summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--CHANGELOG.md11
-rw-r--r--setup.py4
2 files changed, 13 insertions, 2 deletions
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 40467dc..a26a0e7 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -106,3 +106,14 @@ Bug fixes:
- fixed an import of bnb.utils 2e630b55f51d454f3bd723dffda68a07ef93190c
We thank @mryab, @mbrukman, @chessgecko, @dbaranchuk for pull request with bug fixes and new features.
+
+
+### 0.34.0
+
+#### Bug fixes and memory efficient backprop
+
+Features:
+ - Linear8bitLt layer now supports `memory_efficient_backward=True` which enables backprop of gradients through frozen weights.
+
+Bug fixes:
+ - fixed an issue where too many threads were created in blockwise quantization on the CPU for large tensors
diff --git a/setup.py b/setup.py
index 97b99d0..610684b 100644
--- a/setup.py
+++ b/setup.py
@@ -18,13 +18,13 @@ def read(fname):
setup(
name=f"bitsandbytes",
- version=f"0.33.1",
+ version=f"0.34.0",
author="Tim Dettmers",
author_email="dettmers@cs.washington.edu",
description="8-bit optimizers and matrix multiplication routines.",
license="MIT",
keywords="gpu optimizers optimization 8-bit quantization compression",
- url="http://packages.python.org/bitsandbytes",
+ url="https://github.com/TimDettmers/bitsandbytes",
packages=find_packages(),
entry_points={
"console_scripts": ["debug_cuda = bitsandbytes.debug_cli:cli"],