Update README.md

README.md

@@ -10,18 +10,18 @@ license: apache-2.0
 
 [Google's LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/pdf/2112.07916.pdf) introduced as an extension of a successful [T5 model](https://arxiv.org/pdf/1910.10683.pdf).
 
-This is an unofficial *longt5-large-16384-pubmed-3k_steps* checkpoint. I.e., this is a large configuration of the LongT5 model with a `transient-global` attention fine-tuned on [pubmed summarization dataset](https://huggingface.co/datasets/ccdv/pubmed-summarization) for 3,000 training steps.
+This is an unofficial *longt5-large-16384-pubmed-3k_steps* checkpoint. I.e., this is a large configuration of the LongT5 model with a `transient-global` attention fine-tuned on [pubmed summarization dataset](https://huggingface.co/datasets/ccdv/pubmed-summarization) for 3,000 training steps. It may be worth continuing in the fine-tuning as we did not train the model until the convergence.
 
 ## Results and Fine-tuning Details 
 
-The fine-tuned model achieves the following results on the evaluation set using `beam_search=3` and without any specific calibration of generation parameters:
+The fine-tuned model achieves the following results on the evaluation set using `beam_search=3` and without any specific calibration of generation parameters are presented below, altogether with the results from the original paper (the original scores are higher, very likely due to a higher number of training steps).
 
-| Metric | Score |
-| --- | --- |
-| Rouge-1 | 47.44 |
-| Rouge-2 | 22.68 |
-| Rouge-L | 29.83 |
-| Rouge-Lsum | 43.13 |
+| Metric | Score | Score (original paper)
+| --- | --- | --- |
+| Rouge-1 | 47.44 | 49.98 |
+| Rouge-2 | 22.68 | 24.69 |
+| Rouge-L | 29.83 | x |
+| Rouge-Lsum | 43.13 | 46.46 |
 
 The full training hyper-parameters and logs can be found via the following [W&B run](https://wandb.ai/stancld/LongT5/runs/1lwncl8a?workspace=user-stancld). The model was trained using the [HuggingFace's trainer](https://github.com/huggingface/transformers/blob/main/src/transformers/trainer_seq2seq.py).