Repository navigation
Load QNN context binary for faster startup - #2877
Conversation
dosubot
Bot
added
the
size:XL
Summary of ChangesHello @csukuangfj, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed! This pull request enhances the Highlights
Ignored Files
Using Gemini Code AssistThe full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips. Invoking Gemini You can request assistance from Gemini at any point by creating a comment using either
Customization To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a Limitations & Feedback Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here. You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension. Footnotes
|
|
Warning Rate limit exceeded@csukuangfj has exceeded the limit for the number of commits or files that can be reviewed per hour. Please wait 19 minutes and 39 seconds before requesting another review. ⌛ How to resolve this issue?After the wait time has elapsed, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans have higher rate limits than the trial, open-source and free plans. In all cases, we re-allow further reviews after a brief timeout. Please see our FAQ for further information. 📒 Files selected for processing (2)
WalkthroughThis PR introduces QNN (Qualcomm Neural Network) support for SenseVoice with new device information and configuration generation scripts. It significantly overhauls the export workflow to use dynamic matrix-driven SOC-aware builds, modernizes binary versions, and extends model validation to support QNN context binaries as alternatives to full models across C++, Kotlin, and Android codebases. Changes
Sequence DiagramsequenceDiagram
participant CI as GitHub Actions
participant MatrixGen as generate_build_matrix Job
participant Build as export-sense-voice-to-qnn Job
participant Pkg as Packaging Step
participant Release as Release Step
CI->>MatrixGen: Trigger
MatrixGen->>MatrixGen: Run generate_sense_voice.py<br/>(enumerate SOC × duration × framework)
MatrixGen-->>CI: Output dynamic matrix
CI->>Build: Spawn per-SOC build instances<br/>(needs: generate_build_matrix)
loop For each SOC in matrix
Build->>Build: Setup Python 3.10
Build->>Build: Download toolkit v2.40.0.251030
Build->>Build: Generate HTP config<br/>(generate_config.py)
Build->>Build: Build model binary
Build->>Build: Generate context binary<br/>(qnn-context-binary-generator)
Build->>Pkg: Model + context binary ready
end
Pkg->>Pkg: Create directory structure<br/>(d/{SOC}/{language})
Pkg->>Pkg: Copy README, LICENSE, model, binary
Pkg->>Pkg: Generate info.txt
Pkg->>Pkg: Tar/Bz2 packaging
Pkg->>Pkg: Organize into binary/ and so/
Pkg->>Release: Artifacts staged
Release->>Release: Publish to csukuangfj & k2-fsa<br/>with tags (asr-models-qnn,<br/>asr-models-qnn-binary)
Release-->>CI: Release complete
Estimated code review effort🎯 4 (Complex) | ⏱️ ~60 minutes
Possibly related PRs
Poem
Pre-merge checks and finishing touches❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 7
🧹 Nitpick comments (2)
.github/scripts/export-qnn/generate_sense_voice.py (1)
20-39: Consider the potential size of the generated configuration matrix.The script generates all combinations of SOCs, durations (11 values), and frameworks (2 values), resulting in 22 configurations per SOC. Depending on the number of SOCs in
soc_info_dict, this could create a large CI matrix that may impact build times and resource usage.If CI execution time or cost becomes a concern, consider:
- Prioritizing a subset of durations for automated testing (e.g., 10, 20, 30 seconds)
- Testing all combinations only for specific SOCs or on-demand
- Using a staged approach where critical combinations run first
scripts/qnn/generate_config.py (1)
47-97: Consider validatingqnn_sdk_rootpath exists.The function assumes
qnn_sdk_rootis valid. If an invalid path is provided, the JSON files will be written with broken paths that will only fail at runtime when QNN tools try to load the shared library.def generate_config( soc_name: str, graph_name: str, output_dir: str, qnn_sdk_root: str, ): if soc_name not in soc_info_dict: raise ValueError( f"Unsupported SOC {soc_name}. Supported: - {sorted(list(soc_info_dict.keys()))}" ) soc = soc_info_dict[soc_name] + qnn_sdk_path = Path(qnn_sdk_root) + if not qnn_sdk_path.exists(): + raise ValueError(f"QNN SDK root does not exist: {qnn_sdk_root}") + output_dir = Path(output_dir).absolute()
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (12)
.github/scripts/export-qnn/device_info.py(1 hunks).github/scripts/export-qnn/generate_config.py(1 hunks).github/scripts/export-qnn/generate_sense_voice.py(1 hunks).github/workflows/export-sense-voice-to-qnn.yaml(12 hunks).gitignore(1 hunks)android/SherpaOnnxSimulateStreamingAsr/app/src/main/java/com/k2fsa/sherpa/onnx/simulate/streaming/asr/SimulateStreamingAsr.kt(1 hunks)scripts/qnn/device_info.py(1 hunks)scripts/qnn/generate_config.py(1 hunks)sherpa-onnx/csrc/offline-model-config.cc(1 hunks)sherpa-onnx/csrc/offline-recognizer-impl.cc(2 hunks)sherpa-onnx/csrc/offline-sense-voice-model-config.cc(2 hunks)sherpa-onnx/kotlin-api/OfflineRecognizer.kt(1 hunks)
🧰 Additional context used
🧠 Learnings (2)
📚 Learning: 2025-08-06T04:23:50.237Z
Learnt from: litongjava
Repo: k2-fsa/sherpa-onnx PR: 2440
File: sherpa-onnx/java-api/src/main/java/com/k2fsa/sherpa/onnx/core/Core.java:4-6
Timestamp: 2025-08-06T04:23:50.237Z
Learning: The sherpa-onnx JNI library files are stored in Hugging Face repository at https://huggingface.co/csukuangfj/sherpa-onnx-libs under versioned directories like jni/1.12.7/, and the actual Windows JNI library filename is "sherpa-onnx-jni.dll" as defined in Core.java constants.
Applied to files:
.gitignore
📚 Learning: 2025-08-06T04:18:47.981Z
Learnt from: litongjava
Repo: k2-fsa/sherpa-onnx PR: 2440
File: sherpa-onnx/java-api/src/main/java/com/k2fsa/sherpa/onnx/core/Core.java:4-6
Timestamp: 2025-08-06T04:18:47.981Z
Learning: In sherpa-onnx Java API, the native library names in Core.java (WIN_NATIVE_LIBRARY_NAME = "sherpa-onnx-jni.dll", UNIX_NATIVE_LIBRARY_NAME = "libsherpa-onnx-jni.so", MACOS_NATIVE_LIBRARY_NAME = "libsherpa-onnx-jni.dylib") are copied directly from the compiled binary filenames and should not be changed to match other libraries' naming conventions.
Applied to files:
.gitignore
🧬 Code graph analysis (2)
sherpa-onnx/csrc/offline-sense-voice-model-config.cc (1)
sherpa-onnx/kotlin-api/OfflineRecognizer.kt (7)
model(23-25)model(27-29)model(31-33)model(35-38)model(40-42)model(44-46)model(76-81)
scripts/qnn/generate_config.py (3)
.github/scripts/export-qnn/generate_config.py (3)
get_args(14-44)generate_config(47-97)_test(100-107)scripts/qnn/device_info.py (1)
_test(98-108).github/scripts/export-qnn/device_info.py (1)
_test(98-108)
🪛 actionlint (1.7.9)
.github/workflows/export-sense-voice-to-qnn.yaml
27-27: workflow command "set-output" was deprecated. use echo "{name}={value}" >> $GITHUB_OUTPUT instead: https://docs.github.com/en/actions/using-workflows/workflow-commands-for-github-actions
(deprecated-commands)
🪛 Cppcheck (2.18.0)
sherpa-onnx/csrc/offline-recognizer-impl.cc
[information] Limiting analysis of branches. Use --check-level=exhaustive to analyze all branches.
(normalCheckLevelMaxBranches)
[information] Limiting analysis of branches. Use --check-level=exhaustive to analyze all branches.
(normalCheckLevelMaxBranches)
🪛 Flake8 (7.3.0)
scripts/qnn/device_info.py
[error] 3-3: 'enum.unique' imported but unused
(F401)
🪛 Ruff (0.14.8)
scripts/qnn/device_info.py
29-29: An enum class should not be decorated with @dataclass
(RUF049)
52-52: An enum class should not be decorated with @dataclass
(RUF049)
scripts/qnn/generate_config.py
54-56: Avoid specifying long messages outside the exception class
(TRY003)
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (21)
- GitHub Check: Release static tts-OFF
- GitHub Check: Release static tts-ON
- GitHub Check: Release shared tts-ON
- GitHub Check: Debug shared-ON tts-ON
- GitHub Check: Release shared-ON tts-OFF
- GitHub Check: Debug shared-OFF tts-OFF
- GitHub Check: Release shared-OFF tts-OFF
- GitHub Check: Release shared-OFF tts-ON
- GitHub Check: rknn shared OFF
- GitHub Check: Debug shared-ON tts-OFF
- GitHub Check: Release shared-ON tts-ON
- GitHub Check: Debug shared-OFF tts-ON
- GitHub Check: rknn shared ON
- GitHub Check: ubuntu-24.04 3.10
- GitHub Check: ubuntu-24.04 3.13
- GitHub Check: ubuntu-24.04 3.11
- GitHub Check: ubuntu-24.04 3.12
- GitHub Check: swift (macos-latest)
- GitHub Check: ubuntu-24.04 3.9
- GitHub Check: ubuntu-24.04 3.8
- GitHub Check: swift (macos-13)
🔇 Additional comments (10)
.gitignore (1)
166-166: Pattern addition aligns with PR objectives and existing conventions.The new
sherpa-onnx-qnn-*pattern appropriately ignores QNN-related artifacts and binaries, consistent with similar model artifact patterns already in the file (e.g.,sherpa-onnx-sense-voice-*,sherpa-onnx-vits-*). This complements the PR's goal to support loading QNN context binaries for faster startup.sherpa-onnx/csrc/offline-model-config.cc (1)
138-139: LGTM! Validation now supports QNN context binary path.The updated condition correctly triggers sense_voice validation when either a model or a QNN context binary is provided, enabling the faster startup path described in the PR objectives.
sherpa-onnx/csrc/offline-sense-voice-model-config.cc (1)
32-67: LGTM! Comprehensive validation for QNN context binary support.The validation logic correctly handles both traditional model-based and new context-binary-based paths:
- When
context_binaryis empty, requires a valid model file- When
modelis empty butcontext_binaryis provided, verifies the context binary exists- Properly delegates to
qnn_config.Validate()for QNN-specific validationsherpa-onnx/csrc/offline-recognizer-impl.cc (1)
174-175: LGTM! Factory logic updated consistently across both overloads.The QNN branch condition now correctly instantiates the
OfflineSenseVoiceModelQnnimplementation when eithersense_voice.modelorsense_voice.qnn_config.context_binaryis provided. This change is applied consistently in bothCreate()overloads and aligns with the validation logic updates.Also applies to: 495-496
android/SherpaOnnxSimulateStreamingAsr/app/src/main/java/com/k2fsa/sherpa/onnx/simulate/streaming/asr/SimulateStreamingAsr.kt (1)
135-149: LGTM! Asset copying logic correctly handles QNN context binary scenarios.The updated logic properly handles both model-based and context-binary-only paths:
- Checks if either the model or context binary asset exists before attempting to copy
- Only copies the model if it's non-empty
- Always prepares the context binary path (which may not exist initially but will be created on first run)
This aligns with the QNN context binary workflow where the binary can be generated and cached after the first run.
.github/workflows/export-sense-voice-to-qnn.yaml (2)
14-34: LGTM - Dynamic matrix generation looks good.The matrix generation job correctly outputs the matrix via
$GITHUB_OUTPUTand the deprecatedset-outputis appropriately commented out as a reference.
567-605: Release conditions use different SOC filters - verify intent.The release steps have asymmetric SOC filtering:
- Lines 568 and 590:
csukuangfjandk2-fsaonly release.sofiles whenmatrix.soc == 'SM8850'- Lines 579 and 599: Binary releases happen for all SOCs
Is this intentional that
.soreleases are limited to SM8850 only while binaries are released for all SOCs?scripts/qnn/generate_config.py (2)
1-11: LGTM - Clean script structure with good documentation references.The script is well-organized with clear CLI arguments and follows the same pattern as
.github/scripts/export-qnn/generate_config.py.
100-107: Hardcoded test path is fine for local testing.The
_test()function with hardcoded path is acceptable for development/debugging purposes and is properly commented out in__main__.scripts/qnn/device_info.py (1)
63-95: LGTM - Well-structured device info data.The
HtpInfo,SocInfodataclasses and thesoc_info_list/soc_info_dictstructures are well-organized and provide a clean mapping of SOC metadata.
| echo "collect results" | ||
|
|
||
| d=sherpa-onnx-qnn-${{ matrix.soc}}-binary-$t-seconds-sense-voice-zh-en-ja-ko-yue-2024-07-17-int8 | ||
| mkdir -p $d | ||
| mkdir -p $d/test_wavs | ||
|
|
||
| cp -v README.md $d | ||
| cp -v LICENSE $d | ||
| cp -v binary/model.bin $d/ | ||
| cp -v tokens.txt $d | ||
| cp -v *.wav $d/test_wavs | ||
|
|
||
| echo "num_frames=$num_frames" > $d/info.txt | ||
| echo "target=$p" >> $d/info.txt | ||
|
|
||
| ls -lh $d | ||
| tar cjfv $d.tar.bz2 $d | ||
| ls -lh *.tar.bz2 | ||
| rm -rf $d | ||
| mv *.tar.bz2 ../../../binary/ |
There was a problem hiding this comment.
Undefined variable $p will cause incorrect output.
At line 332, $p is referenced but it's not defined until the for loop at line 341. This will result in an empty value being written to info.txt.
Apply this diff to fix the issue:
echo "num_frames=$num_frames" > $d/info.txt
- echo "target=$p" >> $d/info.txt
+ echo "target=binary" >> $d/info.txtOr if you want to indicate the SOC:
echo "num_frames=$num_frames" > $d/info.txt
- echo "target=$p" >> $d/info.txt
+ echo "soc=${{ matrix.soc }}" >> $d/info.txt📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| echo "collect results" | |
| d=sherpa-onnx-qnn-${{ matrix.soc}}-binary-$t-seconds-sense-voice-zh-en-ja-ko-yue-2024-07-17-int8 | |
| mkdir -p $d | |
| mkdir -p $d/test_wavs | |
| cp -v README.md $d | |
| cp -v LICENSE $d | |
| cp -v binary/model.bin $d/ | |
| cp -v tokens.txt $d | |
| cp -v *.wav $d/test_wavs | |
| echo "num_frames=$num_frames" > $d/info.txt | |
| echo "target=$p" >> $d/info.txt | |
| ls -lh $d | |
| tar cjfv $d.tar.bz2 $d | |
| ls -lh *.tar.bz2 | |
| rm -rf $d | |
| mv *.tar.bz2 ../../../binary/ | |
| echo "collect results" | |
| d=sherpa-onnx-qnn-${{ matrix.soc}}-binary-$t-seconds-sense-voice-zh-en-ja-ko-yue-2024-07-17-int8 | |
| mkdir -p $d | |
| mkdir -p $d/test_wavs | |
| cp -v README.md $d | |
| cp -v LICENSE $d | |
| cp -v binary/model.bin $d/ | |
| cp -v tokens.txt $d | |
| cp -v *.wav $d/test_wavs | |
| echo "num_frames=$num_frames" > $d/info.txt | |
| echo "target=binary" >> $d/info.txt | |
| ls -lh $d | |
| tar cjfv $d.tar.bz2 $d | |
| ls -lh *.tar.bz2 | |
| rm -rf $d | |
| mv *.tar.bz2 ../../../binary/ |
🤖 Prompt for AI Agents
.github/workflows/export-sense-voice-to-qnn.yaml around lines 319 to 338: the
script writes "target=$p" to info.txt but $p is not defined yet (defined later
in the for loop), so replace that usage with a defined value — either write the
SOC from the matrix (e.g., matrix.soc) or ensure p is set before this block;
update the echo to use a defined variable (e.g., target=${p:-${{ matrix.soc }}})
or move the info.txt write into the loop after p is assigned so the correct
target is recorded.
| echo "collect results" | ||
|
|
||
| d=sherpa-onnx-qnn-${{ matrix.soc }}-binary-$t-seconds-sense-voice-zh-en-ja-ko-yue-2025-09-09-int8 | ||
| mkdir -p $d | ||
| mkdir -p $d/test_wavs | ||
|
|
||
| cp -v README.md $d | ||
| cp -v binary/model.bin $d/ | ||
| cp -v tokens.txt $d | ||
| cp -v *.wav $d/test_wavs | ||
|
|
||
| echo "num_frames=$num_frames" > $d/info.txt | ||
| echo "target=$p" >> $d/info.txt | ||
|
|
||
| ls -lh $d | ||
| tar cjfv $d.tar.bz2 $d | ||
| ls -lh *.tar.bz2 | ||
| rm -rf $d | ||
| mv *.tar.bz2 ../../../binary/ |
There was a problem hiding this comment.
Undefined variable $p - same issue as FunASR section.
At line 521, $p is referenced but it's not defined until the for loop at line 530. This mirrors the same bug in the FunASR section.
Apply this diff to fix the issue:
echo "num_frames=$num_frames" > $d/info.txt
- echo "target=$p" >> $d/info.txt
+ echo "target=binary" >> $d/info.txt📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| echo "collect results" | |
| d=sherpa-onnx-qnn-${{ matrix.soc }}-binary-$t-seconds-sense-voice-zh-en-ja-ko-yue-2025-09-09-int8 | |
| mkdir -p $d | |
| mkdir -p $d/test_wavs | |
| cp -v README.md $d | |
| cp -v binary/model.bin $d/ | |
| cp -v tokens.txt $d | |
| cp -v *.wav $d/test_wavs | |
| echo "num_frames=$num_frames" > $d/info.txt | |
| echo "target=$p" >> $d/info.txt | |
| ls -lh $d | |
| tar cjfv $d.tar.bz2 $d | |
| ls -lh *.tar.bz2 | |
| rm -rf $d | |
| mv *.tar.bz2 ../../../binary/ | |
| echo "collect results" | |
| d=sherpa-onnx-qnn-${{ matrix.soc }}-binary-$t-seconds-sense-voice-zh-en-ja-ko-yue-2025-09-09-int8 | |
| mkdir -p $d | |
| mkdir -p $d/test_wavs | |
| cp -v README.md $d | |
| cp -v binary/model.bin $d/ | |
| cp -v tokens.txt $d | |
| cp -v *.wav $d/test_wavs | |
| echo "num_frames=$num_frames" > $d/info.txt | |
| echo "target=binary" >> $d/info.txt | |
| ls -lh $d | |
| tar cjfv $d.tar.bz2 $d | |
| ls -lh *.tar.bz2 | |
| rm -rf $d | |
| mv *.tar.bz2 ../../../binary/ |
🤖 Prompt for AI Agents
.github/workflows/export-sense-voice-to-qnn.yaml around lines 509 to 527: the
script writes "target=$p" but $p is undefined at this point (it’s set later in
the for-loop), causing the same bug as in the FunASR section; fix by using the
already-defined variable $t (used earlier to build the directory name) or
otherwise ensure $p is defined before use — specifically replace the "target=$p"
line with "target=$t" so the target value is correct and defined.
| #!/usr/bin/env python3 | ||
| from dataclasses import dataclass | ||
| from enum import IntEnum, unique | ||
|
|
There was a problem hiding this comment.
Remove unused import unique.
The unique decorator is imported but never used.
#!/usr/bin/env python3
from dataclasses import dataclass
-from enum import IntEnum, unique
+from enum import IntEnum📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| #!/usr/bin/env python3 | |
| from dataclasses import dataclass | |
| from enum import IntEnum, unique | |
| #!/usr/bin/env python3 | |
| from dataclasses import dataclass | |
| from enum import IntEnum |
🧰 Tools
🪛 Flake8 (7.3.0)
[error] 3-3: 'enum.unique' imported but unused
(F401)
🤖 Prompt for AI Agents
In scripts/qnn/device_info.py around lines 1 to 4, the imported symbol `unique`
from the enum module is unused; remove the `unique` import from the imports list
(keep the other imports intact) so the file no longer imports an unused name and
update any linter if necessary.
| 9022 -> { | ||
| // for my Xiaomi 17 Pro | ||
| val modelDir = "sherpa-onnx-qnn-SM8850-binary-10-seconds-sense-voice-zh-en-ja-ko-yue-2024-07-17-int8" | ||
| return OfflineModelConfig( | ||
| provider = "qnn", | ||
| senseVoice = OfflineSenseVoiceModelConfig( | ||
| qnnConfig = QnnConfig( | ||
| // Please copy libQnnHtp.so and libQnnSystem.so to jniLibs/arm64-v8a by yourself | ||
| backendLib = "libQnnHtp.so", | ||
| systemLib = "libQnnSystem.so", | ||
| contextBinary = "$modelDir/model.bin", | ||
| ), | ||
| ), | ||
| tokens = "$modelDir/tokens.txt", | ||
| debug = true, | ||
| ) | ||
| } |
There was a problem hiding this comment.
Consider generalizing this device-specific test configuration.
The comment "for my Xiaomi 17 Pro" indicates this is a personal test case for a specific device. While it demonstrates the new QNN context binary feature correctly, including developer-specific test configurations in production code can clutter the codebase.
Consider either:
- Removing the personal comment and making the configuration more generic
- Moving this to an examples directory or test suite
- Using it as a template that users can customize
🤖 Prompt for AI Agents
In sherpa-onnx/kotlin-api/OfflineRecognizer.kt around lines 1117 to 1133, this
block contains a device-specific comment and test configuration ("for my Xiaomi
17 Pro") that should not remain in production code; remove the personal comment,
generalize or rename the modelDir to a neutral example name (e.g.,
"sherpa-onnx-qnn-model-..."), or move the entire case into an examples/tests
file and replace it here with a concise template comment describing how to
customize QNN provider fields; ensure the example still documents the required
backendLib/systemLib/contextBinary fields and keep debug flag as appropriate.
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
The pull request introduces support for loading QNN context binaries to accelerate startup times for SenseVoice models on Android. This involves modifications in C++ and Kotlin code to handle the new context_binary configuration field, ensuring proper validation and asset management. Additionally, new Python scripts are added to generate configuration files related to QNN device information and model settings. The changes appear to be well-integrated and logically sound, enhancing the flexibility and performance of QNN model loading.
| @@ -0,0 +1,112 @@ | |||
| #!/usr/bin/env python3 | |||
| from dataclasses import dataclass | |||
| from enum import IntEnum, unique | |||
There was a problem hiding this comment.
The unique decorator is imported from the enum module but is not used anywhere in this file. Removing unused imports helps maintain code cleanliness and avoids potential confusion.
| from enum import IntEnum, unique | |
| from enum import IntEnum |
1 participant
See the doc at
https://k2-fsa.github.io/sherpa/onnx/qnn/run-executables-on-your-phone-binary.html
Summary by CodeRabbit
Release Notes
New Features
Improvements
✏️ Tip: You can customize this high-level summary in your review settings.