| from huggingface_hub import hf_hub_download |
| import torch |
| from transformers import AutoModelForSequenceClassification as modelSC, AutoTokenizer as token |
|
|
| model_path = hf_hub_download(repo_id="MienOlle/sentiment_analysis_api", |
| filename="sentimentAnalysis.pth" |
| ) |
| modelToken = token.from_pretrained("mdhugol/indonesia-bert-sentiment-classification") |
| model = modelSC.from_pretrained("mdhugol/indonesia-bert-sentiment-classification", num_labels=3) |
| model.load_state_dict(torch.load(model_path, map_location=torch.device("cpu"))) |
| model.eval() |
|
|
| def predict(input): |
| inputs = modelToken(input, return_tensors="pt", padding=True, truncation=True, max_length=512) |
| |
| with torch.no_grad(): |
| outputs = model(**inputs) |
| |
| logits = outputs.logits |
| ret = logits.argmax.item() |
|
|
| labels = ["positive", "neutral", "negative"] |
| return labels[ret] |
