Quick Start Guide for ROCm Support#293
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces a new quick start guide for AMD ROCm, detailing environment setup, model and dataset downloading, weight conversion, and training. The feedback suggests improving the guide's accuracy and usability by using the standard uppercase $WANDB_API_KEY environment variable in the Docker command and replacing the non-standard hf command with huggingface-cli for downloading assets.
| docker run -d --name vime --ulimit nofile=1048576:1048576 \ | ||
| --ipc=host --network=host --device=/dev/kfd --device=/dev/dri \ | ||
| --security-opt seccomp=unconfined --group-add video --privileged \ | ||
| -e WANDB_API_KEY=$wandb_key vllm/vime-rocm |
There was a problem hiding this comment.
Using the standard uppercase environment variable $WANDB_API_KEY is more idiomatic and aligns with how users typically store and export their Weights & Biases API keys in their environment.
| -e WANDB_API_KEY=$wandb_key vllm/vime-rocm | |
| -e WANDB_API_KEY=$WANDB_API_KEY vllm/vime-rocm |
| hf download Qwen/Qwen3-8B --local-dir /root/Qwen3-8B | ||
|
|
||
| # Download training dataset (dapo-math-17k) | ||
| hf download zhuzilin/dapo-math-17k --repo-type dataset --local-dir /root/dapo-math-17k |
There was a problem hiding this comment.
The standard command-line tool provided by the huggingface_hub library is huggingface-cli, not hf. Using huggingface-cli download ensures the commands run successfully without requiring custom shell aliases.
| hf download Qwen/Qwen3-8B --local-dir /root/Qwen3-8B | |
| # Download training dataset (dapo-math-17k) | |
| hf download zhuzilin/dapo-math-17k --repo-type dataset --local-dir /root/dapo-math-17k | |
| huggingface-cli download Qwen/Qwen3-8B --local-dir /root/Qwen3-8B | |
| # Download training dataset (dapo-math-17k) | |
| huggingface-cli download zhuzilin/dapo-math-17k --repo-type dataset --local-dir /root/dapo-math-17k |
1 participant
No description provided.