Add ROLL Docs #93
Add ROLL Docs #93
Conversation
|
Hi @fffrog @noemotiovon @lowdy1, please take a look, thanks! |
| ${custom_env.FrozenLake} | ||
| FrozenLakeThink: | ||
| ${custom_env.FrozenLakeThink} | ||
|
|
There was a problem hiding this comment.
In my opinion, it would be better to make as few changes as possible to the official configuration file. Some parameters should remain as they were.
There was a problem hiding this comment.
Aside from the required backend adaptations, I only adjusted max_steps, batch sizes and related size parameters. These changes are only for the quick-start example—the goal is simply to make the pipeline runnable on common hardware and finish quickly.
There was a problem hiding this comment.
Maybe that is acceptable. Also, please note that Ascend does not currently support colocate mode, and the device_mapping should be updated accordingly.
| 请根据已有昇腾产品型号及 CPU 架构等按照 :doc:`快速安装昇腾环境指引 <../ascend/quick_install>` 进行昇腾环境安装。 | ||
|
|
||
| .. warning:: | ||
| CANN 最低版本为 8.2.RC1，安装 CANN 时，请同时安装 Kernel 算子包以及 nnal 加速库软件包。 |
There was a problem hiding this comment.
Is this limitation from vLLM-Ascend?
|
|
||
| RuntimeError: CMake configuration failed: Command '['/pathto/miniconda3/envs/roll/bin/python3.10', '-m', 'pybind11', '--cmake']' returned non-zero exit status 2. | ||
|
|
||
| 可尝试在 vllm-ascend 目录下 setup.py 文件 151-158 行进行如下修改并重新进行编译： |
There was a problem hiding this comment.
Currently, installing vLLM version 0.8.4 is a bit complicated. Later, once version 0.11 is supported, we’ll also update the documentation.
| - DeepSpeed | ||
| - vLLM | ||
| - Atlas 900 A2 PODc | ||
| * - RLVR |
There was a problem hiding this comment.
Please check the VLM script when you have time.
4 participants
This PR Add ROLL Docs with Ascend NPU.