640ab465
Merge pull request #28 from fairinternal/nar_t2u_offline
|
2 years ago |
|
Kaushik Ram Sadagopan
|
85c1b894cd
Address comments.
|
2 years ago |
|
Kaushik Ram Sadagopan
|
a55c74eb41
Remove unnecessary apply_padding_mask operations.
|
2 years ago |
|
Kaushik Ram Sadagopan
|
afbcd665fb
Fix Conv1dBlock to use padding_mask.
|
2 years ago |
|
Kaushik Ram Sadagopan
|
65851ae4cc
Apply padding_mask in VariancePredictor, enforce TagManager parity with fairseq.
|
2 years ago |
|
Kaushik Ram Sadagopan
|
157a1628f0
Apply padding_mask to output unit sequences.
|
2 years ago |
|
Kaushik Ram Sadagopan
|
8fac405bcb
Made unity_nar_multilingual compatible with Translator.
|
2 years ago |
