423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-08-02 03:32:50 +08:00
Code Issues Packages Projects Releases Wiki Activity

[data] fix mm template (#7181)

Browse Source 
Former-commit-id: be66df1f0211cd2d90eac3ab407dced653c9e443
This commit is contained in:
hoshi-hiyouga 2025-03-06 15:18:32 +08:00 committed by GitHub
parent 002f58ef8e
commit 2b21c749c1
2 changed files with 12 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
src/llamafactory/data/mm_plugin.py
View File

@ -1042,15 +1042,15 @@ class Qwen2VLPlugin(BasePlugin):
image = super()._preprocess_image(image, **kwargs)
if min(image.width, image.height) < 28:
width, height = max(image.width, 28), max(image.height, 28)
image = image.resize((width, height), resample=Image.Resampling.NEAREST)
image = image.resize((width, height))
if image.width / image.height > 200:
width, height = image.height * 180, image.height
image = image.resize((width, height), resample=Image.Resampling.NEAREST)
image = image.resize((width, height))
if image.height / image.width > 200:
width, height = image.width, image.width * 180
image = image.resize((width, height), resample=Image.Resampling.NEAREST)
image = image.resize((width, height))
return image

10
src/llamafactory/data/template.py
View File

@ -1268,9 +1268,17 @@ register_template(
)
# copied from gemma template
register_template(
name="paligemma",
format_user=StringFormatter(slots=["{{content}}\n"]),
format_prefix=EmptyFormatter(slots=[{"bos_token"}]),
mm_plugin=get_mm_plugin(name="paligemma", image_token="<image>"),
)
# copied from gemma template
register_template(
name="paligemma_chat",
format_user=StringFormatter(slots=["<start_of_turn>user\n{{content}}<end_of_turn>\n<start_of_turn>model\n"]),
format_assistant=StringFormatter(slots=["{{content}}<end_of_turn>\n"]),
format_observation=StringFormatter(