Merge branch 'GestaltCogTeam:master' into master

Author: superarthurlx

tutorial/training_with_BLAST.md

@@ -2,7 +2,8 @@
 
 ## 1  BLAST Dataset Overview
 
-The [BLAST](https://arxiv.org/abs/2505.17871) dataset is a large-scale time-series corpus created specifically for **Universal Forecasting Models**. Thanks to its rich and diverse samples, BLAST enables **faster convergence**, **notable reductions in computational cost**, and superior performance even with limited resources.  
+The [BLAST](https://arxiv.org/abs/2505.17871) dataset is a large-scale time-series corpus created specifically for **Universal Forecasting Models**. Thanks to its rich and diverse samples, BLAST enables **faster convergence**, **notable reductions in computational cost**, and superior performance even with limited resources. The BLAST data is now available at [huggingface](https://huggingface.co/datasets/ZezhiShao/BLAST).
+
 BasicTS provides native support for BLAST and can be used to train models such as **TimeMoE** (decoder-only architecture) and **ChronosBolt** (encoder-decoder architecture). Before you start, follow the BasicTS [README](../README.md) to clone the codebase and set up the environment.
 
 ## 2  Data Preparation
@@ -124,4 +125,4 @@ BasicTS currently supports training **TimeMoE** and **ChronosBolt** on BLAST.
 > [!TIP]
 > - The default config is tuned for BLAST (batch size, learning rate, and other hyper-parameters). Adjust them if your GPUs or hardware differ.  
 
-If you encounter any issues, feel free to open an issue on [BasicTS Issues](https://github.com/GestaltCogTeam/BasicTS/issues). Happy training!
+If you encounter any issues, feel free to open an issue on [BasicTS Issues](https://github.com/GestaltCogTeam/BasicTS/issues). Happy training!