Update README.md
Browse files
README.md
CHANGED
|
@@ -67,15 +67,28 @@ model_name = "InfocubeSrl/LexCube"
|
|
| 67 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 68 |
model = AutoModelForMaskedLM.from_pretrained(model_name)
|
| 69 |
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 73 |
|
| 74 |
-
mask_index = (inputs["input_ids"][0] == tokenizer.mask_token_id).nonzero()[0]
|
| 75 |
-
predicted_id = outputs.logits[0, mask_index].argmax()
|
| 76 |
-
predicted_token = tokenizer.decode(predicted_id)
|
| 77 |
-
|
| 78 |
-
print("Prediction:", predicted_token)
|
| 79 |
```
|
| 80 |
|
| 81 |
|
|
@@ -90,3 +103,5 @@ print("Prediction:", predicted_token)
|
|
| 90 |
- Structured format with numbered provisions and cross-citations
|
| 91 |
- Avg. length: ~909 words (≈2,193 tokens per document); some documents exceed 11k tokens
|
| 92 |
- **Confidentiality:** Raw dataset cannot be shared due to contractual agreements, but it has been statistically and linguistically analyzed for research
|
|
|
|
|
|
|
|
|
| 67 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 68 |
model = AutoModelForMaskedLM.from_pretrained(model_name)
|
| 69 |
|
| 70 |
+
# Examples with [MASK]
|
| 71 |
+
examples = [
|
| 72 |
+
"[MASK] il Decreto Legislativo 18 agosto 2000, n. 267 (Testo Unico delle leggi sull'ordinamento degli Enti Locali)",
|
| 73 |
+
"ACQUISITI, ai sensi dell'art. [MASK] del D.Lgs. 267/2000, i pareri favorevoli di regolarità tecnica e di regolarità contabile",
|
| 74 |
+
"Visto gli art. [MASK] e 42 del D.Lgs n.267/2000, Testo unico degli enti locali.",
|
| 75 |
+
"DI DICHIARARE la presente deliberazione immediatamente [MASK] ai sensi dell'art. 134, comma 4, del D.Lgs. n. 267/2000."
|
| 76 |
+
]
|
| 77 |
+
|
| 78 |
+
for text in examples:
|
| 79 |
+
inputs = tokenizer(text, return_tensors="pt")
|
| 80 |
+
outputs = model(**inputs)
|
| 81 |
+
|
| 82 |
+
# Find mask token position
|
| 83 |
+
mask_index = (inputs["input_ids"][0] == tokenizer.mask_token_id).nonzero(as_tuple=True)[0]
|
| 84 |
+
|
| 85 |
+
# Get top prediction
|
| 86 |
+
predicted_id = outputs.logits[0, mask_index].argmax(dim=-1)
|
| 87 |
+
predicted_token = tokenizer.decode(predicted_id)
|
| 88 |
+
|
| 89 |
+
print(f"Input: {text}")
|
| 90 |
+
print(f"Prediction: {predicted_token}\n")
|
| 91 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 92 |
```
|
| 93 |
|
| 94 |
|
|
|
|
| 103 |
- Structured format with numbered provisions and cross-citations
|
| 104 |
- Avg. length: ~909 words (≈2,193 tokens per document); some documents exceed 11k tokens
|
| 105 |
- **Confidentiality:** Raw dataset cannot be shared due to contractual agreements, but it has been statistically and linguistically analyzed for research
|
| 106 |
+
|
| 107 |
+
|