elixir-chatbot
所属分类:自然语言处理
开发工具:Elixir
文件大小:0KB
下载次数:0
上传日期:2023-11-18 14:45:13
上 传 者:
sh-1993
说明: 查特博特长生不老药
(elixir chatbot)
文件列表:
.formatter.exs (95, 2023-12-27)
.tool-versions (52, 2023-12-27)
apps/ (0, 2023-12-27)
apps/chat_web/ (0, 2023-12-27)
apps/chat_web/.formatter.exs (146, 2023-12-27)
apps/chat_web/assets/ (0, 2023-12-27)
apps/chat_web/assets/css/ (0, 2023-12-27)
apps/chat_web/assets/css/app.css (349, 2023-12-27)
apps/chat_web/assets/js/ (0, 2023-12-27)
apps/chat_web/assets/js/app.js (1777, 2023-12-27)
apps/chat_web/assets/tailwind.config.js (2410, 2023-12-27)
apps/chat_web/assets/vendor/ (0, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/ (0, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/LICENSE.md (1075, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/UPGRADE.md (347, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/ (0, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/ (0, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/ (0, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/academic-cap.svg (1010, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/adjustments-horizontal.svg (576, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/adjustments-vertical.svg (578, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/archive-box-arrow-down.svg (456, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/archive-box-x-mark.svg (501, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/archive-box.svg (359, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/arrow-down-circle.svg (330, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/arrow-down-left.svg (304, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/arrow-down-on-square-stack.svg (569, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/arrow-down-on-square.svg (385, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/arrow-down-right.svg (250, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/arrow-down-tray.svg (454, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/arrow-down.svg (317, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/arrow-left-circle.svg (475, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/arrow-left-on-rectangle.svg (628, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/arrow-left.svg (319, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/arrow-long-down.svg (309, 2023-12-27)
apps/chat_web/assets/vendor/heroicons/optimized/20/solid/arrow-long-left.svg (312, 2023-12-27)
... ...
# ElixirChatbot
## Using EXLA with the GPU
This project uses the EXLA backend for Nx. While it has support for computing on GPU, it will not be able to do so without proper configuration. This section will explain how to get EXLA running on a CUDA capable GPU. For further reading (for example, running the app on other architectures, there seems to be an experimental support for ROCm among others), refer to the [EXLA documentation](https://github.com/elixir-nx/nx/tree/main/exla) and [XLA binaries documentation](https://github.com/elixir-nx/xla).
### Find out what CUDA version your GPU supports
You can find it out either with nVidia's GUI tools or with the `nvidia-smi` command.
### Download CUDA and cuDNN libraries
The official releases are available here: [CUDA](https://developer.nvidia.com/cuda-downloads), [cuDNN](https://developer.nvidia.com/cudnn). For many platforms, there are special considerations and configuration needed - refer to online documentation for there.
For platforms which are not oficially supported (such as Arch Linux or NixOS) there are community packages available.
### Configure XLA to use correct binaries
If prebuilt XLA bineries for the CUDA version your GPU supports is also available on [XLA](https://github.com/elixir-nx/xla), you need to export several environment variables. Before doing so, make sure there are no old versions of binaries:
```shell
mix deps.clean xla exla
```
Then, export these variables with correct values. Refer to the documentation to find out what values your platform needs.
```shell
export CUDA_HOME="/opt/cuda" # path to where the CUDA libraries have been installed to
export PATH="/opt/cuda/bin:$PATH" # possibly optional
export XLA_TARGET=cuda120 # depending on the CUDA version on your GPU
export ELIXIR_ERL_OPTIONS="+sssdio 128" # grant additional stack space for the XLA compiler
export TF_CUDA_VERSION="12.2" # possibly optional, couldn't get it to work without it, though. Depends on CUDA version
export XLA_FLAGS=--xla_gpu_cuda_data_dir=/opt/cuda #path to where CUDA libraries have been installed to
```
After restarting the shell or sourcing appropriate files to update the environment, download and build the libraries again:
```shell
mix deps.get
mix deps.compile
```
You can check whether it works by launhing the `ElixirChatbotCore` app in interactive mode:
```shell
iex -S mix
```
```iex
iex(1)> Nx.tensor([1, 2, 3])
#Nx.Tensor<
s64[3]
EXLA.Backend
[1, 2, 3]
>
```
近期下载者:
相关文件:
收藏者: