Views
No views yet
optimum-cli1optimum-cli export openvino \
2 --model openai/whisper-large-v3-turbo \
3 --weight-format fp16 \
4 whisper-large-v3-turbo-fp16| Parameter | Value |
|---|---|
| Source model | openai/whisper-large-v3-turbo |
| Export format | OpenVINO IR (.xml + .bin) |
| Weight format | FP16 (half precision) |
| Optimum version | Latest at time of conversion |
| OpenVINO version | 2026.0.0.0 |
| File | Description |
|---|---|
openvino_encoder_model.xml/bin | Whisper encoder (audio → features) |
openvino_decoder_model.xml/bin | Whisper decoder (features → tokens, with KV-cache support) |
openvino_tokenizer.xml/bin | Tokenizer (text → token IDs) |
openvino_detokenizer.xml/bin | Detokenizer (token IDs → text) |
config.json | Model configuration |
generation_config.json | Generation parameters |
preprocessor_config.json | Audio preprocessing configuration |
tokenizer.json | Tokenizer vocabulary and rules |
tokenizer_config.json | Tokenizer configuration |
vocab.json | Token vocabulary mapping |
merges.txt | BPE merge rules |
special_tokens_map.json | Special token definitions |
added_tokens.json | Additional tokens |
normalizer.json | Text normalization rules |
WhisperPipeline.1#include <openvino/genai/whisper_pipeline.hpp>
2
3ov::genai::WhisperPipeline pipeline("path/to/model", "GPU");
4ov::genai::WhisperGenerationConfig config;
5config.language = "<|en|>";
6config.task = "transcribe";
7config.return_timestamps = true;
8
9auto result = pipeline.generate(audio_samples, config);