From e2b351e04ce309b1f94d6281e33d0e84474a4ae5 Mon Sep 17 00:00:00 2001
From: Christian Medina <37550954+cmedinasoriano@users.noreply.github.com>
Date: Thu, 2 Jul 2026 13:34:33 -0400
Subject: [PATCH] fix: actually quantize model with quantize_4bit()

---
 test_model_loading.py | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/test_model_loading.py b/test_model_loading.py
index be80840..328f7dc 100644
--- a/test_model_loading.py
+++ b/test_model_loading.py
@@ -248,6 +248,12 @@ def test_strategy_6():
             model,
             use_gradient_checkpointing=False,
         )
+        print("   ✓ Model prepared for k-bit training")
+        
+        # Actually quantize the model
+        from bitsandbytes.nn.modules import Params4bit
+        print("   Quantizing weights to 4-bit...")
+        model.quantize_4bit()
         print("   ✓ Model quantized to 4-bit (~17.5GB)")
         
         print("\n   Step 3: Move to GPU 0...")