Use default docker paths

rossumai · Mar 13, 2023 · 9b1136e · 9b1136e
1 parent e19c817
commit 9b1136e
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 4 deletions.
diff --git a/baselines/README.md b/baselines/README.md
@@ -31,7 +31,7 @@ We provide code to reproduce all results that can be found in the [paper](../REA
 
 The code is structured into three subfolders:
 * [NER](NER/) contains most of the baselines code, including training code for RoBERTa, LayoutLMv3 and RoBERTa pretraining, and the inference code.
-* [layoutlmv3_pretraing](layoutlmv3_pretrain/) [coming soon] contains code for LayoutLMv3 pretraining
+* [layoutlmv3_pretraing](layoutlmv3_pretrain/) contains code for LayoutLMv3 pretraining.
 * [table-transformer](table-transformer/) [coming soon] contains code for DETR used for table and Line Item detection.
 
 ## Results on the validation set

diff --git a/baselines/layoutlmv3_pretrain/pretrain.py b/baselines/layoutlmv3_pretrain/pretrain.py
@@ -568,8 +568,8 @@ def __len__(self):
 
 
 def main():
-    dataset_path = Path("../../data/docile")
-    split = "unlabeled.json"
+    dataset_path = Path("/app/data/docile")
+    split = "unlabeled"
     model_name = "microsoft/layoutlmv3-base"
     batch_size = 128
     num_workers = 0
@@ -578,7 +578,7 @@ def main():
     weight_decay = 0.5
     gradient_norm_clip_val = 5.0
     optimizer = "AdamW"
-    ckpt_dir = "./saved_models"
+    ckpt_dir = "/app/data/baselines/trainings/layoutlmv3_pretraining"
     num_gpus = 8
     num_nodes = 1
     max_epochs = 30