mrq/bitsandbytes-rocm
1
0
Fork 1
Code Issues 1 Pull Requests Packages Projects Releases Wiki Activity

Bumped version.

Browse Source
This commit is contained in:
Tim Dettmers 2023-05-23 19:55:52 -07:00
parent 1b8772a8f3
commit 0f40fa3f0a
3 changed files with 14 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
CHANGELOG.md
View File

@ -228,3 +228,14 @@ Deprecated:
Features: Features:
- Added Int8 SwitchBack layers - Added Int8 SwitchBack layers
- Added Fake FP8 layers for research purposes (available under `bnb.research.nn. ...`) - Added Fake FP8 layers for research purposes (available under `bnb.research.nn. ...`)
### 0.39.0
Features:
- 4-bit matrix multiplication for Float4 and NormalFloat4 data types.
- Added 4-bit quantization routines
- Doubled quantization routines for 4-bit quantization
- Paged optimizers for Adam and Lion.
- bfloat16 gradient / weight support for Adam and Lion with 8 or 32-bit states.

3
Makefile
View File

@ -25,8 +25,7 @@ FILES_CUDA := $(CSRC)/ops.cu $(CSRC)/kernels.cu
FILES_CPP := $(CSRC)/common.cpp $(CSRC)/cpu_ops.cpp $(CSRC)/pythonInterface.c FILES_CPP := $(CSRC)/common.cpp $(CSRC)/cpu_ops.cpp $(CSRC)/pythonInterface.c
INCLUDE := -I $(CUDA_HOME)/include -I $(ROOT_DIR)/csrc -I $(CONDA_PREFIX)/include -I $(ROOT_DIR)/include INCLUDE := -I $(CUDA_HOME)/include -I $(ROOT_DIR)/csrc -I $(CONDA_PREFIX)/include -I $(ROOT_DIR)/include
INCLUDE_10x := -I $(CUDA_HOME)/include -I $(ROOT_DIR)/csrc -I $(ROOT_DIR)/dependencies/cub -I $(ROOT_DIR)/include LIB := -L $(CUDA_HOME)/lib64 -lcudart -lcublas -lcublasLt -lcusparse -L $(CONDA_PREFIX)/lib
LIB := -L $(CUDA_HOME)/lib64 -lcudart -lcublas -lcublasLt -lcurand -lcusparse -L $(CONDA_PREFIX)/lib
# NVIDIA NVCC compilation flags # NVIDIA NVCC compilation flags
COMPUTE_CAPABILITY += -gencode arch=compute_50,code=sm_50 # Maxwell COMPUTE_CAPABILITY += -gencode arch=compute_50,code=sm_50 # Maxwell

4
setup.py
View File

@ -18,10 +18,10 @@ def read(fname):
setup( setup(
name=f"bitsandbytes", name=f"bitsandbytes",
version=f"0.38.1", version=f"0.39.0",
author="Tim Dettmers", author="Tim Dettmers",
author_email="dettmers@cs.washington.edu", author_email="dettmers@cs.washington.edu",
description="8-bit optimizers and matrix multiplication routines.", description="k-bit optimizers and matrix multiplication routines.",
license="MIT", license="MIT",
keywords="gpu optimizers optimization 8-bit quantization compression", keywords="gpu optimizers optimization 8-bit quantization compression",
url="https://github.com/TimDettmers/bitsandbytes", url="https://github.com/TimDettmers/bitsandbytes",