UPLOAD.md

Upload Guide

This release is ready up to the pre-upload stage. The local package is:

/home/ps/sjj/project/Memory/dataset/general/general_science_release.tar.gz

The folder to upload is:

/home/ps/sjj/project/Memory/dataset/general/general_science_release

1. Pre-upload Check

Run:

cd /home/ps/sjj/project/Memory/dataset/general/general_science_release
conda run -n memory python scripts/validate_general_science_release.py

Expected result:

validation=ok

2. Hugging Face Upload

Option A: Web Upload

1. Create a new dataset repository on Hugging Face.
2. Suggested repo name:

gineven/GeneralScience-MLLM-22K

3. Upload the contents of general_science_release/.
4. Keep the folder structure unchanged:

README.md
general_science_card.md
stats.json
train.jsonl
test.jsonl
images/scienceqa/*.png
scripts/*.py

Option B: CLI Upload

Install or update the HF hub client if needed:

pip install -U huggingface_hub

Login:

hf auth login

Create the dataset repo:

# You already created this repository:
# https://huggingface.co/datasets/gineven/GeneralScience-MLLM-22K

Upload the folder:

cd /home/ps/sjj/project/Memory/dataset/general
hf upload gineven/GeneralScience-MLLM-22K general_science_release . \
  --repo-type dataset \
  --exclude "*/__pycache__/*" \
  --exclude "*.pyc"

If your installed CLI uses the older command name, use:

huggingface-cli login
# Repository already exists:
# https://huggingface.co/datasets/gineven/GeneralScience-MLLM-22K

Then upload with Python API as shown below.

Option C: Python API Upload

from huggingface_hub import HfApi

api = HfApi()
repo_id = "gineven/GeneralScience-MLLM-22K"

api.create_repo(repo_id=repo_id, repo_type="dataset", exist_ok=True)
api.upload_folder(
    repo_id=repo_id,
    repo_type="dataset",
    folder_path="/home/ps/sjj/project/Memory/dataset/general/general_science_release",
    path_in_repo=".",
    ignore_patterns=["*/__pycache__/*", "*.pyc"],
)

After upload, the expected download link is:

https://huggingface.co/datasets/gineven/GeneralScience-MLLM-22K

3. ModelScope Upload

ModelScope upload is similar in spirit: create a dataset repository, then upload the same folder contents. Keep README.md, train.jsonl, test.jsonl, stats.json, and images/scienceqa/ at the repository root.

If using modelscope CLI/API, first make sure your account is logged in and then upload the general_science_release/ directory as a dataset repository.

4. Netdisk Fallback

If HF/ModelScope credentials are not available, upload:

general_science_release.tar.gz

Then share the netdisk URL together with:

general_science_release/general_science_card.md
general_science_release/stats.json

5. Important License Check

Before public upload, manually verify the upstream ScienceQA license because the local snapshot used for this build does not include license metadata. The current README.md marks the combined license as other and records the missing ScienceQA license note.