skyrl-agent osworld integration

[shuo-yuan]

No description provided.

[gemini-code-assist]

Code Review

The code changes include adding a bash script for Tinker RL training for the MemAgent task in the OSWorld environment and a YAML configuration file for the OSWorld task. The review comments point out the hardcoded absolute paths for datasets and the output directory in the bash script, suggesting the use of environment variables with sensible defaults to make the script more portable.

[gemini-code-assist]

The script contains hardcoded absolute paths for datasets and the output directory, which include a specific user's home directory (/home/ubuntu/shuo). This makes the script not portable and difficult for other users to run without modification.

It's a best practice to avoid hardcoding user-specific paths. You can make these paths configurable, for example, by using environment variables with sensible defaults, similar to how other variables like MODEL_NAME are handled in this script.

Suggested change

	DATASET_FILE="/home/ubuntu/shuo/osworld/OSWorld_llm_agentsynth/osworld_train_8.parquet"
	EVAL_DATASET_FILE="/home/ubuntu/shuo/osworld/OSWorld_llm_agentsynth/osworld_train_8.parquet"
	# Output directory
	NAME="${NAME:-jan03_qwen3_8b_osworld_tinker_lr4e_5_rank128}"
	OUTPUT_DIR="/home/ubuntu/shuo/osworld/checkpoints/${NAME}"
	DATASET_FILE="${DATASET_FILE:-/path/to/your/osworld_train_8.parquet}"
	EVAL_DATASET_FILE="${EVAL_DATASET_FILE:-/path/to/your/osworld_train_8.parquet}"
	# Output directory
	NAME="${NAME:-jan03_qwen3_8b_osworld_tinker_lr4e_5_rank128}"
	OUTPUT_DIR="${OUTPUT_DIR:-/path/to/your/checkpoints/${NAME}}"