Modelpublic

healthforallofus/are-bambara-asr-mindspore

sourceHugging Faceupdated 4mo agoView on Hugging Face

0likes39downloads

Model Card

ARE Bambara ASR (MindSpore)

Checkpoint MindSpore (.ckpt) serialise depuis le modele HuggingFace fine-tune `openai/whisper-small`. Noms de tenseurs = state_dict HuggingFace, type Float32 (relisible par mindspore.load_checkpoint).

Rechargement (MindSpore + MindNLP)

python

import os; os.environ["HF_ENDPOINT"] = "https://huggingface.co"
import mindspore as ms, torch            # torch == mindtorch sous mindnlp
from mindnlp.transformers import WhisperForConditionalGeneration, WhisperConfig

cfg   = WhisperConfig.from_pretrained("healthforallofus/are-bambara-asr-mindspore")
model = WhisperForConditionalGeneration(cfg)
pd = ms.load_checkpoint("are-bambara-asr.ckpt")
model.load_state_dict({k: torch.from_numpy(v.asnumpy()) for k, v in pd.items()}, strict=False)
model.eval()