Update README.md

README.md

@@ -19,8 +19,8 @@ Beyond Token-Level Cross-Entropy: Fréchet Distributional Post-Training for Auto
 </p>
 
 <p align="center">
-  <a href="https://arxiv.org/abs/2608.00562">
-    <img src="https://img.shields.io/badge/arXiv-2608.00562-b31b1b.svg" alt="arXiv">
+  <a href="https://arxiv.org/abs/2505.12742v3">
+    <img src="https://img.shields.io/badge/arXiv-2505.12742-b31b1b.svg" alt="arXiv">
   </a>
   <a href="https://github.com/CVL-UESTC/FDPT-AR">
     <img src="https://img.shields.io/github/stars/CVL-UESTC/FDPT-AR?style=social" alt="GitHub stars">
@@ -83,8 +83,6 @@ bash scripts/download_assets.sh llamagen
 bash scripts/download_assets.sh stats
 ```
 
-Replace `llamagen` with `titok`, `var`, or `gigatok` when needed.
-
 ---
 
 ## 🤗 Post-trained Checkpoints
@@ -125,9 +123,6 @@ using `scripts/download_assets.sh`.
 
 ## 🔥 Training
 
-Only the model name, pretrained checkpoint directory, and per-GPU batch size
-need to be specified:
-
 ```bash
 bash scripts/train.sh \
   --model llamagen-b \
@@ -155,8 +150,7 @@ Available models and their required checkpoint files are listed below.
 | VAR-d24 | `var-d24` | `var_d24.pth`, `vae_ch160v4096z32.pth` |
 | GigaTok-B-L | `gigatok-bl` | `GPT_B256_e300_VQ_BL.pt`, `VQ_BL256_e200.pt` |
 
-The scripts automatically use all visible GPUs. To select specific GPUs, set
-`CUDA_VISIBLE_DEVICES` before running the command:
+To select specific GPUs, set `CUDA_VISIBLE_DEVICES` before running the command:
 
 ```bash
 CUDA_VISIBLE_DEVICES=0,1 bash scripts/train.sh \
@@ -165,8 +159,6 @@ CUDA_VISIBLE_DEVICES=0,1 bash scripts/train.sh \
   --bs 8
 ```
 
-Here, `--bs` denotes the batch size on each GPU.
-
 ---
 
 ## 🖼️ Sampling
@@ -234,4 +226,4 @@ Upstream-derived files retain their original licenses. See
 
 For questions or collaborations, please contact
 [Jinhua Zhang](mailto:jinhua.zjh@gmail.com) or
-[Yisong Lin](mailto:yisongl164@gmail.com).
+[Yisong Lin](mailto:yisongl164@gmail.com).