Update README.md

README.md

@@ -40,17 +40,17 @@ Timer-S1 is a time series foundation model with 8.3B total parameters, 0.75B act
 
 ![image](https://cdn-uploads.huggingface.co/production/uploads/64fbe24a2d20ced4e91de38a/7Udz1nO2V1Nk0pw5cW4gG.png)
 
-**Features**: Timer-S1 is a decoder-only Transformer consisting of **TimeMoE** and **TimeSTP** blocks. TimeSTP enables **Serial Forecasting**. It refers to initial input series, iteratively reuses learned representations from the previous block, and generates multi-step predictions without rolling-style autoregression. The model used a shared forecasting head to produce quantile forecasts. 
+**Architecture**: Timer-S1 is a decoder-only Transformer consisting of **TimeMoE** and **TimeSTP** blocks. TimeSTP enables **Serial Forecasting**. It refers to initial input series, iteratively reuses representations from the previous block, and generates multi-step predictions without rolling-style autoregression. The model used a shared forecasting head to produce quantile forecasts. 
 
 ![image](https://cdn-uploads.huggingface.co/production/uploads/64fbe24a2d20ced4e91de38a/1XsUZDPw8DJebZwH-Ievd.png)
 
-**Performance**: Timer-S1 achieves **1st CRPS/MASE** scores on [GIFT-Eval](https://huggingface.co/spaces/Salesforce/GIFT-Eval). The model is particularly good at medium-term and long-term forecasting tasks.
+**Performance**: Timer-S1 achieves **1st CRPS/MASE** (2026.03) scores on [GIFT-Eval](https://huggingface.co/spaces/Salesforce/GIFT-Eval). The model is particularly good at medium-term and long-term forecasting tasks.
 
 ![image](https://cdn-uploads.huggingface.co/production/uploads/64fbe24a2d20ced4e91de38a/XDOekWBIGBoc8nTDI-WBI.png)
 
 ![image](https://cdn-uploads.huggingface.co/production/uploads/64fbe24a2d20ced4e91de38a/r7eGVKBIRI8h7lMre4-lP.png)
 
-**Inference Speed**: Due to its sparse MoE configuration and serial-forecasting style, Timer-S1 has a fast inference speed (compared to NTP: Next-Token Prediction, MTP: Multi-Token Prediction). 
+**Inference Speed**: Due to sparse MoE configuration and serial-forecasting pipeline, Timer-S1 has a fast inference speed (compared to NTP: Next-Token Prediction, MTP: Multi-Token Prediction). 
 
 ![image](https://cdn-uploads.huggingface.co/production/uploads/683a85ec73d7c4f85b8fe9c1/HfsnFr_Mn8UPLD_wcfLPn.png)