[Fix] preserve image data in preprocess for VLM training on multimodal data

specforge/utils.py

@@ -405,6 +405,10 @@ def safe_conversations_generator(file_path):
                     else:
                         result["tools"] = []
 
+                # preserve image - required in preprocess_vlm_conversations for vlm
+                if "image" in row:
+                    result["image"] = row["image"]
+
                 yield result
 
             except Exception as e:

tests/test_data/data/vlm_conversation.jsonl

@@ -0,0 +1,2 @@
+
+{"id": 1, "image": "/data/images/test.jpg", "conversations": [{"role": "user", "content": "Describe this image."}, {"role": "assistant", "content": "A cat."}]}

tests/test_data/test_build_eagle3_dataset.py

@@ -194,6 +194,26 @@ def test_build_eagle3_dataset_basic(self):
             assistant_indices = torch.where(loss_mask == 1)[0]
             self.assertTrue(len(assistant_indices) > 0, "No assistant tokens found")
 
+    def test_vlm_image_field_preserved(self):
+        """Regression test - preprocess_vlm_conversations relies on 'image'
+        field being preserved when the dataset is built with build_eagle3_dataset."""
+        data_file = os.path.join(
+            os.path.dirname(__file__), "data", "vlm_conversation.jsonl"
+        )
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            dataset = Dataset.from_generator(
+                generator=safe_conversations_generator,
+                gen_kwargs={"file_path": data_file},
+                cache_dir=tmp_dir,
+                keep_in_memory=True,
+            )
+            self.assertIn(
+                "image",
+                dataset.column_names,
+                "image column missing from dataset — results in KeyError in preprocess_vlm_conversations",
+            )
+            self.assertEqual(dataset[0]["image"], "/data/images/test.jpg")
+
 
 if __name__ == "__main__":
     unittest.main(verbosity=2)