metadata: migrate score_type -> score_pipeline (atomic stage contract; see mm-eval scorer docs/en/SCORING.md)

metadata.json

@@ -1,56 +1,59 @@
 {
- "name": "MMVP",
- "release_date": "2024-01-01",
- "subsets": {
-  "main": {
-   "language": [
-    "en"
-   ],
-   "modalities": [
-    "single_image_start"
-   ],
-   "task_type": "multiple_choice_qa",
-   "score_type": "llm_extract",
-   "score_protocol": {
-    "reference": "vlmevalkit@vlmeval/dataset/image_mcq.py:249-324 (ImageMCQDataset.evaluate_heuristic) -> vlmevalkit@vlmeval/dataset/utils/multiple_choice.py:475-508 mcq_vanilla_eval: can_infer rule prefetch, then GPT extractor (extract_answer_from_item @vlmeval/dataset/utils/multiple_choice.py:359-407) maps verbose output to a letter, rule-compare vs GT; MMVP registered @vlmeval/dataset/image_mcq.py:121; headline via report_acc_MMVP @vlmeval/dataset/utils/multiple_choice.py:613-622: pairs of consecutive rows (2i, 2i+1) — 'Overall' counts a pair only if BOTH hits are 1",
-    "note": "Official headline 'Overall' is PAIR accuracy (a CLIP-blind pair scores only if both questions are correct); a per-sample scorer reports per-question accuracy, systematically higher. Pair = consecutive ids (mm-eval ids 1..300, pair k = ids 2k-1 & 2k). GT stored as lowercase parenthesized letters '(a)'/'(b)' with options in options_text ('(a) Open (b) Closed'); matcher must normalize case/parens."
-   },
-   "prompt_template": "<image>{{ question }} {{ options.options_text }}",
-   "prompt_template_source": {
-    "origin": "official",
-    "reference": "https://github.com/tsb0601/MMVP/blob/main/scripts/evaluate_mllm.py (cur_prompt = row['Question'] + ' ' + row['Options'])",
-    "notes": "Tier 1: MMVP official evaluation prompt per tsb0601/MMVP/scripts/evaluate_mllm.py line 59: cur_prompt = row['Question'] + ' ' + row['Options']. No trailer is appended by the official script. Options field uses '(a) Open (b) Closed' inline format from the source CSV/HF parquet."
-   },
-   "mapping_from_source": {
-    "media": {
-     "from": "image",
-     "type": "list",
-     "min_items": 1,
-     "max_items": 1
-    },
-    "id": {
-     "from": "Index"
-    },
-    "question": {
-     "from": "Question"
-    },
-    "answer": {
-     "from": "Correct Answer",
-     "optional": true
-    },
-    "extra": {
-     "options_text": {
-      "from": "Options",
-      "note": "Pre-rendered options string in '(a) opt1 (b) opt2' format from MMVP source."
-     }
-    },
-    "source": {
-     "format": "huggingface",
-     "url": {
-      "test": "https://huggingface.co/datasets/lmms-lab-eval/MMVP"
-     }
+  "name": "MMVP",
+  "release_date": "2024-01-01",
+  "subsets": {
+    "main": {
+      "language": [
+        "en"
+      ],
+      "modalities": [
+        "single_image_start"
+      ],
+      "task_type": "multiple_choice_qa",
+      "score_pipeline": [
+        "rule-match",
+        "llm-match"
+      ],
+      "score_protocol": {
+        "reference": "vlmevalkit@vlmeval/dataset/image_mcq.py:249-324 (ImageMCQDataset.evaluate_heuristic) -> vlmevalkit@vlmeval/dataset/utils/multiple_choice.py:475-508 mcq_vanilla_eval: can_infer rule prefetch, then GPT extractor (extract_answer_from_item @vlmeval/dataset/utils/multiple_choice.py:359-407) maps verbose output to a letter, rule-compare vs GT; MMVP registered @vlmeval/dataset/image_mcq.py:121; headline via report_acc_MMVP @vlmeval/dataset/utils/multiple_choice.py:613-622: pairs of consecutive rows (2i, 2i+1) — 'Overall' counts a pair only if BOTH hits are 1",
+        "note": "Official headline 'Overall' is PAIR accuracy (a CLIP-blind pair scores only if both questions are correct); a per-sample scorer reports per-question accuracy, systematically higher. Pair = consecutive ids (mm-eval ids 1..300, pair k = ids 2k-1 & 2k). GT stored as lowercase parenthesized letters '(a)'/'(b)' with options in options_text ('(a) Open (b) Closed'); matcher must normalize case/parens."
+      },
+      "prompt_template": "<image>{{ question }} {{ options.options_text }}",
+      "prompt_template_source": {
+        "origin": "official",
+        "reference": "https://github.com/tsb0601/MMVP/blob/main/scripts/evaluate_mllm.py (cur_prompt = row['Question'] + ' ' + row['Options'])",
+        "notes": "Tier 1: MMVP official evaluation prompt per tsb0601/MMVP/scripts/evaluate_mllm.py line 59: cur_prompt = row['Question'] + ' ' + row['Options']. No trailer is appended by the official script. Options field uses '(a) Open (b) Closed' inline format from the source CSV/HF parquet."
+      },
+      "mapping_from_source": {
+        "media": {
+          "from": "image",
+          "type": "list",
+          "min_items": 1,
+          "max_items": 1
+        },
+        "id": {
+          "from": "Index"
+        },
+        "question": {
+          "from": "Question"
+        },
+        "answer": {
+          "from": "Correct Answer",
+          "optional": true
+        },
+        "extra": {
+          "options_text": {
+            "from": "Options",
+            "note": "Pre-rendered options string in '(a) opt1 (b) opt2' format from MMVP source."
+          }
+        },
+        "source": {
+          "format": "huggingface",
+          "url": {
+            "test": "https://huggingface.co/datasets/lmms-lab-eval/MMVP"
+          }
+        }
+      }
     }
-   }
   }
- }
-}
+}