Update README.md

README.md

@@ -78,17 +78,24 @@ UBTree support for Ling3 in SGLang is coming soon. We will share installation
 instructions and serving examples once the integration is publicly available.
 **Stay tuned!**
 
-## Evaluation notes
+## Benchmark results
 
-The [UBTree paper](https://arxiv.org/abs/2609.39972) reports results under its
-specified models and experimental configurations. Those results should not
-be treated as measurements of every released checkpoint or serving setup.
+Results for this checkpoint will be added here. **TBD** means not yet reported.
 
-Runtime version, hardware, tree budget, concurrency, and decoding settings
-all affect acceptance length and speed. Checkpoint-specific reproducible
-benchmark recipes will accompany the public runtime integration. Numerical
-or token-level equivalence and task quality require validation for the chosen
-serving configuration.
+| Dataset | Acceptance length | Speedup |
+|---|---:|---:|
+| GSM8K | TBD | TBD |
+| MATH-500 | TBD | TBD |
+| AIME 2025 | TBD | TBD |
+| HumanEval | TBD | TBD |
+| MBPP | TBD | TBD |
+| LiveCodeBench | TBD | TBD |
+
+Acceptance length is the mean number of tokens committed per verification
+step, including the target token. Speedup is measured against target-only
+autoregressive decoding with the same hardware, inputs, and decoding settings.
+The runtime version, tree budget, concurrency, and timing protocol will be
+reported alongside the results.
 
 ## Citation
 
@@ -106,5 +113,6 @@ If you use UBTree in your work, please cite:
 
 ## License
 
-The model weights are released under the **MIT License**. The reproduced paper
-figure is separately licensed under **CC BY 4.0**, as credited above.
+The model weights are released under the **MIT License**. The pipeline figure
+is from [Liang et al., Figure 2](https://arxiv.org/html/2609.39972v1#S2.F2),
+reproduced without modification under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).