Fix GitHub URL, add leaderboard link, update paper abstract quote

README.md

@@ -15,39 +15,41 @@ tags:
   <p align="center">
     <a href="https://arxiv.org/abs/2601.16175"><img src="https://img.shields.io/badge/arXiv-2601.16175-b31b1b.svg" alt="arXiv"></a>
     <a href="https://test-time-training.github.io/discover/"><img src="https://img.shields.io/badge/Project-Page-blue" alt="Project Page"></a>
-    <a href="https://github.com/prannayHexo/discover"><img src="https://img.shields.io/badge/GitHub-discover-black?logo=github" alt="GitHub"></a>
+    <a href="https://github.com/test-time-training/discover"><img src="https://img.shields.io/badge/GitHub-discover-black?logo=github" alt="GitHub"></a>
     <a href="https://colab.research.google.com/github/Pran-Ker/gpu-mode-trimul/blob/main/gpu_mode_trimul.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"></a>
   </p>
 </p>
 
 LoRA adapter for **gpt-oss-120b**, trained with reinforcement learning on the
-[GPU Mode TriMul](https://github.com/gpu-mode) competition — triangular matrix multiplication on H100.
+[GPU Mode TriMul](https://www.gpumode.com/v2/leaderboard/496?tab=rankings) competition —
+triangular matrix multiplication on H100.
 
-Produced via **TTT-Discover** ([Yuksekgonul et al., 2026](https://arxiv.org/abs/2601.16175)),
-a method that runs RL *at test time* to let the model discover better solutions by trying and
-learning from real execution feedback.
+Produced via **TTT-Discover** ([Yuksekgonul et al., 2026](https://arxiv.org/abs/2601.16175)):
+
+> *"We perform reinforcement learning at test time, allowing the LLM to continue training
+> with experience specific to the problem at hand … Our test-time training runs are performed
+> using Tinker, an API by Thinking Machines, with a cost of only a few hundred dollars per problem."*
 
 ---
 
 ## What is TTT-Discover?
 
-> *"We perform reinforcement learning at test time, allowing the LLM to continue training with
-> experience specific to the problem at hand."*
-> — [Learning to Discover at Test Time](https://arxiv.org/abs/2601.16175)
-
-The model receives a reward signal based on measured GPU runtime (`reward = 1500 / runtime_μs`)
-and learns to write faster Triton kernels through trial and error — no human-written examples needed.
+Instead of prompting a frozen model (like AlphaEvolve), TTT-Discover **keeps training** on your
+specific problem at test time. The model earns a reward signal from real execution feedback
+(`reward = 1500 / runtime_μs`) and learns to write faster Triton kernels through trial and error —
+no human-written examples needed.
 
 **Published results (H100, TriMul):**
 
-| | Runtime ↓ |
-|---|---|
-| Best Human | 1371 μs |
-| Prev. Best AI | — |
-| **TTT-Discover** | **1161 μs** |
+| | A100 ↓ | H100 ↓ | B200 ↓ | MI300x ↓ |
+|---|---|---|---|---|
+| Best Human | 4531 μs | 1371 μs | 1005 μs | 2462 μs |
+| **TTT-Discover** | **2198 μs** | **1161 μs** | **905 μs** | **1596 μs** |
+
+[Verify on GPU Mode leaderboard →](https://www.gpumode.com/v2/leaderboard/496?tab=rankings)
 
-This checkpoint reached **~3638 μs** at step 14 (best reward 0.412), plateauing at ~3570 μs through step 30.
-Starting from this checkpoint instead of scratch saves ~14 steps of cold-start exploration.
+This checkpoint reached **~3638 μs** at step 14 (best reward 0.412).
+Starting from it instead of scratch saves ~14 steps of cold-start exploration.
 
 ---
 
@@ -64,7 +66,7 @@ Starting from this checkpoint instead of scratch saves ~14 steps of cold-start e
 | 25 | 0.281 | 0.156 | ~5340 μs |
 | 30 | 0.280 | 0.210 | ~5360 μs |
 
-> Steps 15–17 had zero reward due to a billing limit on the eval cluster — not a training failure.
+> Steps 15–17 had zero reward due to an eval cluster billing limit — not a training failure.
 
 ---
 
@@ -78,7 +80,7 @@ sampler_weights/
 gpu_mode_trimul.ipynb         # Google Colab notebook
 ```
 
-The training state (optimizer + model state for resuming RL) is stored on Tinker at:
+Training state (for resuming RL with fresh optimizer):
 `tinker://681a070d-2ef4-5b8c-a216-d4f22dca1efb:train:0/weights/000030`
 
 ---
@@ -87,7 +89,7 @@ The training state (optimizer + model state for resuming RL) is stored on Tinker
 
 [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/Pran-Ker/gpu-mode-trimul/blob/main/gpu_mode_trimul.ipynb)
 
-Open the notebook above to run TTT-Discover on the TriMul task, fork from this checkpoint,
+Open the notebook above to run TTT-Discover on TriMul, fork from this checkpoint,
 or plug in your own GPU kernel problem.
 
 ### What you need
@@ -100,14 +102,14 @@ or plug in your own GPU kernel problem.
 
 > No local GPU required — training runs on Tinker's cluster; kernel evals run on Modal H100s.
 
-### Run on TriMul from this checkpoint
+### Warm-start RL from this checkpoint
 
 ```python
 import asyncio, os
 from ttt_discover.rl.train import Config, main as rl_main
 from ttt_discover.tinker_utils import misc_utils
 from ttt_discover.tinker_utils.dataset_builder import DatasetConfig, get_single_problem_dataset_builder
-from examples.gpu_mode.env import GpuModeEnv
+from examples.gpu_mode.env import GpuModeEnv  # from github.com/test-time-training/discover
 
 CHECKPOINT = "tinker://681a070d-2ef4-5b8c-a216-d4f22dca1efb:train:0/weights/000030"
 EXPERIMENT  = "my-trimul-run"
@@ -166,8 +168,8 @@ print(asyncio.run(ask("Write a fast Triton kernel for triangular matmul on H100.
 **Learning to Discover at Test Time**  
 Mert Yuksekgonul\*, Daniel Koceja\*, Xinhao Li\*, Federico Bianchi\*, Jed McCaleb,
 Xiaolong Wang, Jan Kautz, Yejin Choi, James Zou†, Carlos Guestrin†, Yu Sun  
-Stanford · NVIDIA · Astera Institute · UC San Diego · Together AI  
-[arXiv:2601.16175](https://arxiv.org/abs/2601.16175) · [Project page](https://test-time-training.github.io/discover/)
+*Stanford · NVIDIA · Astera Institute · UC San Diego · Together AI*  
+[arXiv:2601.16175](https://arxiv.org/abs/2601.16175) · [Project page](https://test-time-training.github.io/discover/) · [PDF](https://test-time-training.github.io/discover.pdf)
 
 ```bibtex
 @article{ttt-discover2026,
@@ -186,4 +188,4 @@ Stanford · NVIDIA · Astera Institute · UC San Diego · Together AI
 ## Acknowledgments
 
 - **[GPU Mode](https://github.com/gpu-mode)** — community for GPU kernel optimization and the TriMul competition
-- **[Tinker](https://tinker.thinkingmachines.dev)** — LLM training and RL infrastructure
+- **[Tinker](https://tinker.thinkingmachines.dev)** — LLM training and RL infrastructure by Thinking Machines