Update README.md

README.md

@@ -70,6 +70,10 @@ To transfer knowledge from English model to Czech, we developed a simple method
 <img src="figures/tllama_test.png" width="900"/>
 Figure 4: Test perplexity over the course of training for vocabulary swap (swapping 1.7K tokens) method on TinyLLAMA. Our method (green curve) vs TinyLLAMA training from scratch (blue curve).
 
+We also verify that finetuning from English to Czech is beneficial for MPT-7B model, compared from training a new model, at least on the first 10K steps. The training also seems to be more stable (notice yellow spike around 10k steps).
+Figure 5: Test cross-entropy over the course of training on CSMPT7B (yellow-red). Comparison with TinyLLAMA (blue-green). Our method (red&green curve) vs TinyLLAMA training from scratch (yellow&blue curve).
+<img src="figures/csmpt_tllama_test.png" width="900"/>
+
 The vocabulary swap was done the same way as our [Czech-GPT-2](https://huggingface.co/BUT-FIT/Czech-GPT-2-XL-133k) model (check it out for comprehensive description.)
 For CSMPT7b, we managed to align 4,177 english tokens with corresponding czech tokens. 
 
@@ -97,7 +101,6 @@ Not mentioned hyperparameters were kept the same as for MPT.
 | scheduler_steps            | 170,000       |                                                                                              |
 | scheduler_alpha            | 0.1           | So LR on last step is 0.1*(vanilla LR)                                                       |
 
-
 # Usage
 ## How to Setup Environment
 ```bash