Run vLLM inference with quantized models and benchmark on Arm servers
Learn how to run Llama 3.1 8B and Whisper inference with BF16 and W8A8 quantized models on Arm servers, then compare throughput and accuracy with vLLM benchmark tools.
Learn how to run Llama 3.1 8B and Whisper inference with BF16 and W8A8 quantized models on Arm servers, then compare throughput and accuracy with vLLM benchmark tools.
This Learning Path is for software developers familiar with basic machine learning concepts and looking to run the OpenAI Whisper Automatic Speech Recognition (ASR) model efficiently, using an Arm-based cloud instance.
Learn how to deploy the ModelScope FunASR Chinese automatic speech recognition model on Arm-based servers with real-time transcription and sentiment analysis.
Evaluation Conditions
Automatic Speech Recognition · AWS Graviton G4 · INT8 · LibriSpeech (test)
| Metric | Resultsvs baseline | Change | |
|---|---|---|---|
| End to end latency p50 | 3113.55ms4012.95ms | −899.40 ms | |
| Peak memory | 1728.44MB4058.66MB | −2330.21 MB | |
| Word error rate | 0.02%0.02% | 0.00% | |
| Character error rate | 0.02%0.02% | 0.00% |
The Optimized Model is provided as a reference implementation to demonstrate and evaluate execution and performance on Arm-based systems. It is not a production-ready or supported solution.
Arm’s publication of the Optimized Model does not constitute an endorsement or certification of the Original Model or a representation that the Optimized Model is suitable for production use or any particular purpose.
To the fullest extent permitted by applicable law (i) the Optimized Model is provided “as is.” Arm makes no representations or warranties that the Original Model, the Optimized Model or their outputs are accurate, safe, secure, non-infringing, legally compliant, suitable for production use or fit for any particular purpose; and (ii) Arm will not be liable for any loss or damage arising from or in connection with the Optimized Model, its use or its outputs.
You are responsible for independently evaluating the Optimized Model, its outputs and its suitability for your intended use, including compliance with applicable legal, regulatory, safety and security requirements.
Arm does not commit to provide ongoing support, maintenance or updates for the Optimized Model. Any use of or reliance on the Optimized Model or its outputs is at your own risk.
Use of the Portal is subject to the Arm Website Terms and Conditions of Use.
| Configuration | Baseline | Arm Optimized |
|---|---|---|
| Dataset | LibriSpeech (test) | LibriSpeech (test) |
| Sample count | 100 | 100 |
| Batch size | 1 | 1 |
| Mode | offline | offline |
| Normalization | Whisper-style | Whisper-style |
| Sample rate (Hz) | 16000 | 16000 |
| Audio length (s) | 30 | 30 |
| Generation length | 1 | 1 |
| Runs | 30 | 30 |