Repository navigation
Releases: PNL-toshiyaishihara/nemo-speech-python
Release list
nemo-speech-python v0.3.0rc1
Unofficial Python bindings for NeMo-Speech.cpp, v0.3.0rc1.
Changes
What's Changed
Fixes
- Allocate the ggml scheduler's split buffers on demand instead of 1.3 GB per session by @PNL-toshiyaishihara in #65
- Bound the RNNT greedy decoder's joint window with a NeMo-Speech.cpp patch by @PNL-toshiyaishihara in #67
Documentation
- Allow small, result-preserving patches to NeMo-Speech.cpp sources by @PNL-toshiyaishihara in #66
Full Changelog: v0.2.0...v0.3.0rc1
Vendored sources
| Source | Revision | Commit |
|---|---|---|
| NeMo-Speech.cpp | v0.2.0 | 6a3ca3693707 |
| llama.cpp | bd4f514 | bd4f514db14d |
| SentencePiece | v0.2.0 | 17d7580d6407 |
NeMo-Speech.cpp is unchanged since v0.2.0.
Wheels
Python: 3.10, 3.11, 3.12, 3.13, 3.14 (Requires-Python >=3.10). Every wheel is py3-none, so one file serves all of them.
The x86-64 wheels (Linux x86_64 and Windows x64, every variant) carry ggml's CPU backend once per CPU generation and load the best one for the CPU at run time, from plain x86-64 to AVX-VNNI and AVX-512.
| Variant | Platform | Requirements | File |
|---|---|---|---|
| CPU | Linux aarch64 (glibc 2.28+) | nemo_speech-0.3.0rc1-py3-none-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl |
|
| CPU | Linux x86_64 (glibc 2.28+) | nemo_speech-0.3.0rc1-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl |
|
| CPU | Windows x64 | nemo_speech-0.3.0rc1-py3-none-win_amd64.whl |
|
| CPU + Metal | macOS arm64 (13.0+) | nemo_speech-0.3.0rc1-py3-none-macosx_13_0_arm64.whl |
|
| Vulkan | Windows x64 | GPU driver with Vulkan | nemo_speech-0.3.0rc1+vulkan-py3-none-win_amd64.whl |
| CUDA 12.8 | Linux x86_64 (glibc 2.28+) | NVIDIA driver R570+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 12.0; newer GPUs through PTX (9.0+) | nemo_speech-0.3.0rc1+cu128-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl |
| CUDA 12.8 | Windows x64 | NVIDIA driver R570+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 12.0; newer GPUs through PTX (9.0+) | nemo_speech-0.3.0rc1+cu128-py3-none-win_amd64.whl |
| CUDA 13.0 | Linux aarch64 (glibc 2.28+) | NVIDIA driver R580+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 11.0, 12.0, 12.1; newer GPUs through PTX (9.0+) | nemo_speech-0.3.0rc1+cu130-py3-none-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl |
| CUDA 13.0 | Linux x86_64 (glibc 2.28+) | NVIDIA driver R580+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 12.0; newer GPUs through PTX (9.0+) | nemo_speech-0.3.0rc1+cu130-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl |
| CUDA 13.0 | Windows x64 | NVIDIA driver R580+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 12.0; newer GPUs through PTX (9.0+) | nemo_speech-0.3.0rc1+cu130-py3-none-win_amd64.whl |
Install
All variants share the distribution name nemo-speech, so install the file for your platform and backend directly (a version specifier with -f would let pip pick a GPU variant). For example:
pip install https://github.com/PNL-toshiyaishihara/nemo-speech-python/releases/download/v0.3.0rc1/nemo_speech-0.3.0rc1+cu128-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whlGPU wheels bundle their runtime libraries (CUDA runtime, cuBLAS) and need only the GPU driver. nemo_speech.build_info() reports how an installation was built. SHA256SUMS lists the checksums of all files.
Known issues
Upstream issues worked around in this release: https://github.com/PNL-toshiyaishihara/nemo-speech-python/issues?q=label%3Aupstream
Contributors
Assets 15
nemo-speech-python v0.2.0
Unofficial Python bindings for NeMo-Speech.cpp, v0.2.0.
Changes
What's Changed
Features
- Build the CUDA 13.0 wheel for Linux aarch64 (DGX Spark, Jetson Thor) by @PNL-toshiyaishihara in #54
- Build the x86-64 CPU backend per CPU generation by @PNL-toshiyaishihara in #57
Fixes
- Load ggml backends from libggml's directory, not the working directory by @PNL-toshiyaishihara in #61
Packaging and CI
- Pin actions to commit SHAs and declare workflow permissions by @PNL-toshiyaishihara in #45
- Add Dependabot for GitHub Actions and Python security updates by @PNL-toshiyaishihara in #44
- Verify the CUDA and Vulkan downloads against pinned hashes by @PNL-toshiyaishihara in #50
- Pin the build tools CI installs from PyPI by @PNL-toshiyaishihara in #51
Documentation
- Mark upstream issues that are worked around here by @PNL-toshiyaishihara in #40
- Add a security policy by @PNL-toshiyaishihara in #47
- Add issue forms for bug reports and feature requests by @PNL-toshiyaishihara in #48
- Correct the Vulkan ASR performance explanation in the README by @PNL-toshiyaishihara in #52
Full Changelog: v0.1.0...v0.2.0
Vendored sources
| Source | Revision | Commit |
|---|---|---|
| NeMo-Speech.cpp | v0.2.0 | 6a3ca3693707 |
| llama.cpp | bd4f514 | bd4f514db14d |
| SentencePiece | v0.2.0 | 17d7580d6407 |
NeMo-Speech.cpp is unchanged since v0.1.0.
Wheels
Python: 3.10, 3.11, 3.12, 3.13, 3.14 (Requires-Python >=3.10). Every wheel is py3-none, so one file serves all of them.
The x86-64 wheels (Linux x86_64 and Windows x64, every variant) carry ggml's CPU backend once per CPU generation and load the best one for the CPU at run time, from plain x86-64 to AVX-VNNI and AVX-512.
| Variant | Platform | Requirements | File |
|---|---|---|---|
| CPU | Linux aarch64 (glibc 2.28+) | nemo_speech-0.2.0-py3-none-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl |
|
| CPU | Linux x86_64 (glibc 2.28+) | nemo_speech-0.2.0-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl |
|
| CPU | Windows x64 | nemo_speech-0.2.0-py3-none-win_amd64.whl |
|
| CPU + Metal | macOS arm64 (13.0+) | nemo_speech-0.2.0-py3-none-macosx_13_0_arm64.whl |
|
| Vulkan | Windows x64 | GPU driver with Vulkan | nemo_speech-0.2.0+vulkan-py3-none-win_amd64.whl |
| CUDA 12.8 | Linux x86_64 (glibc 2.28+) | NVIDIA driver R570+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 12.0; newer GPUs through PTX (9.0+) | nemo_speech-0.2.0+cu128-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl |
| CUDA 12.8 | Windows x64 | NVIDIA driver R570+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 12.0; newer GPUs through PTX (9.0+) | nemo_speech-0.2.0+cu128-py3-none-win_amd64.whl |
| CUDA 13.0 | Linux aarch64 (glibc 2.28+) | NVIDIA driver R580+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 11.0, 12.0, 12.1; newer GPUs through PTX (9.0+) | nemo_speech-0.2.0+cu130-py3-none-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl |
| CUDA 13.0 | Linux x86_64 (glibc 2.28+) | NVIDIA driver R580+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 12.0; newer GPUs through PTX (9.0+) | nemo_speech-0.2.0+cu130-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl |
| CUDA 13.0 | Windows x64 | NVIDIA driver R580+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 12.0; newer GPUs through PTX (9.0+) | nemo_speech-0.2.0+cu130-py3-none-win_amd64.whl |
Install
All variants share the distribution name nemo-speech, so install the file for your platform and backend directly (a version specifier with -f would let pip pick a GPU variant). For example:
pip install https://github.com/PNL-toshiyaishihara/nemo-speech-python/releases/download/v0.2.0/nemo_speech-0.2.0+cu128-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whlGPU wheels bundle their runtime libraries (CUDA runtime, cuBLAS) and need only the GPU driver. nemo_speech.build_info() reports how an installation was built. SHA256SUMS lists the checksums of all files.
Known issues
Upstream issues worked around in this release: https://github.com/PNL-toshiyaishihara/nemo-speech-python/issues?q=label%3Aupstream
Contributors
Assets 15
- 742 MB
2026-10-13T10:13:11Z - 711 MB
2026-10-13T10:13:11Z - 724 MB
2026-10-13T10:13:11Z - 577 MB
2026-10-13T10:13:11Z - 554 MB
2026-10-13T10:13:12Z - 35.6 MB
2026-10-13T10:13:36Z - 20.6 MB
2026-10-13T10:13:40Z - 21.1 MB
2026-10-13T10:13:42Z - 27.5 MB
2026-10-13T10:13:44Z - 22.8 MB
2026-10-13T10:13:47Z -
2026-10-13T08:09:23Z -
2026-10-13T08:09:23Z -
2026-10-13T08:09:23Z - Loading
nemo-speech-python v0.1.0
Unofficial Python bindings for NeMo-Speech.cpp, v0.1.0.
Changes
What's Changed
Features
- Add the release pipeline: variants, CUDA 13, Python 3.10-3.14, build info by @PNL-toshiyaishihara in #8
- Ship the Japanese and Mandarin TTS tokenizers with their data by @PNL-toshiyaishihara in #34
Fixes
- TTS: keep the requested sample rate after a cancellation, and close safely during a synthesis by @PNL-toshiyaishihara in #18
- Close Recognizer, Diarizer and Translator safely while other threads use them by @PNL-toshiyaishihara in #19
- Read WAVE_FORMAT_EXTENSIBLE files on Python 3.10 and 3.11 by @PNL-toshiyaishihara in #20
- Make model downloads resumable, retrying and safe to share with the CLI by @PNL-toshiyaishihara in #21
- ASR: honor interim_results=False in streams by @PNL-toshiyaishihara in #22
- Loader: keep wheels on their bundled CUDA libraries, and name missing symbols by @PNL-toshiyaishihara in #23
Vendored dependencies
- Update NeMo-Speech.cpp to v0.2.0 by @PNL-toshiyaishihara in #7
Packaging and CI
- tests: check ctypes types, enum values and callbacks against the headers by @PNL-toshiyaishihara in #27
- Harden the release pipeline: tag, asset and variant checks by @PNL-toshiyaishihara in #24
- Ship the license texts of the bundled compiler runtimes by @PNL-toshiyaishihara in #25
- ci: run cuda-smoke on every PR so it can be a required check by @PNL-toshiyaishihara in #28
- Build CUDA wheels with PTX for newer GPUs and native code for B200 by @PNL-toshiyaishihara in #17
- Release notes: list every change in the first final release by @PNL-toshiyaishihara in #38
Documentation
- docs: add CONTRIBUTING.md and CLAUDE.md by @PNL-toshiyaishihara in #15
- docs: document the x86-64 CPU baseline and library clashes; refresh known issues by @PNL-toshiyaishihara in #26
Full Changelog: https://github.com/PNL-toshiyaishihara/nemo-speech-python/commits/v0.1.0
Vendored sources
| Source | Revision | Commit |
|---|---|---|
| NeMo-Speech.cpp | v0.2.0 | 6a3ca3693707 |
| llama.cpp | bd4f514 | bd4f514db14d |
| SentencePiece | v0.2.0 | 17d7580d6407 |
Wheels
Python: 3.10, 3.11, 3.12, 3.13, 3.14 (Requires-Python >=3.10). Every wheel is py3-none, so one file serves all of them.
The x86-64 wheels (Linux x86_64 and Windows x64, every variant) need a CPU with AVX2, FMA, F16C and BMI2 (Intel Haswell, AMD Zen or newer).
| Variant | Platform | Requirements | File |
|---|---|---|---|
| CPU | Linux aarch64 (glibc 2.28+) | nemo_speech-0.1.0-py3-none-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl |
|
| CPU | Linux x86_64 (glibc 2.28+) | nemo_speech-0.1.0-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl |
|
| CPU | Windows x64 | nemo_speech-0.1.0-py3-none-win_amd64.whl |
|
| CPU + Metal | macOS arm64 (13.0+) | nemo_speech-0.1.0-py3-none-macosx_13_0_arm64.whl |
|
| Vulkan | Windows x64 | GPU driver with Vulkan | nemo_speech-0.1.0+vulkan-py3-none-win_amd64.whl |
| CUDA 12.8 | Linux x86_64 (glibc 2.28+) | NVIDIA driver R570+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 12.0; newer GPUs through PTX (9.0+) | nemo_speech-0.1.0+cu128-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl |
| CUDA 12.8 | Windows x64 | NVIDIA driver R570+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 12.0; newer GPUs through PTX (9.0+) | nemo_speech-0.1.0+cu128-py3-none-win_amd64.whl |
| CUDA 13.0 | Linux x86_64 (glibc 2.28+) | NVIDIA driver R580+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 12.0; newer GPUs through PTX (9.0+) | nemo_speech-0.1.0+cu130-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl |
| CUDA 13.0 | Windows x64 | NVIDIA driver R580+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 12.0; newer GPUs through PTX (9.0+) | nemo_speech-0.1.0+cu130-py3-none-win_amd64.whl |
Install
All variants share the distribution name nemo-speech, so install the file for your platform and backend directly (a version specifier with -f would let pip pick a GPU variant). For example:
pip install https://github.com/PNL-toshiyaishihara/nemo-speech-python/releases/download/v0.1.0/nemo_speech-0.1.0+cu128-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whlGPU wheels bundle their runtime libraries (CUDA runtime, cuBLAS) and need only the GPU driver. nemo_speech.build_info() reports how an installation was built. SHA256SUMS lists the checksums of all files.
Known issues
Upstream issues worked around in this release: https://github.com/PNL-toshiyaishihara/nemo-speech-python/issues?q=label%3Aupstream
Contributors
Assets 14
nemo-speech-python v0.1.0rc2
Unofficial Python bindings for NeMo-Speech.cpp, v0.1.0rc2.
Changes
What's Changed
Features
- Ship the Japanese and Mandarin TTS tokenizers with their data by @PNL-toshiyaishihara in #34
Fixes
- TTS: keep the requested sample rate after a cancellation, and close safely during a synthesis by @PNL-toshiyaishihara in #18
- Close Recognizer, Diarizer and Translator safely while other threads use them by @PNL-toshiyaishihara in #19
- Read WAVE_FORMAT_EXTENSIBLE files on Python 3.10 and 3.11 by @PNL-toshiyaishihara in #20
- Make model downloads resumable, retrying and safe to share with the CLI by @PNL-toshiyaishihara in #21
- ASR: honor interim_results=False in streams by @PNL-toshiyaishihara in #22
- Loader: keep wheels on their bundled CUDA libraries, and name missing symbols by @PNL-toshiyaishihara in #23
Packaging and CI
- tests: check ctypes types, enum values and callbacks against the headers by @PNL-toshiyaishihara in #27
- Harden the release pipeline: tag, asset and variant checks by @PNL-toshiyaishihara in #24
- Ship the license texts of the bundled compiler runtimes by @PNL-toshiyaishihara in #25
- ci: run cuda-smoke on every PR so it can be a required check by @PNL-toshiyaishihara in #28
- Build CUDA wheels with PTX for newer GPUs and native code for B200 by @PNL-toshiyaishihara in #17
Documentation
- docs: document the x86-64 CPU baseline and library clashes; refresh known issues by @PNL-toshiyaishihara in #26
Full Changelog: v0.1.0rc1...v0.1.0rc2
Vendored sources
| Source | Revision | Commit |
|---|---|---|
| NeMo-Speech.cpp | v0.2.0 | 6a3ca3693707 |
| llama.cpp | bd4f514 | bd4f514db14d |
| SentencePiece | v0.2.0 | 17d7580d6407 |
NeMo-Speech.cpp is unchanged since v0.1.0rc1.
Wheels
Python: 3.10, 3.11, 3.12, 3.13, 3.14 (Requires-Python >=3.10). Every wheel is py3-none, so one file serves all of them.
The x86-64 wheels (Linux x86_64 and Windows x64, every variant) need a CPU with AVX2, FMA, F16C and BMI2 (Intel Haswell, AMD Zen or newer).
| Variant | Platform | Requirements | File |
|---|---|---|---|
| CPU | Linux aarch64 (glibc 2.28+) | nemo_speech-0.1.0rc2-py3-none-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl |
|
| CPU | Linux x86_64 (glibc 2.28+) | nemo_speech-0.1.0rc2-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl |
|
| CPU | Windows x64 | nemo_speech-0.1.0rc2-py3-none-win_amd64.whl |
|
| CPU + Metal | macOS arm64 (13.0+) | nemo_speech-0.1.0rc2-py3-none-macosx_13_0_arm64.whl |
|
| Vulkan | Windows x64 | GPU driver with Vulkan | nemo_speech-0.1.0rc2+vulkan-py3-none-win_amd64.whl |
| CUDA 12.8 | Linux x86_64 (glibc 2.28+) | NVIDIA driver R570+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 12.0; newer GPUs through PTX (9.0+) | nemo_speech-0.1.0rc2+cu128-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl |
| CUDA 12.8 | Windows x64 | NVIDIA driver R570+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 12.0; newer GPUs through PTX (9.0+) | nemo_speech-0.1.0rc2+cu128-py3-none-win_amd64.whl |
| CUDA 13.0 | Linux x86_64 (glibc 2.28+) | NVIDIA driver R580+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 12.0; newer GPUs through PTX (9.0+) | nemo_speech-0.1.0rc2+cu130-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl |
| CUDA 13.0 | Windows x64 | NVIDIA driver R580+; compute capability 7.5, 8.0, 8.6, 8.9, 9.0, 10.0, 12.0; newer GPUs through PTX (9.0+) | nemo_speech-0.1.0rc2+cu130-py3-none-win_amd64.whl |
Install
All variants share the distribution name nemo-speech, so install the file for your platform and backend directly (a version specifier with -f would let pip pick a GPU variant). For example:
pip install https://github.com/PNL-toshiyaishihara/nemo-speech-python/releases/download/v0.1.0rc2/nemo_speech-0.1.0rc2+cu128-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whlGPU wheels bundle their runtime libraries (CUDA runtime, cuBLAS) and need only the GPU driver. nemo_speech.build_info() reports how an installation was built. SHA256SUMS lists the checksums of all files.
Known issues
Upstream issues worked around in this release: https://github.com/PNL-toshiyaishihara/nemo-speech-python/issues?q=label%3Aupstream
Contributors
Assets 13
nemo-speech-python v0.1.0rc1
Unofficial Python bindings for NeMo-Speech.cpp, v0.1.0rc1.
Changes
What's Changed
Features
- Add the release pipeline: variants, CUDA 13, Python 3.10-3.14, build info by @PNL-toshiyaishihara in #8
Vendored dependencies
- Update NeMo-Speech.cpp to v0.2.0 by @PNL-toshiyaishihara in #7
Documentation
- docs: add CONTRIBUTING.md and CLAUDE.md by @PNL-toshiyaishihara in #15
New Contributors
- @PNL-toshiyaishihara made their first contribution in #7
Full Changelog: https://github.com/PNL-toshiyaishihara/nemo-speech-python/commits/v0.1.0rc1
Vendored sources
| Source | Revision | Commit |
|---|---|---|
| NeMo-Speech.cpp | v0.2.0 | 6a3ca3693707 |
| llama.cpp | bd4f514 | bd4f514db14d |
| SentencePiece | v0.2.0 | 17d7580d6407 |
Wheels
Python: 3.10, 3.11, 3.12, 3.13, 3.14 (Requires-Python >=3.10). Every wheel is py3-none, so one file serves all of them.
| Variant | Platform | Requirements | File |
|---|---|---|---|
| CPU | Linux aarch64 (glibc 2.28+) | nemo_speech-0.1.0rc1-py3-none-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl |
|
| CPU | Linux x86_64 (glibc 2.28+) | nemo_speech-0.1.0rc1-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl |
|
| CPU | Windows x64 | nemo_speech-0.1.0rc1-py3-none-win_amd64.whl |
|
| CPU + Metal | macOS arm64 (13.0+) | nemo_speech-0.1.0rc1-py3-none-macosx_13_0_arm64.whl |
|
| Vulkan | Windows x64 | GPU driver with Vulkan | nemo_speech-0.1.0rc1+vulkan-py3-none-win_amd64.whl |
| CUDA 12.8 | Linux x86_64 (glibc 2.28+) | NVIDIA driver R570+; 75-real, 80-real, 86-real, 89-real, 90-real, 120 | nemo_speech-0.1.0rc1+cu128-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl |
| CUDA 12.8 | Windows x64 | NVIDIA driver R570+; 75-real, 80-real, 86-real, 89-real, 90-real, 120 | nemo_speech-0.1.0rc1+cu128-py3-none-win_amd64.whl |
| CUDA 13.0 | Linux x86_64 (glibc 2.28+) | NVIDIA driver R580+; 75-real, 80-real, 86-real, 89-real, 90-real, 120 | nemo_speech-0.1.0rc1+cu130-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl |
| CUDA 13.0 | Windows x64 | NVIDIA driver R580+; 75-real, 80-real, 86-real, 89-real, 90-real, 120 | nemo_speech-0.1.0rc1+cu130-py3-none-win_amd64.whl |
Install
All variants share the distribution name nemo-speech, so install the file for your platform and backend directly (a version specifier with -f would let pip pick a GPU variant). For example:
pip install https://github.com/PNL-toshiyaishihara/nemo-speech-python/releases/download/v0.1.0rc1/nemo_speech-0.1.0rc1+cu128-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whlGPU wheels bundle their runtime libraries (CUDA runtime, cuBLAS) and need only the GPU driver. nemo_speech.build_info() reports how an installation was built. SHA256SUMS lists the checksums of all files.
Known issues
Upstream issues worked around in this release: https://github.com/PNL-toshiyaishihara/nemo-speech-python/issues?q=label%3Aupstream