From 7149c03d3691197d42b4c060e449e59ac4da45a7 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=EA=B9=80=EC=9E=AC=EC=9A=B0?= Date: Tue, 3 Mar 2026 01:07:09 +0900 Subject: [PATCH] fix: use processor.tokenizer for apply_chat_template in VLM preprocessing (#416) `AutoProcessor.from_pretrained()` returns a generic processor that does not have `apply_chat_template()`. The tokenizer accessible via `processor.tokenizer` does have this method with the chat template loaded from the model config. --- specforge/data/preprocessing.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/specforge/data/preprocessing.py b/specforge/data/preprocessing.py index 71a63820..e7df490e 100644 --- a/specforge/data/preprocessing.py +++ b/specforge/data/preprocessing.py @@ -245,7 +245,7 @@ def preprocess_vlm_conversations( else: messages.append({"role": role, "content": sentence["content"]}) - conversation = processor.apply_chat_template( + conversation = processor.tokenizer.apply_chat_template( messages, tokenize=False, add_generation_prompt=False,