| import os |
| import json |
| import torch |
| import numpy as np |
| from sklearn.model_selection import train_test_split |
| from sklearn.metrics import f1_score |
| from transformers import AutoTokenizer, AutoModelForSequenceClassification, Trainer, TrainingArguments, DataCollatorWithPadding |
| from datasets import Dataset |
|
|
| |
| MODEL_NAME = "answerdotai/ModernBERT-large" |
| REAL_DATA = "real_10k.jsonl" |
| SYNTH_DATA = "synth_10k.jsonl" |
| OUTPUT_DIR = "MODERNBERT_MITRE_COMMUNITY" |
|
|
| |
| PARAMS = { |
| 'learning_rate': 3.5e-05, |
| 'weight_decay': 0.05, |
| 'num_train_epochs': 5, |
| 'batch_size': 16 |
| } |
|
|
| def load_data(): |
| all_data = [] |
| for f in [REAL_DATA, SYNTH_DATA]: |
| with open(f, 'r', encoding='utf-8') as fin: |
| for line in fin: |
| all_data.append(json.loads(line)) |
| |
| all_labels = sorted(list(set(l for item in all_data for l in item['labels']))) |
| label2id = {l: i for i, l in enumerate(all_labels)} |
| id2label = {i: l for i, l in enumerate(all_labels)} |
| |
| texts = [x['text'] for x in all_data] |
| matrix = np.zeros((len(texts), len(all_labels)), dtype=np.float32) |
| for i, item in enumerate(all_data): |
| for lbl in item['labels']: |
| matrix[i, label2id[lbl]] = 1.0 |
| |
| return texts, matrix, id2label, label2id |
|
|
| def train(): |
| texts, labels, id2label, label2id = load_data() |
| train_texts, val_texts, train_labels, val_labels = train_test_split(texts, labels, test_size=0.1, random_state=42) |
| |
| tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME) |
| train_ds = Dataset.from_dict({"text": train_texts, "labels": train_labels}).map( |
| lambda x: tokenizer(x["text"], truncation=True, max_length=1024), batched=True |
| ) |
| val_ds = Dataset.from_dict({"text": val_texts, "labels": val_labels}).map( |
| lambda x: tokenizer(x["text"], truncation=True, max_length=1024), batched=True |
| ) |
| |
| model = AutoModelForSequenceClassification.from_pretrained( |
| MODEL_NAME, num_labels=len(id2label), problem_type="multi_label_classification", |
| id2label=id2label, label2id=label2id |
| ) |
| |
| args = TrainingArguments( |
| output_dir=f"{OUTPUT_DIR}_checkpoints", |
| per_device_train_batch_size=PARAMS['batch_size'], |
| learning_rate=PARAMS['learning_rate'], |
| weight_decay=PARAMS['weight_decay'], |
| num_train_epochs=PARAMS['num_train_epochs'], |
| fp16=torch.cuda.is_available(), |
| eval_strategy="epoch", |
| save_strategy="epoch", |
| load_best_model_at_end=True, |
| metric_for_best_model="f1_micro", |
| report_to="none" |
| ) |
| |
| def compute_metrics(eval_pred): |
| preds = (1 / (1 + np.exp(-eval_pred.predictions)) > 0.5).astype(int) |
| return {'f1_micro': f1_score(eval_pred.label_ids, preds, average='micro', zero_division=0)} |
| |
| trainer = Trainer( |
| model=model, args=args, train_dataset=train_ds, eval_dataset=val_ds, |
| tokenizer=tokenizer, data_collator=DataCollatorWithPadding(tokenizer=tokenizer), |
| compute_metrics=compute_metrics |
| ) |
| |
| print("🚀 Démarrage de l'entraînement communautaire...") |
| trainer.train() |
| trainer.save_model(OUTPUT_DIR) |
| tokenizer.save_pretrained(OUTPUT_DIR) |
| print(f"✅ Modèle sauvegardé dans {OUTPUT_DIR}") |
|
|
| if __name__ == "__main__": |
| train() |
|
|
