remind users to use glove embeddings for reproducing the results

allanj · Mar 11, 2020 · 25bd85c · 25bd85c
1 parent 33f714e
commit 25bd85c
Showing 1 changed file with 2 additions and 0 deletions.
diff --git a/docs/benchmark.md b/docs/benchmark.md
@@ -1,6 +1,8 @@
 ### Benchmark Performance
 
 Empirically, although `ADAM` optimizer converges faster, we found that using `SGD` with learning rate of `0.01` and `100` epochs is better than `ADAM`.
+The following results are obtained with the Glove embeddings (`glove.6B.100d.txt`).
+Download the embeddings and specify the path as arguments.
 
 * Experiments on the CoNLL-2003 dataset