From 48615ddb076e2413c6535848f016aaf401441f13 Mon Sep 17 00:00:00 2001 From: davidlightmysterion Date: Tue, 5 Aug 2025 20:35:07 +0800 Subject: [PATCH] [train] fix adjusting logits size after adding special tokens (#8823) --- src/llamafactory/model/model_utils/embedding.py | 1 + 1 file changed, 1 insertion(+) diff --git a/src/llamafactory/model/model_utils/embedding.py b/src/llamafactory/model/model_utils/embedding.py index c10e34f4..5604d854 100644 --- a/src/llamafactory/model/model_utils/embedding.py +++ b/src/llamafactory/model/model_utils/embedding.py @@ -67,4 +67,5 @@ def resize_embedding_layer(model: "PreTrainedModel", tokenizer: "PreTrainedToken _noisy_mean_initialization(model.get_input_embeddings().weight.data, num_new_tokens) _noisy_mean_initialization(model.get_output_embeddings().weight.data, num_new_tokens) + model.config.vocab_size = new_embedding_size logger.info_rank0(f"Resized token embeddings from {current_embedding_size} to {new_embedding_size}.")