Handling block_size value for models with no sequence length limit. (#…

…1487) * handle block size -1 * change where to fetch --------- Co-authored-by: Adam Louly <[email protected]@orttrainingdev9.d32nl1ml4oruzj4qz3bqlggovf.px.internal.cloudapp.net>
huggingface · Oct 27, 2023 · e164827 · e164827
1 parent 313e1bd
commit e164827
Showing 1 changed file with 2 additions and 0 deletions.
diff --git a/examples/onnxruntime/training/language-modeling/run_clm.py b/examples/onnxruntime/training/language-modeling/run_clm.py
@@ -495,6 +495,8 @@ def tokenize_function(examples):
 
     if hasattr(config, "max_position_embeddings"):
         max_pos_embeddings = config.max_position_embeddings
+        if max_pos_embeddings < 0:
+            max_pos_embeddings = 1024
     else:
         # Define a default value if the attribute is missing in the config.
         max_pos_embeddings = 1024