diff --git a/examples/experiments/onto-joint/pretrain.cfg b/examples/experiments/onto-joint/pretrain.cfg
index 1637cceae..4f1898d69 100644
--- a/examples/experiments/onto-joint/pretrain.cfg
+++ b/examples/experiments/onto-joint/pretrain.cfg
@@ -53,17 +53,7 @@ n_save_every = null
 batch_size = 3000
 seed = ${training:seed}
 use_pytorch_for_gpu_memory = ${training:use_pytorch_for_gpu_memory}
-
-[pretraining.model]
-@architectures = "spacy.HashEmbedCNN.v1"
-pretrained_vectors = ${nlp:vectors}
-width = 256
-depth = 6
-window_size = 1
-embed_size = 2000
-maxout_pieces = 3
-subword_features = true
-dropout = null
+tok2vec_model = "nlp.pipeline.tok2vec.model"
 
 [pretraining.optimizer]
 @optimizers = "Adam.v1"
diff --git a/spacy/cli/pretrain.py b/spacy/cli/pretrain.py
index 96564b98b..921eb38ab 100644
--- a/spacy/cli/pretrain.py
+++ b/spacy/cli/pretrain.py
@@ -109,7 +109,10 @@ def pretrain(
     with msg.loading(f"Loading model '{vectors_model}'..."):
         nlp = util.load_model(vectors_model)
     msg.good(f"Loaded model '{vectors_model}'")
-    tok2vec = pretrain_config["model"]
+    tok2vec_path = pretrain_config["tok2vec_model"]
+    tok2vec = config
+    for subpath in tok2vec_path.split("."):
+        tok2vec = tok2vec.get(subpath)
     model = create_pretraining_model(nlp, tok2vec)
     optimizer = pretrain_config["optimizer"]
 
diff --git a/spacy/ml/models/multi_task.py b/spacy/ml/models/multi_task.py
index 970d31899..8000d1aff 100644
--- a/spacy/ml/models/multi_task.py
+++ b/spacy/ml/models/multi_task.py
@@ -102,4 +102,4 @@ def _replace_word(word, random_words, mask="[MASK]"):
     elif roll < 0.9:
         return random_words.next()
     else:
-        return word
\ No newline at end of file
+        return word