EngineX-Iluvatar/enginex_bi_series-sherpa-onnx
Archived
16
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Add int8 quantized whisper large models (#1126)

Browse Source
This commit is contained in:
Fangjun Kuang
2024-07-13 22:30:06 +08:00
committed by GitHub
parent 3951a12f8d
commit ab71c3976d
3 changed files with 41 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
scripts/whisper/test.py
View File

@@ -90,6 +90,7 @@ class OnnxModel:
self.n_text_layer = int(meta["n_text_layer"])
self.n_text_ctx = int(meta["n_text_ctx"])
self.n_text_state = int(meta["n_text_state"])
self.n_mels = int(meta["n_mels"])
self.sot = int(meta["sot"])
self.eot = int(meta["eot"])
self.translate = int(meta["translate"])
@@ -294,8 +295,9 @@ def main():
args = get_args()
model = OnnxModel(args.encoder, args.decoder)
dim = 80 if "large-v3" not in args.encoder else 128
mel = compute_features(args.sound_file, dim=dim)
n_mels = model.n_mels
mel = compute_features(args.sound_file, dim=n_mels)
n_layer_cross_k, n_layer_cross_v = model.run_encoder(mel)