From c171573a8f1e4c243a26e6a99cf031afee8bc6aa Mon Sep 17 00:00:00 2001 From: Roland Tannous Date: Thu, 5 Mar 2026 06:10:10 +0000 Subject: [PATCH] fix: check for http(s) prefix instead of bare string type for URL detection --- studio/backend/utils/datasets/format_conversion.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/studio/backend/utils/datasets/format_conversion.py b/studio/backend/utils/datasets/format_conversion.py index c5c9a4d6e7..41a9617857 100644 --- a/studio/backend/utils/datasets/format_conversion.py +++ b/studio/backend/utils/datasets/format_conversion.py @@ -350,7 +350,8 @@ def convert_to_vlm_format( return {"messages": messages} total = len(dataset) - has_urls = isinstance(next(iter(dataset))[image_column], str) + first_image = next(iter(dataset))[image_column] + has_urls = isinstance(first_image, str) and first_image.startswith(("http://", "https://")) # ── URL probe: 200 samples with parallel workers to estimate speed + failure rate ── PROBE_SIZE = 200