Testing image dataset upload

faceid-data2.py

@@ -0,0 +1,72 @@
+# dataset_script.py
+import os
+import json
+from pathlib import Path
+
+import datasets
+from datasets import (
+    GeneratorBasedBuilder,
+    DatasetInfo,
+    SplitGenerator,
+    Features,
+    Sequence,
+    Value,
+    Image
+)
+
+_CITATION = """\
+@misc{faceid2025,
+  title={FaceID for nanoVLM},
+  author={Your Name},
+  year={2025},
+}
+"""
+
+_DESCRIPTION = """
+FaceID dataset: for each image, a single-question Q&A pair:
+"Question: Who is the person?" → "Answer: <Name>"
+"""
+
+class FaceIDDataset(GeneratorBasedBuilder):
+
+    VERSION = datasets.Version("1.0.0")
+
+    def _info(self):
+        return DatasetInfo(
+            description=_DESCRIPTION,
+            features=Features({
+                # a *list* of images (we only have one per example, but must match your JSONL)
+                "images": Sequence(feature=Image(decode=True)),
+                # a list of QA dicts
+                "texts":  Sequence({
+                    "user":      Value("string"),
+                    "assistant": Value("string"),
+                    "source":    Value("string"),
+                }),
+            }),
+            supervised_keys=None,
+            homepage="https://huggingface.co/datasets/your-username/faceid-data2",
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        # we already have JSONL + images in the same folder, no download needed
+        data_dir = Path(__file__).parent
+        return [
+            SplitGenerator(name=datasets.Split.TRAIN, 
+                           gen_kwargs={"jsonl": data_dir / "train.jsonl"}),
+            SplitGenerator(name=datasets.Split.VALIDATION,
+                           gen_kwargs={"jsonl": data_dir / "val.jsonl"}),
+        ]
+
+    def _generate_examples(self, jsonl):
+        base = Path(__file__).parent
+        with open(jsonl, "r", encoding="utf-8") as f:
+            for idx, line in enumerate(f):
+                ex = json.loads(line)
+                # images is a list of relative paths under this repo
+                img_list = [ str(base / p) for p in ex["images"] ]
+                yield idx, {
+                    "images": img_list,
+                    "texts":  ex["texts"]
+                }