README.md

---
license: apache-2.0
tags:
- executorch
- xnnpack
- pte
- on-device
- automatic-speech-recognition
base_model:
- openai/whisper-small
---
# Whisper-small — ExecuTorch (encoder + decoder)

Speech recognition in two `.pte` files: the encoder runs once per 30-second window, the
decoder once per generated token. Putting them in one graph would re-encode the audio on
every step.

| graph | build | file | size (MB) | corr vs fp32 eager | ms | eager ms |
|---|---|---|---|---|---|---|
| encoder | XNNPACK fp32 | `whisper_small_encoder_xnnpack_fp32.pte` | 352.8 | 1.000000 | 357.4 | 144.7 |
| encoder | XNNPACK fp16 | `whisper_small_encoder_xnnpack_fp16.pte` | 180.5 | 1.000000 | 619.5 | 144.4 |
| encoder | XNNPACK int8 | `whisper_small_encoder_xnnpack_int8.pte` | 98.3 | 0.999052 | 339.3 | 151.8 |
| encoder | Core ML | `whisper_small_encoder_coreml_all.pte` | 176.8 | 0.999942 | 95.5 | 143.4 |
| decoder | XNNPACK fp32 | `whisper_small_decoder_xnnpack_fp32.pte` | 774.0 | 1.000000 | 91.2 | 55.3 |
| decoder | XNNPACK fp16 | `whisper_small_decoder_xnnpack_fp16.pte` | 387.3 | 0.999994 | 188.6 | 55.3 |
| decoder | XNNPACK int8 | `whisper_small_decoder_xnnpack_int8.pte` | 315.5 | 0.994395 | 85.2 | 57.6 |
| decoder | Core ML | `whisper_small_decoder_coreml_all.pte` | 307.6 | 0.999863 | 13.7 | 55.6 |

Every file takes and returns fp32 tensors (token ids stay int64), so any encoder pairs with
any decoder. The lightest working pair is 405.9 MB.

- **Source**: [openai/whisper-small](https://huggingface.co/openai/whisper-small)
- **License**: Apache-2.0
- **Encoder input**: log-mel spectrogram `[1, 80, 3000]` — 30 s at 16 kHz, 80 mel bins, hop
  160, window 400, exactly what `WhisperFeatureExtractor` produces
- **Decoder input**: the encoder output plus `decoder_input_ids [1, 128]` int64,
  left-aligned and padded. Start with `<|startoftranscript|>`, a language token,
  `<|transcribe|>`, `<|notimestamps|>`.

## Decoding

No KV cache: the decoder is a static graph over a fixed 128-token window, so a greedy step
is take `argmax` of row `len-1`, append it, run again. Stop at `<|endoftext|>` (50257). 128
tokens covers a 30-second window of ordinary speech; past that, start a new window.

That costs a full 128-position forward pass per token, which is the price of a static graph
that runs unchanged across runtimes and precisions.

## Verification (Mac arm64, executorch 1.4.0, torch 2.13.0)

The two wrappers compose back to `WhisperForConditionalGeneration` exactly — max_abs_diff
**0.000e+00** — and every graph matches torch fp32 eager at the correlations above. Timings
are medians over 5 runs in one process: a relative reference, not a device number.

## Two things worth knowing about the sizes

**The decoder `.pte` is larger than the decoder's weights.** Whisper ties `proj_out.weight`
to `decoder.embed_tokens.weight`, but the two uses need different representations: an
embedding table the portable kernels index into, and the same values packed into the XNNPACK
delegate's blob for the output matmul. Tying them in PyTorch does not tie them here.
Referencing the weight through `F.linear` instead of the `proj_out` module does not either —
exported both ways, whisper-tiny's decoder comes out at 198.0 MB exactly.

**The decoder's int8 build is the smallest portable one.** Dynamic int8 quantizes the linear weights and
leaves the token embedding table in fp32, and that table is 159.3 MB — 51,865 tokens at
768 dimensions. On this size the rest of the decoder finally outweighs it: int8 comes in at 315.5 MB against fp16's 387.3 MB, so it is the smallest portable build and it ships.

Until recently there was no decoder int8 build at all, and this card said PT2E was observing
the int64 `decoder_input_ids`. That was wrong on both halves.
`XNNPACKQuantizer.transform_for_annotation` rewrites every scalar argument of
`add.Tensor`/`mul.Tensor` as `torch.tensor(float(arg))` whatever the node's dtype — one line
in ExecuTorch's `backends/xnnpack/quantizer/xnnpack_quantizer_utils.py`, still present on
main. In this decoder the casualty is `position_ids = torch.arange(...) + past_key_values_length`
(`modeling_whisper.py:749`, `past_key_values_length` being a python `int`): it comes back
float32, and the failure lands on `self.weight[position_ids]` — the **position** embedding
lookup, not the token ids, and no observer involved. Measured by running `prepare_pt2e`
with an empty quantizer and printing the failing node.

### Checked in the task's own units

Correlation is a first filter. These are the numbers that decide:

- **encoder int8** — measured end to end — word error rate against the fp32 encoder: mean WER 0.0% (worst clip 0.0%) over 5 spoken sentences, int8 encoder against the fp32 encoder with the same fp32 decoder and the same waveform; the fp32 arm transcribes all five correctly, so the comparison is against a working control rather than against noise.
- **decoder int8** — measured end to end — word error rate against the fp32 decoder: mean WER 0.0% (worst clip 0.0%) over 5 spoken sentences, the int8 decoder against the fp32 one with the other half and the waveform held identical; the fp32 arm transcribes all five correctly, so the comparison is against a working control rather than against noise.

The sensitivity of that test, measured by injecting random noise into whisper-tiny's encoder output: rel_l2 0.03 (what int8 actually costs) and 0.10 both give WER 0.000; 0.20 and 0.40 give 0.025. Five clean sentences leave headroom, so a pass means *does not break the transcript*, not *indistinguishable at any error level*.

## Conversion

```bash
python convert/export_whisper.py small
```

The ExecuTorch tree ships a single-graph Whisper example under `examples/models/whisper`;
this is that model with the halves separated.

(conversion scripts: [executorch-models](https://github.com/john-rocky/executorch-models))