Update README.md

README.md

@@ -2,20 +2,25 @@
 license: apache-2.0
 library_name: tempo-pfn
 tags:
-- time-series-forecasting
-- zero-shot
-- rnn
-- linear-rnn
-- synthetic-data
-- foundation-model
-- automl
-- time-series
+  - TempoPFN
+  - time-series
+  - forecasting
+  - time-series-forecasting
+  - foundation models
+  - pretrained models
+  - zero-shot
+  - synthetic-data
+  - rnn
+  - linear-rnn
+leaderboards:
+  - Salesforce/GIFT-Eval
+pipeline_tag: time-series-forecasting
 arxiv: 2510.25502
 ---
 
 # TempoPFN: Synthetic Pre-Training of Linear RNNs for Zero-Shot Time Series Forecasting
 
-[![arXiv](https://img.shields.io/badge/arXiv-2510.25502-b31b1b.svg)](https://arxiv.org/abs/2510.25502) [![License](https://img.shields.io/badge/License-Apache_2.0-green.svg)](https://github.com/automl/TempoPFN/blob/main/LICENSE)
+[![preprint](https://img.shields.io/static/v1?label=Paper&message=2510.25502&color=B31B1B&logo=arXiv)](https://arxiv.org/abs/2510.25502) [![GIFT-Eval](https://img.shields.io/badge/%F0%9F%8F%86%20GIFT--Eval-Leaderboard-0078D4)](https://huggingface.co/spaces/Salesforce/GIFT-Eval) [![github](https://img.shields.io/badge/%F0%9F%92%BB%20GitHub-Repo-grey)](https://github.com/automl/TempoPFN) [![License: Apache 2.0](https://img.shields.io/badge/License-Apache%202.0-green.svg)](https://github.com/automl/TempoPFN/blob/main/LICENSE)
 
 ---
 
@@ -52,49 +57,41 @@ This repository includes the **pretrained 38M parameter model** (`models/checkpo
 
 2.  **Set up the environment:**
 ```bash
-    python -m venv venv && source venv/bin/activate
+python -m venv venv && source venv/bin/activate
 
-    # 1. Install PyTorch version matching your CUDA version
-    # Example for CUDA 12.8:
-    pip install torch --index-url https://download.pytorch.org/whl/cu128
+# 1. Install PyTorch version matching your CUDA version
+# Example for CUDA 12.8:
+pip install torch --index-url https://download.pytorch.org/whl/cu128
 
-    # 2. Install TempoPFN and all other dependencies
-    pip install -r requirements.txt
-    export PYTHONPATH=$PWD
-```
+# 2. Clone the Hugging Face repository
+git clone https://huggingface.co/AutoML-org/TempoPFN
+cd TempoPFN
 
-## 🚀 Quick Start: Run the Demo
+# 3. Set up the environment
+python3.12 -m venv venv & source venv/bin/activate
+export PYTHONPATH=$PWD
 
-**Prerequisites:**
-* You must have a **CUDA-capable GPU** with a matching PyTorch version installed.
-* You have run `export PYTHONPATH=$PWD` from the repo's root directory (see Installation).
+# 4. Install PyTorch version matching your CUDA version
+pip install torch --index-url https://download.pytorch.org/whl/cu128
 
-### 1. Run the Quick Start Script
+# 5. Install dependencies
+pip install .
+pip install .[dev]
 
-Run a demo forecast on a synthetic sine wave. This script will automatically find and load the `models/checkpoint_38M.pth` file included in this repository.
-```bash
+# 4. Run the Quick Start Script 
 python examples/quick_start_tempo_pfn.py
-```
-
-### 2. Run with a Different Checkpoint (Optional)
-
-If you have trained your own model, you can point the script to it:
-```bash
-python examples/quick_start_tempo_pfn.py --checkpoint /path/to/your/checkpoint.pth
-```
 
-### 3. Run the Notebook version
-```bash
+# 5. Alternatively, you can run the Notebook version
 jupyter notebook examples/quick_start_tempo_pfn.ipynb
 ```
 
 ### Hardware & Performance Tips
 
-**GPU Required:** Inference requires a CUDA-capable GPU. Tested on NVIDIA A100/H100.
+**GPU Required:** Inference requires a CUDA-capable GPU with a matching PyTorch version installed. Tested on NVIDIA A100/H100.
 
-**First Inference May Be Slow:** Initial calls for unseen sequence lengths trigger Triton kernel compilation. Subsequent runs are cached and fast.
+**First Run:** The first inference for a new sequence length will be slow due to Triton kernel compilation. Subsequent runs will be fast.
 
-**Triton Caches:** To prevent slowdowns from writing caches to a network filesystem, route caches to a local directory (like `/tmp`) before running:
+**Cache Tip:** If using a network filesystem, prevent slowdowns by routing caches to a local directory (like `/tmp`) *before* running:
 ```bash
 LOCAL_CACHE_BASE="${TMPDIR:-/tmp}/tsf-$(date +%s)"
 mkdir -p "${LOCAL_CACHE_BASE}/triton" "${LOCAL_CACHE_BASE}/torchinductor"
@@ -106,21 +103,16 @@ python examples/quick_start_tempo_pfn.py
 
 ## 🚂 Training
 
-### Single-GPU Training (for debugging)
+All training and model parameters are controlled via YAML files in `configs/`.  
+
 ```bash
+# Single-GPU (Debug)
 torchrun --standalone --nproc_per_node=1 src/training/trainer_dist.py --config ./configs/train.yaml
-```
 
-### Multi-GPU Training (Single-Node)
-
-This example uses 8 GPUs. The training script uses PyTorch DistributedDataParallel (DDP).
-```bash
+# Multi-GPU (e.g., 8 GPUs)
 torchrun --standalone --nproc_per_node=8 src/training/trainer_dist.py --config ./configs/train.yaml
 ```
 
-### Configuration
-
-All training and model parameters are controlled via YAML files in `configs/` (architecture, optimizers, paths).
 
 ## 💾 Synthetic Data Generation