Releases: likelovewant/ollama-for-amd
v0.5.9
- Detailed Installation Guide: please refer to the wiki guide:
Demo Release Version:
This releaseOllamaSetup.exebuild with `ROCm6.1.2 hipsdk 6.1.2， make sure to replace the libs with v0.6.1.2
Tip
ROCm5.7 version * The upstream llama.cpp broken the rocm5.7 , need update update hipsdk5.7 clang17 to clang19 in building process. More from wiki
ollama-windows-amd64-rocm5.7z ( build for gfx803 gfx900:xnack- gfx1103(test only))
ROCmlibs for 5.7 available at ROCmlibs for 5.7
ROCmlibs for 6.1.2 available at ROCmlibs for 6.1.2
Support lists gfx1010:xnack- gfx1011 gfx1012:xnack- gfx1030 gfx1031 gfx1032 gfx1034 gfx1035 gfx1100 gfx1101 gfx1103 gfx1150
Note
Install OllamaSetup.exe first in this release , 2nd, upzip ollama-windows-amd64.7z to replace the libs in C:\Users\usrname\AppData\Local\Programs\Ollama\lib\ollama , 3rd, replace files in your Ollama program ROCm folder with the rocblas.dll and rocblas/library folder matches your GPU architecture with the correct ROCmlibs for 6.1.2 or ROCmlibs for 5.7 for ollama-windows-amd64-rocm5.7z .(eg. the file in C:\Users\usrname\AppData\Local\Programs\Ollama\lib\ollama ( or you may skip step 1 ,remove any ollama clients on your local , simply use 7z package with step 2 and step3. by run ./ollama serve )
If there is error log show amdgpu is not supported (supported types:[gfx1030 gfx1100 gfx1101 gfx1102 gfx906] it's means you are missing some steps . please check the above steps or possible replace the rocmlibs in your hipsdk
Full Changelog: v0.5.8...v0.5.9
For a complete list of changes and bug fixes, please check ollama changelog:
ollama/releases
v0.5.8
Full Changelog: v0.5.4...v0.5.8
v0.5.4
- Detailed Installation Guide: please refer to the wiki guide:
Demo Release Version: - Add experimental support for gfx900:xnack-
This releaseOllamaSetup.exebuild with `ROCm6.1.2 hipsdk 6.1.2， make sure to replace the libs with v0.6.1.2
Tip
ROCm5.7 version *No need to update for this release
ROCmlibs for 6.1.2 available at ROCmlibs for 6.1.2
Support lists gfx803 (may not detected ) gfx900:xnack- gfx902 gfx90c:xnack- gfx906:xnack- gfx1010:xnack- gfx1011 gfx1012:xnack- gfx1030 gfx1031 gfx1032 gfx1034 gfx1035 gfx1036 gfx1100 gfx1101 gfx1103 gfx1150
ollama-windows-amd64-rocm5.7-v2.7z ( rebuild for gfx803 gfx900:xnack- gfx902 gfx906:xnack- gfx90c:xnack- gfx1010:xnack-)
Note
OllamaSetup.exe is same as ollama-windows-amd64.7z ,you don't need them both . * Refer to the Standalone CLI guide:
For a complete list of changes and bug fixes, please check ollama changelog:
ollama/releases
v0.5.1
- Detailed Installation Guide: please refer to the wiki guide:
Demo Release Version: - Add experimental support for gfx1150 （890M,880M）: please refer to the wiki guide:
This releaseOllamaSetup.exebuild with `ROCm6.1.2 hipsdk 6.1.2， make sure to replace the libs with v0.6.1.2
Tip
ROCm5.7 version *ollama-windows-amd64-rocm5.7.7z.
ROCmlibs for 6.1.2 available at ROCmlibs for 6.1.2
ROCmlibs for 5.7 available at ROCmlibs for 5.7
Support lists gfx803 gfx900 gfx902 gfx90c:xnack- gfx906:xnack- gfx90a:xnack- gfx1010:xnack- gfx1012:xnack- gfx1030 gfx1031 gfx1032 gfx1034 gfx1035 gfx1036 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150
OllamaSetup.exe is same as ollama-windows-amd64.7z ,you don't need them both . * Refer to the Standalone CLI guide:
For a complete list of changes and bug fixes, please check ollama changelog:
ollama/releases
v0.4.6
- Detailed Installation Guide: please refer to the wiki guide:
Demo Release Version:
This release OllamaSetup.exe build with `ROCm6.1.2 hipsdk 6.1.2， make sure to replace the libs with v0.6.1.2 otherwise ,it's may fail to detected
Note the drop replacement with ollama-windows-amd64 not working if have earlier version v0.4.0 installed .
ROCmlibs for 6.1.2 available at ROCmlibs for 6.1.2
If you encounter issues with certain GPUs being undetected by ROCm 6.1.2, It is recommend using ROCm 5.7 as an
alternative. You can download the appropriate version for your system: ollama-windows-amd64-rocm5.7.7z
ROCmlibs for 5.7 available at ROCmlibs for 5.7
Important
(update info for those who skip reading the Detailed Installation Guide , Demo Release Version: )
Install and Replace Files: Run the OllamaSetup.exe installer first, replace files in your Ollama program folder with the rocblas.dll and rocblas/library folder matches your GPU architecture with the correct ROCmlibs for 6.1.2.(eg. the file in `C:\Users\usrname\AppData\Local\Programs\Ollama\lib\ollama)
Support lists gfx803 gfx900 gfx902 gfx90c:xnack- gfx906:xnack- gfx90a:xnack- gfx1010:xnack- gfx1012:xnack- gfx1030 gfx1031 gfx1032 gfx1034 gfx1035 gfx1036 gfx1100 gfx1101 gfx1102 gfx1103 ( if you arches are not on the lists or multi-gpu , please build yourself with the guide available at wiki , or feel free to share you arches info by type hipinfo in terminal when you have hipsdk availalbe . then the arch lists can add it to the lists)
For a complete list of changes and bug fixes, please check ollama changelog:
ollama/releases
v0.4.4
- Detailed Installation Guide: please refer to the wiki guide:
Demo Release Version:
V0.4.4 is equal to V0.4.5 from ollama/releases
This release OllamaSetup.exe build with `ROCm6.1.2 hipsdk 6.1.2， make sure to replace the libs with v0.6.1.2 otherwise ,it's may fail to detected
Note the drop replacement with ollama-windows-amd64 not working if have earlier version v0.4.0 installed .
ROCmlibs for 6.1.2 available at ROCmlibs for 6.1.2
Important
(update info for those who skip reading the Detailed Installation Guide , Demo Release Version: )
Install and Replace Files: Run the OllamaSetup.exe installer first, replace files in your Ollama program folder with the rocblas.dll and rocblas/library folder matches your GPU architecture with the correct ROCmlibs for 6.1.2.(eg. the file in `C:\Users\usrname\AppData\Local\Programs\Ollama\lib\ollama)
Support lists gfx900 gfx902 gfx906:xnack- gfx90a:xnack- gfx1010:xnack- gfx1012:xnack- gfx1030 gfx1031 gfx1032 gfx1034 gfx1035 gfx1036 gfx1100 gfx1101 gfx1102 gfx1103 ( if you arches are not on the lists or multi-gpu , please build yourself with the guide available at wiki , or feel free to share you arches info by type hipinfo in terminal when you have hipsdk availalbe . then the arch lists can add it to the lists)
Tip
When a model partially loads onto shared VRAM, even with sufficient total VRAM, response times become excessively
slow. Reinstalling the graphics drivers typically resolves this issue.
Note
For GFX803 and GFX90C users, recommend using the earlier release bundled with ROCm 5.7. However, due to its
limited user and lower performance as reported by some gfx803 users, ... considering removing it from supported
options. As an alternative, could potentially downgrade the entire project from ROCm 6.1.2 to 5.7. Do you have
any other suggestions?
For a complete list of changes and bug fixes, please check ollama changelog:
ollama/releases
v0.4.2
- Detailed Installation Guide: please refer to the wiki guide:
Demo Release Version:
This release OllamaSetup.exe build with `ROCm6.1.2 hipsdk 6.1.2， make sure to replace the libs with v0.6.1.2 otherwise ,it's may fail to detected
Note the drop replacement with ollama-windows-amd64 not working if have earlier version v0.4.0 installed .
ROCmlibs for 6.1.2 available at ROCmlibs for 6.1.2
If you encounter issues with certain GPUs being undetected by ROCm 6.1.2, It is recommend using ROCm 5.7 as an
alternative. You can download the appropriate version for your system: ollama-windows-amd64-rocm5.7.7z
ROCmlibs for 5.7 available at ROCmlibs for 5.7
Important
(update info for those who skip reading the Detailed Installation Guide , Demo Release Version: )
Install and Replace Files: Run the OllamaSetup.exe installer first, replace files in your Ollama program folder with the rocblas.dll and rocblas/library folder matches your GPU architecture with the correct ROCmlibs for 6.1.2.(eg. the file in `C:\Users\usrname\AppData\Local\Programs\Ollama\lib\ollama)
Support lists gfx803 gfx900 gfx902 gfx90c:xnack- gfx906:xnack- gfx90a:xnack- gfx1010:xnack- gfx1012:xnack- gfx1030 gfx1031 gfx1032 gfx1034 gfx1035 gfx1036 gfx1100 gfx1101 gfx1102 gfx1103 ( if you arches are not on the lists or multi-gpu , please build yourself with the guide available at wiki , or feel free to share you arches info by type hipinfo in terminal when you have hipsdk availalbe . then the arch lists can add it to the lists)
For a complete list of changes and bug fixes, please check ollama changelog:
ollama/releases
v0.4.1
- Detailed Installation Guide: please refer to the wiki guide:
Demo Release Version:
This release OllamaSetup.exe build with ROCm5.7 （hipsdk 5.7） due to the upstream ollama breaking change， make sure to replace the libs with v0.5.7 otherwise ,it's may fail to detected
Note the drop replacement with ollama-windows-amd64 not working if have earlier version installed .
ROCmlibs for 5.7 available at ROCmlibs for 5.7
Important
(update info for those who skip reading the first line , the wiki guide )
Install and Replace Files: Run the OllamaSetup.exe installer first, replace files in your Ollama program folder with the rocblas.dll and rocblas/library folder matches your GPU architecture with the correct ROCmlibs for 5.7.(eg. the file in `C:\Users\usrname\AppData\Local\Programs\Ollama\lib\ollama)
Support lists for rocm5.7 gfx803 gfx900 gfx902 gfx90c:xnack- gfx940 gfx941 gfx942 gfx1010:xnack- gfx1012:xnack- gfx1030 gfx1031 gfx1032 gfx1034 gfx1035 gfx1036 gfx1100 gfx1101 gfx1102 gfx1103 ( if you arches are not on the lists or multi-gpu , please build yourself with the guide available at wiki , or feel free to share you arches info by type hipinfo in terminal when you have hipsdk availalbe . then the arch lists can add it to the lists)
For a complete list of changes and bug fixes, please check ollama changelog:
ollama/releases
v0.3.14
- Detailed Installation Guide: please refer to the wiki guide:
Demo Release Version:
For a complete list of changes and bug fixes, please check ollama changelog:
ollama/releases
v0.3.13
- Detailed Installation Guide: please refer to the wiki guide:
Demo Release Version:
For a complete list of changes and bug fixes, please check ollama changelog:
ollama/releases