Text Generation
PEFT
Safetensors
Portuguese
text-classification
community-notes
portuguese
reranker
lora
misinformation
Instructions to use histlearn/community-notes-reranker-ptbr with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use histlearn/community-notes-reranker-ptbr with PEFT:
Task type is invalid.
- Notebooks
- Google Colab
- Kaggle
| """Inferencia rapida usando um unico fold (fold 1). Ideal para demos. | |
| Em CPU: ~5-10s por par. Em GPU T4: ~50ms. | |
| """ | |
| import json, torch | |
| from transformers import AutoTokenizer, AutoModelForCausalLM | |
| from peft import PeftModel | |
| from huggingface_hub import snapshot_download | |
| REPO = "histlearn/community-notes-reranker-ptbr" | |
| path = snapshot_download(REPO, allow_patterns=["manifesto.json", "adapter_fold_1/*"]) | |
| m = json.load(open(f"{path}/manifesto.json")) | |
| tok = AutoTokenizer.from_pretrained(m["base_model"], padding_side="left") | |
| model = AutoModelForCausalLM.from_pretrained( | |
| m["base_model"], torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32) | |
| model = PeftModel.from_pretrained(model, f"{path}/adapter_fold_1") | |
| if torch.cuda.is_available(): model.cuda() | |
| model.eval() | |
| def score(tweet, nota): | |
| text = (m["prompt_prefixo"] + "<Instruct>: " + m["instrucao"] + | |
| "\n<Query>: " + tweet + "\n<Document>: " + nota + m["prompt_sufixo"]) | |
| enc = tok(text, return_tensors="pt", truncation=True, max_length=m["max_length"]).to(model.device) | |
| with torch.no_grad(): | |
| logits = model(**enc).logits[:, -1, :] | |
| return float(torch.sigmoid(logits[:, m["id_yes"]] - logits[:, m["id_no"]]).item()) | |
| print(score("Bolsonaro disse que a Terra e plana", | |
| "Bolsonaro nunca afirmou isso; checagem em https://exemplo.org")) | |