fix: disable WandB

train_sft.py

@@ -1,10 +1,12 @@
-# train_sft.py
+import os
 import sys
 import json
 from datasets import load_dataset
 from transformers import AutoModelForCausalLM, AutoTokenizer, TrainingArguments
 from trl import SFTTrainer
 
+os.environ["WANDB_MODE"] = "disabled"
+
 DRY_RUN = "--dry-run" in sys.argv
 
 MODEL_ID = "Salesforce/codegen-350M-multi"
@@ -55,7 +57,7 @@ training_args = TrainingArguments(
     save_steps       = 200,
 
     fp16             = False,            # no half‐precision on CPU
-    report_to        = None,             # disable WandB/others
+    report_to        = [],             # disable WandB/others
 )
 
 # 4) instantiate trainer
@@ -65,8 +67,9 @@ trainer = SFTTrainer(
     train_dataset=(tokenized if DRY_RUN else ds),
 )
 
-print(f"\n✅ DRY-RUN: Trainer instantiated:\n  – model: {type(model)}\n  – tokenizer: {type(tokenizer)}\n  – train_dataset size: {len(tokenized if DRY_RUN else ds)}")
-print(f"  – SFTTrainingArguments: {training_args}")
+if DRY_RUN:
+    print(f"\n✅ DRY-RUN: Trainer instantiated:\n  – model: {type(model)}\n  – tokenizer: {type(tokenizer)}\n  – train_dataset size: {len(tokenized if DRY_RUN else ds)}")
+    print(f"  – SFTTrainingArguments: {training_args}")
 
 if not DRY_RUN:
     # only run the real training if you didn’t pass --dry-run