Update README.md

README.md

@@ -48,7 +48,9 @@ lighteval accelerate \
     --use-chat-template
 ```
 
-Llama-3.1-8B-Instruct got 50.67% accuracy. I haven't been able to run all of the questions by OpenAI, but ChatGPT was doing pretty good on a sample of questions.
+Llama-3.1-8B-Instruct got 50.67% accuracy on the full dataset.<br/>
+In a test of reasoning models on a small sample, `DeepSeek-R1-Distill-Llama-8B` got 4/9 and `DeepSeek-R1-Distill-Qwen-14B` got 7/9: [LightEval / CoLab notebook](https://colab.research.google.com/drive/1e4rZoEHWe3kVPDsdSGcUMhcVAAXLVOkD?usp=sharing),<br/>
+ChatGPT was doing pretty good on the questions that I tested, but I didn't do a full evaluation.<br/>
 
 ## Source information