Upload train_ner.py with huggingface_hub

train_ner.py

@@ -28,16 +28,14 @@ DATASET_NAME = "ai4privacy/pii-masking-200k"
 HUB_MODEL_ID = "jefftherover/modernbert-pii-ner"
 OUTPUT_DIR   = "modernbert-pii-ner"
 MAX_LENGTH   = 512
-SUBSET_SIZE  = 20_000
 
-# 1. Load data
+# ── 1. Load full English dataset ─────────────────────────────────────────────
 print("Loading dataset...")
 full = load_dataset(DATASET_NAME, split="train")
 en   = full.filter(lambda x: x["language"] == "en")
 print(f"English rows: {len(en)}")
 
-subset   = en.select(range(min(SUBSET_SIZE, len(en))))
-splits   = subset.train_test_split(test_size=0.1, seed=42)
+splits   = en.train_test_split(test_size=0.1, seed=42)
 train_ds = splits["train"]
 eval_ds  = splits["test"]
 print(f"Train: {len(train_ds)}  Eval: {len(eval_ds)}")
@@ -134,22 +132,23 @@ model = AutoModelForTokenClassification.from_pretrained(
 )
 
 # 7. Trackio
-trackio.init(project="modernbert-pii-ner", name="modernbert-pii-ner-20k-v1")
+trackio.init(project="modernbert-pii-ner", name="modernbert-pii-ner-43k-v2")
 
-# 8. Training args
+# ── 8. Training args ─────────────────────────────────────────────────────────
 args = TrainingArguments(
     output_dir=OUTPUT_DIR,
-    num_train_epochs=3,
+    num_train_epochs=5,
     per_device_train_batch_size=16,
     per_device_eval_batch_size=32,
-    gradient_accumulation_steps=2,
-    learning_rate=2e-5,
+    gradient_accumulation_steps=2,        # effective batch = 32
+    learning_rate=5e-5,
     weight_decay=0.01,
-    warmup_ratio=0.1,
+    warmup_ratio=0.2,
+    lr_scheduler_type="cosine_with_restarts",
     eval_strategy="steps",
-    eval_steps=200,
+    eval_steps=500,
     save_strategy="steps",
-    save_steps=200,
+    save_steps=500,
     save_total_limit=3,
     load_best_model_at_end=True,
     metric_for_best_model="f1",
@@ -158,9 +157,9 @@ args = TrainingArguments(
     hub_model_id=HUB_MODEL_ID,
     hub_strategy="every_save",
     report_to="trackio",
-    run_name="modernbert-pii-ner-20k-v1",
+    run_name="modernbert-pii-ner-43k-v2",
     fp16=True,
-    logging_steps=50,
+    logging_steps=100,
     dataloader_num_workers=2,
 )