metadata: 2026-07-07 revised taxonomy (task_type format vocabulary multiple_choice/yes_no/short_answer/free_form; score_type incl. none; re-verified protocol evidence; repaired prompt provenance)
Browse files- metadata.json +1 -1
metadata.json
CHANGED
|
@@ -9,7 +9,7 @@
|
|
| 9 |
"modalities": [
|
| 10 |
"single_image_start"
|
| 11 |
],
|
| 12 |
-
"task_type": "
|
| 13 |
"score_type": "llm_extract",
|
| 14 |
"score_protocol": {
|
| 15 |
"reference": "VLMEvalKit@vlmeval/dataset/image_vqa.py:1703-1760 + vlmeval/dataset/utils/logicvista.py:42-72 — every sample sent to GPT extractor (evaluate asserts a working OpenAI API); extracted letters are lowercased+sorted and exact-compared against gold answer split on ', ' (sorted letter-set compare); evaluate_logicvista aggregates Overall + per-skill accuracy. Matches official Yijia-Xiao/LogicVista GPT-extraction protocol.",
|
|
|
|
| 9 |
"modalities": [
|
| 10 |
"single_image_start"
|
| 11 |
],
|
| 12 |
+
"task_type": "multiple_choice",
|
| 13 |
"score_type": "llm_extract",
|
| 14 |
"score_protocol": {
|
| 15 |
"reference": "VLMEvalKit@vlmeval/dataset/image_vqa.py:1703-1760 + vlmeval/dataset/utils/logicvista.py:42-72 — every sample sent to GPT extractor (evaluate asserts a working OpenAI API); extracted letters are lowercased+sorted and exact-compared against gold answer split on ', ' (sorted letter-set compare); evaluate_logicvista aggregates Overall + per-skill accuracy. Matches official Yijia-Xiao/LogicVista GPT-extraction protocol.",
|