cmedina/agenx-lora-training
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: actually quantize model with quantize_4bit()

Browse Source
This commit is contained in:
Christian Medina
2026-07-02 13:34:33 -04:00
parent aff91c990b
commit e2b351e04c
1 changed files with 6 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
test_model_loading.py
View File

@@ -248,6 +248,12 @@ def test_strategy_6():
model, model,
use_gradient_checkpointing=False, 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(" ✓ Model quantized to 4-bit (~17.5GB)")
print("\n Step 3: Move to GPU 0...") print("\n Step 3: Move to GPU 0...")