tsukumi
ebd249bca8
Fix: spm.model is missing
2024-12-19 13:16:25 +09:00
tsukumi
2833fb5eeb
Fix: The conversion script for English BERT to Fast Tokenizer was incorrect, so tokenization was not performed correctly
2024-12-19 11:10:33 +09:00
tsukumi
c1fce3fec7
Improve: Make it possible to convert BERT language models to FP16
2024-12-19 07:16:59 +09:00
tsukumi
e642a4cb69
Merge branch 'litagin02:master' into master
2024-12-08 03:37:27 +09:00
litagin02
065a7ffa0a
Merge pull request #177 from aka7774/master
...
Add /g2p (Update server_fastapi.py)
2024-11-17 12:35:02 +09:00
tsukumi
4a92a1abad
Fix: Missing onnxsim dependency
2024-11-13 10:50:44 +09:00
tsukumi
f135a79316
Fix: Missing onnx dependency
2024-11-13 10:43:58 +09:00
litagin02
875186f0c7
Change faster-whisper as default
2024-11-11 10:25:30 +09:00
litagin02
038d07f4e7
Bump ver only
2024-11-11 10:18:55 +09:00
litagin02
2cb5b24244
Merge pull request #165 from tsukumijima/master
...
ONNX への変換と ONNXRuntime による推論サポートを追加
2024-11-11 09:44:32 +09:00
Aka Diffusion
3155e2acc1
Update server_fastapi.py
...
g2pの内容をjsonで取れるAPI
2024-11-10 15:27:19 +09:00
tsukumi
689217e350
Refactor: TTSModel.convert_to_16_bit_wav() to @staticmethod
2024-11-10 07:11:53 +09:00
tsukumi
20b0b0908b
Fix: Ensure memory is cleared on unload
2024-11-10 07:06:20 +09:00
tsukumi
9723f75f99
Fix: g2p process fails in PyTorch-independent environment
2024-11-08 02:56:05 +09:00
tsukumi
b0b12696f3
Add: Type Hint to null_model_params
2024-10-25 22:37:09 +09:00
tsukumi
375eb7f3a9
Refactor: run "hatch run style:fmt"
2024-10-25 22:18:57 +09:00
tsukumi
d762bf4401
Merge branch 'dev' of github.com:litagin02/Style-Bert-VITS2
2024-10-25 22:16:19 +09:00
tsukumi
d397a58874
Fix: errors during ONNX CPU inference
2024-10-06 16:18:21 +09:00
tsukumi
c4b6c169fc
Fix: Inference failure with some ExecutionProviders
...
length_scale, etc. were passed as one-dimensional arrays when they should have been passed as scalar values.
2024-09-28 17:36:26 +09:00
tsukumi
7793e9da43
Remove: Unused import
2024-09-28 13:50:10 +09:00
tsukumi
d935d30e4c
Add: test code for CoreMLExecutionProvider on a trial basis
...
Currently ONNXRuntime's CoreML support is not very good, and on my environment, I get errors like “coreml_execution_provider.cc:192 operator() Input (/sdp/flows.3/GatherND_2_output_0) has a dynamic shape ({-1,-1}) but the runtime shape ({0,10}) has zero elements. This is not supported by the CoreML EP.” and inference fails.
I hope this will be fixed in a future version of ONNXRuntime, and leave the test code as it is.
2024-09-23 21:07:27 +09:00
tsukumi
b67e50fd8d
Improve: DirectML inference performance
2024-09-23 20:20:46 +09:00
tsukumi
08af692835
Fix: PyTorch inference test
2024-09-23 15:20:52 +09:00
tsukumi
e91b69862c
Fix: ONNX inference with DirectML fails
2024-09-23 15:06:19 +09:00
tsukumi
758696c59c
Fix: Device ID specified in ExecutionProvider options is respected when moving input tensors
2024-09-23 12:57:14 +09:00
tsukumi
c792bf9a0d
Improve: Make ONNX inference tests more rigorous
2024-09-23 12:27:29 +09:00
tsukumi
2cdb5844cb
Fix: ONNX inference fails in Windows environments
...
ref: https://qiita.com/gomasa/items/f6337087988685d00ebb
2024-09-23 11:55:50 +09:00
tsukumi
53777e315e
Fix: Log in real time during pytest execution
2024-09-23 11:39:10 +09:00
tsukumi
5ca6e184d7
Fix: ONNX inference of BERT models fails in some Windows environments
2024-09-23 11:33:17 +09:00
tsukumi
f9607ad891
Fix: Omitted consideration of macOS support
2024-09-23 11:10:41 +09:00
tsukumi
f21f6435ad
Improve: Add test code for ONNX inference with DmlExecutionProvider
2024-09-23 10:54:29 +09:00
tsukumi
eb5d245903
Add: Test for ONNX inference code (without PyTorch dependency)
2024-09-23 10:04:02 +09:00
tsukumi
5a82105df3
Improve: Since the use cases for speech synthesis in English and Chinese are limited, only Japanese BERT models/tokenizers will be pre-loaded to save VRAM
2024-09-23 08:58:57 +09:00
tsukumi
faddb0a8a8
Fix: Don't list hidden files as models
...
Copying files from macOS to Linux would produce files starting with . _ that were incorrectly identified as model files.
2024-09-23 08:51:31 +09:00
tsukumi
245bfe54ae
Fix: In the “Merge” tab of the Web UI, re-create the TTSModelHolder using the passed TTSModelHolder instance variable so that ONNX models do not get mixed up
2024-09-23 08:13:16 +09:00
tsukumi
b7cd512978
Fix: During ONNX inference, the tensor of the return value of the tokenizer is now obtained in Numpy's NDArray format to eliminate the dependency on PyTorch
2024-09-23 07:40:17 +09:00
tsukumi
dcca37b9a8
Add: Support for speech synthesis in English and Chinese for ONNX inference in Non-JP-Extra models
2024-09-23 07:14:44 +09:00
tsukumi
fe253611d3
Fix: nltk to 3.8.1 or lower
...
Due to reports that updating nltk to 3.8.2 or later significantly slows down learning, including English.
2024-09-23 06:22:47 +09:00
tsukumi
db0acd7b5f
Add: Support for ONNX inference and ONNX conversion of Non-JP-Extra models
2024-09-23 05:59:12 +09:00
tsukumi
cc1d6120fe
Fix: Default value of False for compatibility with older models where the HyperParameters.data.use_jp_extra field does not exist
...
JP-Extra models always have use_jp_extra=True, but non-JP-Extra models do not always.
2024-09-23 05:26:19 +09:00
tsukumi
0e0b4bea31
Fix: Disable optimization on InferenceSession initialization to speed up loading of ONNX models
2024-09-22 23:08:00 +09:00
tsukumi
16115a366b
Improve: .safetensors models under a directory specified with --model can be automatically converted to ONNX
2024-09-22 16:37:15 +09:00
tsukumi
291e635e9e
Improve: ONNX conversion logic improved, and support for noise_scale(_w) model argument
2024-09-22 16:14:25 +09:00
tsukumi
9b27ba9777
Improve: Graphical display of ONNX conversion script logs
2024-09-18 06:11:10 +09:00
tsukumi
449927616a
Fix: Suppress many warning logs
2024-09-18 05:48:37 +09:00
tsukumi
fa7123c422
Fix: ONNX BERT model/tokenizer is not preloaded by default to avoid wasting VRAM
2024-09-18 05:11:15 +09:00
tsukumi
327e038977
Improve: I/O binding during ONNX inference
...
I honestly don't see a big difference...
2024-09-18 04:49:56 +09:00
tsukumi
7d721fdf2b
Fix: CUDAExecutionProvider was not being used to infer Style-Bert-VITS2 ONNX models even though CUDA was available
2024-09-18 04:37:23 +09:00
tsukumi
e5a05a30cb
Improve: Outputs model load time and inference time
2024-09-17 18:37:00 +09:00
tsukumi
b95420325b
Improve: Changed PyTorch to an optional dependency when used style_bert_vits2 as a library
...
ONNXRuntime has relied on CUDA 12.x / cuDNN 9.x by default since v1.19.0, while PyTorch finally supports cuDNN 9 since v2.4.0.
Therefore, unless the pinning of torch<2.4 is eliminated, CUDA cannot be used with the latest version of ONNXRuntime.
2024-09-17 17:56:59 +09:00
