Use direct links for dataset files and benchmark guide

README.md

@@ -22,8 +22,8 @@ configs:
 
 A 15-scene development benchmark for [Code2Worlds-2](https://github.com/AIGeeksGroup/Code2Worlds-2), spanning indoor, industrial, natural outdoor, service, and urban tasks.
 
-- [Scene index](data/scenes.jsonl): case IDs, task prompts, and links to the corresponding runtime models in GitHub. The **Dataset Viewer** displays these 15 validation cases.
-- [Blender sources](blender_sources): 12 original `.blend` projects. The six natural and service scenes each include three animation preview frames beside the project file.
+- [Scene index](https://huggingface.co/datasets/AIGeeksGroup/SimCode4D/blob/main/data/scenes.jsonl): case IDs, task prompts, and links to the corresponding runtime models in GitHub. The **Dataset Viewer** displays these 15 validation cases.
+- [Blender sources](https://huggingface.co/datasets/AIGeeksGroup/SimCode4D/tree/main/blender_sources): 12 original `.blend` projects. The six natural and service scenes each include three animation preview frames beside the project file.
 - [Benchmark code and runtime](https://github.com/AIGeeksGroup/Code2Worlds-2/tree/main/benchmark/simcode4d_unified): all 15 executable MuJoCo worlds, required assets, frozen case definitions, validators, metrics, and submission scorer. Follow the [benchmark guide](https://github.com/AIGeeksGroup/Code2Worlds-2/blob/main/benchmark/simcode4d_unified/README.md) to run an evaluation.
 
 The runtime files are hosted in GitHub; this dataset adds the authoring scenes and visual previews. See the [third-party notices](https://github.com/AIGeeksGroup/Code2Worlds-2/blob/main/THIRD_PARTY_NOTICES.md) for the runtime assets.