third-party-mirrors/ollama
1
0
Fork 1
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix README.md

Browse Source
This commit is contained in:
jmorganca 2024-06-11 22:54:31 -07:00
parent 18662d1180
commit f1f54c5bd5
1 changed files with 28 additions and 65 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

93
llama/README.md
View File

@ -1,10 +1,6 @@
# `llama` # `llama`
<<<<<<< Updated upstream
This package integrates llama.cpp as a Go package that's easy to build with tags for different CPU and GPU processors.
=======
This package integrates the [llama.cpp](https://github.com/ggerganov/llama.cpp) library as a Go package and makes it easy to build it with tags for different CPU and GPU processors. This package integrates the [llama.cpp](https://github.com/ggerganov/llama.cpp) library as a Go package and makes it easy to build it with tags for different CPU and GPU processors.
>>>>>>> Stashed changes
Supported: Supported:
@ -16,15 +12,12 @@ Supported:
- [x] Linux CUDA - [x] Linux CUDA
- [x] Linux ROCm - [x] Linux ROCm
- [x] Llava - [x] Llava
<<<<<<< Updated upstream
=======
- [ ] Parallel Requests - [ ] Parallel Requests
Extra build steps are required for CUDA and ROCm on Windows since `nvcc` and `hipcc` both require using msvc as the host compiler. For these small dlls are created: Extra build steps are required for CUDA and ROCm on Windows since `nvcc` and `hipcc` both require using msvc as the host compiler. For these small dlls are created:
- `ggml-cuda.dll` - `ggml-cuda.dll`
- `ggml-hipblas.dll` - `ggml-hipblas.dll`
>>>>>>> Stashed changes
> Note: it's important that memory is allocated and freed by the same compiler (e.g. entirely by code compiled with msvc or mingw). Issues from this should be rare, but there are some places where pointers are returned by the CUDA or HIP runtimes and freed elsewhere, causing a a crash. In a future change the same runtime should be used in both cases to avoid crashes. > Note: it's important that memory is allocated and freed by the same compiler (e.g. entirely by code compiled with msvc or mingw). Issues from this should be rare, but there are some places where pointers are returned by the CUDA or HIP runtimes and freed elsewhere, causing a a crash. In a future change the same runtime should be used in both cases to avoid crashes.
@ -54,84 +47,54 @@ go build -tags=avx,avx2 .
### CUDA ### CUDA
<<<<<<< Updated upstream
Install the [CUDA toolkit v11.3.1](https://developer.nvidia.com/cuda-11-3-1-download-archive): Install the [CUDA toolkit v11.3.1](https://developer.nvidia.com/cuda-11-3-1-download-archive):
=======
Install the [CUDA toolkit v11.3.1](https://developer.nvidia.com/cuda-11-3-1-download-archive) then build `libggml-cuda.so`:
```shell ```shell
./build_cuda.sh make ggml_cuda.so
``` go build -tags=avx,cuda .
>>>>>>> Stashed changes ```
Then build the package with the `cuda` tag: ### ROCm
```shell Install the [CUDA toolkit v11.3.1](https://developer.nvidia.com/cuda-11-3-1-download-archive):
go build -tags=cuda .
``` ```shell
make ggml_hipblas.so
## Windows go build -tags=avx,rocm .
```
<<<<<<< Updated upstream
Download [w64devkit](https://github.com/skeeto/w64devkit/releases/latest) for a simple MinGW development environment. ## Windows
======= Download [w64devkit](https://github.com/skeeto/w64devkit/releases/latest) for a simple MinGW development environment.
>>>>>>> Stashed changes
### CUDA ### CUDA
Install the [CUDA toolkit v11.3.1](https://developer.nvidia.com/cuda-11-3-1-download-archive) then build the cuda code: Install the [CUDA toolkit v11.3.1](https://developer.nvidia.com/cuda-11-3-1-download-archive) then build the cuda code:
Build `ggml-cuda.dll`: ```shell
make ggml_cuda.dll
```shell go build -tags=avx,cuda .
<<<<<<< Updated upstream
make ggml_cuda.dll
=======
./build_cuda.ps1
>>>>>>> Stashed changes
```
Then build the package with the `cuda` tag:
```shell
go build -tags=cuda .
``` ```
### ROCm ### ROCm
<<<<<<< Updated upstream
Install [ROCm 5.7.1](https://rocm.docs.amd.com/en/docs-5.7.1/). Install [ROCm 5.7.1](https://rocm.docs.amd.com/en/docs-5.7.1/).
```shell ```shell
make ggml_hipblas.dll make ggml_hipblas.dll
======= go build -tags=rocm .
Install [ROCm 5.7.1](https://rocm.docs.amd.com/en/docs-5.7.1/) and [Strawberry Perl](https://strawberryperl.com/).
Then, build `ggml-hipblas.dll`:
```shell
./build_hipblas.sh
>>>>>>> Stashed changes
``` ```
Then build the package with the `rocm` tag: Then build the package with the `rocm` tag:
```shell ```shell
go build -tags=rocm . go build -tags=avx,rocm .
``` ```
## Syncing with llama.cpp ## Syncing with llama.cpp
<<<<<<< Updated upstream To update this package to the latest llama.cpp code, use the `sync_llama.sh` script:
To update this package to the latest llama.cpp code, use the `sync_llama.sh` script from the root of this repo:
``` ```
./sync_llama.sh ../../llama.cpp ../sync_llama.sh ../../llama.cpp
=======
To update this package to the latest llama.cpp code, use the `scripts/sync_llama.sh` script from the root of this repo, providing the location of a llama.cpp checkout:
```
cd ollama
./scripts/sync_llama.sh ../llama.cpp
>>>>>>> Stashed changes
``` ```