github-starred/TinyTorch
2
0
Fork 0
mirror of https://github.com/MLSysBook/TinyTorch.git synced 2026-04-29 02:00:30 -05:00
Code Issues 7 Packages Projects Releases Wiki Activity

Clean up formatting in verification functions

Browse Source
This commit is contained in:
Vijay Janapa Reddi
2025-12-05 12:12:38 -08:00
parent a5bfffe48f
commit 3aa6a9b040
3 changed files with 13 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
modules/15_quantization/quantization.py
View File

@@ -1361,18 +1361,17 @@ def verify_quantization_works(original_model, quantized_model):
print("🔬 Verifying actual memory reduction with .nbytes...")
# Collect actual bytes from original FP32 model
original_bytes = 0
for param in original_model.parameters():
if hasattr(param, 'data') and hasattr(param.data, 'nbytes'):
original_bytes += param.data.nbytes
original_bytes = sum(
param.data.nbytes for param in original_model.parameters()
if hasattr(param, 'data') and hasattr(param.data, 'nbytes')
)
# Collect actual bytes from quantized INT8 model
quantized_bytes = 0
for layer in quantized_model.layers:
if isinstance(layer, QuantizedLinear):
quantized_bytes += layer.q_weight.data.nbytes
if layer.q_bias is not None:
quantized_bytes += layer.q_bias.data.nbytes
quantized_bytes = sum(
layer.q_weight.data.nbytes + (layer.q_bias.data.nbytes if layer.q_bias is not None else 0)
for layer in quantized_model.layers
if isinstance(layer, QuantizedLinear)
)
# Calculate actual reduction
actual_reduction = original_bytes / max(quantized_bytes, 1)