use the same tokenizer

This commit is contained in:
Shizhe Diao 2025-10-20 11:11:31 -07:00
parent cee6a17d9e
commit fc23c1aa71

View File

@ -82,7 +82,7 @@ def build_model(checkpoint_dir, step, device, phase):
else:
model.train()
# Load the Tokenizer
tokenizer_name = meta_data["tokenizer_name"]
tokenizer_name = meta_data["tokenizer_name"] if "tokenizer_name" in meta_data else "tokenizer"
print(f"Loading tokenizer: {tokenizer_name}")
tokenizer = get_tokenizer(tokenizer_name)
# Sanity check: compatibility between model and tokenizer