Yevhenii Hordashnyk

I can optimize the Dockerfile and create a PR in a day or so. Thanks for sharing it @nxdong

I wasn't able to mount the `AWS_WEB_IDENTITY_TOKEN_FILE` file because of the directory structure it has. `AWS_WEB_IDENTITY_TOKEN_FILE` is a symlink and docker does not follow them during the mount. So I...

I do have the same problem with the latest image, but was able to launch meilisync with older version

Additionally, based on this [commit](https://github.com/Cinnamon/kotaemon/blame/main/libs/ktem/ktem/index/file/graph/pipelines.py#L180), it seems like groups are intended not to work. Maybe the "select all" option is intended not to work as well?

I see the same exact behavior with GraphRAG. It works fine if I manually select a document, but it doesn't work with "Search All" option. File Collection is fine

> @vonhuy1 Please upgrade to 0.4.21 and then try, if same issue, please share the code snippet. I check it asap. Just an FYI - I was able to get...

@youkaichao 1. I'm able to successfully launch a non-quantized model with this setup 2. I did test a setup where ray head also has access to GPU, but it doesn't...

@youkaichao Using a suggestion from this [issue](https://github.com/vllm-project/vllm/issues/7890) I hardcoded a setting of CUDA_VISIBLE_DEVICES and was able to make it work. I can only assume that either vllm or ray overrides...

Unfortunately, I keep getting the same error even though ray cluster shows available GPUs

I noticed, that if i set `ray_actor_options.num_gpus: 0.95`, vllm will show that i have 0.05 GPU. `ValueError: Current node has no GPU available. current_node_resource={'memory': 36000000000.0, 'GPU': 0.05....}.` Meaning, Ray for...