From 06a5c59a476bd4cc740283ca1dce85a3872676f4 Mon Sep 17 00:00:00 2001 From: Maurizio Dipierro Date: Wed, 19 Mar 2025 17:06:26 +0100 Subject: [PATCH] truncation true --- app.py | 13 ++++++++----- 1 file changed, 8 insertions(+), 5 deletions(-) diff --git a/app.py b/app.py index 07e232d..7e7a02c 100644 --- a/app.py +++ b/app.py @@ -41,10 +41,11 @@ def normalize_values(text, target_max=500): return normalized_text processor = AutoProcessor.from_pretrained("ds4sd/SmolDocling-256M-preview") -model = AutoModelForVision2Seq.from_pretrained("ds4sd/SmolDocling-256M-preview", - torch_dtype=torch.bfloat16, - # _attn_implementation="flash_attention_2" - ).to("cuda") +model = AutoModelForVision2Seq.from_pretrained( + "ds4sd/SmolDocling-256M-preview", + torch_dtype=torch.bfloat16, + # _attn_implementation="flash_attention_2" +).to("cuda") def model_inference(input_dict, history): text = input_dict["text"] @@ -77,7 +78,9 @@ def model_inference(input_dict, history): } ] prompt = processor.apply_chat_template(resulting_messages, add_generation_prompt=True) - inputs = processor(text=prompt, images=[images], return_tensors="pt").to('cuda') + + # Added truncation=True to explicitly activate truncation. + inputs = processor(text=prompt, images=[images], return_tensors="pt", truncation=True).to('cuda') generation_args = { "input_ids": inputs.input_ids,