some kind of warning or something when this is first executed to make people aware that a cast happens and the operation quantization is performed in fp16.

1 files changed, 4 insertions, 0 deletions

diff --git a/bitsandbytes/autograd/_functions.py b/bitsandbytes/autograd/_functions.py
index bdcbec5..6d473e9 100644
--- a/bitsandbytes/autograd/_functions.py
+++ b/bitsandbytes/autograd/_functions.py
@@ -1,4 +1,6 @@
 import operator
+import warnings
+
 import torch
 import bitsandbytes.functional as F
 
@@ -229,6 +231,8 @@ class MatMul8bitLt(torch.autograd.Function):
 
         # Cast A to fp16
         A_dtype = A.dtype
+        if A_dtype != torch.float16:
+            warnings.warn(f"MatMul8bitLt: temporarily casting input matrix from {A_dtype} to float16")
         A = A.to(torch.float16)
 
         # 1. Quantize A