eval logs: minicpm_o_interleave (lite)

eval_logs/minicpm_o_interleave/lite/summary.json

@@ -1,165 +1,165 @@
 {
   "by_config": {
     "a2t": 58.99318913454119,
-    "a2v": 19.84899034240562,
+    "a2v": 31.976348205496578,
     "t2a": 44.95670045494452,
-    "t2v": 40.699736611062335,
-    "v2a": 35.97892888498683,
-    "v2t": 64.1642855242504
+    "t2v": 62.80099502487562,
+    "v2a": 41.52853380158034,
+    "v2t": 76.37983877404422
   },
   "by_family": {
-    "perception": 63.75,
-    "spatial": 20.77365696768682,
-    "temporal": 17.08251670938238,
-    "linguistic": 70.75,
-    "knowledge": 44.0
+    "perception": 66.91666666666667,
+    "spatial": 37.88632594602743,
+    "temporal": 32.963968038594906,
+    "linguistic": 71.0,
+    "knowledge": 52.0
   },
   "disparity": {
-    "T_vs_V": -48.12197036209326,
-    "T_vs_A": -49.03610290792029,
-    "V_vs_A": -0.9141325458270302
+    "T_vs_V": -30.445007582408792,
+    "T_vs_A": -65.67595179184292,
+    "V_vs_A": -35.23094420943413
   },
   "imbalance": {
-    "T_vs_V": -23.464548913188068,
+    "T_vs_V": -13.578843749168598,
     "T_vs_A": -14.036488679596673,
-    "V_vs_A": 16.129938542581208
+    "V_vs_A": 9.55218559608376
   },
   "level1": {
     "knowledge/emotion_classification": {
       "a2t": 74.0,
-      "a2v": 0.0,
+      "a2v": 26.0,
       "t2a": 54.0,
-      "t2v": 0.0,
-      "v2a": 0.0,
-      "v2t": 22.0
+      "t2v": 62.0,
+      "v2a": 26.0,
+      "v2t": 60.0
     },
     "knowledge/movie_matching": {
       "a2t": 58.0,
-      "a2v": 48.0,
+      "a2v": 52.0,
       "t2a": 0.0,
-      "t2v": 94.0,
+      "t2v": 98.0,
       "v2a": 0.0,
-      "v2t": 96.0
+      "v2t": 92.0
     },
     "knowledge/music_genre_classification": {
       "a2t": 82.0,
-      "a2v": 18.0,
+      "a2v": 30.0,
       "t2a": 72.0,
-      "t2v": 58.0,
-      "v2a": 60.0,
-      "v2t": 82.0
+      "t2v": 76.0,
+      "v2a": 58.0,
+      "v2t": 86.0
     },
     "knowledge/singer_identification": {
       "a2t": 52.0,
-      "a2v": 34.0,
+      "a2v": 36.0,
       "t2a": 0.0,
       "t2v": 82.0,
       "v2a": 0.0,
-      "v2t": 70.0
+      "v2t": 72.0
     },
     "linguistic/recognition": {
       "a2t": 93.0,
-      "a2v": 40.0,
+      "a2v": 44.0,
       "t2a": 68.0,
-      "t2v": 51.0,
-      "v2a": 76.0,
+      "t2v": 48.0,
+      "v2a": 77.0,
       "v2t": 96.0
     },
     "linguistic/translation": {
       "a2t": 93.0,
-      "a2v": 44.0,
+      "a2v": 42.0,
       "t2a": 81.0,
-      "t2v": 41.0,
-      "v2a": 79.0,
-      "v2t": 87.0
+      "t2v": 45.0,
+      "v2a": 77.0,
+      "v2t": 88.0
     },
     "perception/finegrained": {
       "a2t": 52.5,
-      "a2v": 27.5,
+      "a2v": 32.5,
       "t2a": 70.0,
-      "t2v": 87.5,
-      "v2a": 67.5,
-      "v2t": 90.0
+      "t2v": 85.0,
+      "v2a": 52.5,
+      "v2t": 87.5
     },
     "perception/general_activities": {
       "a2t": 72.5,
-      "a2v": 42.5,
+      "a2v": 45.0,
       "t2a": 85.0,
-      "t2v": 92.5,
-      "v2a": 75.0,
+      "t2v": 95.0,
+      "v2a": 77.5,
       "v2t": 92.5
     },
     "perception/instruments": {
       "a2t": 60.0,
-      "a2v": 32.5,
+      "a2v": 30.0,
       "t2a": 40.0,
-      "t2v": 75.0,
-      "v2a": 72.5,
+      "t2v": 72.5,
+      "v2a": 67.5,
       "v2t": 92.5
     },
     "perception/instruments_comp": {
       "a2t": 37.5,
-      "a2v": 0.0,
+      "a2v": 20.0,
       "t2a": 62.5,
-      "t2v": 0.0,
-      "v2a": 52.5,
-      "v2t": 95.0
+      "t2v": 92.5,
+      "v2a": 60.0,
+      "v2t": 90.0
     },
     "perception/natures": {
       "a2t": 82.5,
       "a2v": 37.5,
       "t2a": 80.0,
-      "t2v": 90.0,
+      "t2v": 85.0,
       "v2a": 57.5,
-      "v2t": 90.0
+      "v2t": 92.5
     },
     "spatial/3D_movements": {
       "a2t": 73.13432835820896,
-      "a2v": 0.0,
+      "a2v": 26.865671641791046,
       "t2a": 35.82089552238806,
-      "t2v": 0.0,
-      "v2a": 7.462686567164179,
-      "v2t": 20.895522388059703
+      "t2v": 50.74626865671642,
+      "v2a": 34.32835820895522,
+      "v2t": 58.208955223880594
     },
     "spatial/arrangements": {
       "a2t": 31.34328358208955,
-      "a2v": 7.462686567164179,
+      "a2v": 19.402985074626866,
       "t2a": 23.880597014925375,
-      "t2v": 16.417910447761194,
-      "v2a": 26.865671641791046,
-      "v2t": 77.61194029850746
+      "t2v": 37.3134328358209,
+      "v2a": 23.880597014925375,
+      "v2t": 76.11940298507463
     },
     "spatial/panaroma": {
       "a2t": 27.272727272727273,
-      "a2v": 0.0,
+      "a2v": 33.333333333333336,
       "t2a": 24.242424242424242,
-      "t2v": 0.0,
-      "v2a": 0.0,
-      "v2t": 1.5151515151515151
+      "t2v": 34.84848484848485,
+      "v2a": 24.242424242424242,
+      "v2t": 46.96969696969697
     },
     "temporal/calculation": {
       "a2t": 41.791044776119406,
-      "a2v": 2.985074626865672,
+      "a2v": 25.37313432835821,
       "t2a": 11.940298507462687,
-      "t2v": 2.985074626865672,
-      "v2a": 17.91044776119403,
-      "v2t": 26.865671641791046
+      "t2v": 31.34328358208955,
+      "v2a": 22.388059701492537,
+      "v2t": 38.80597014925373
     },
     "temporal/count": {
       "a2t": 25.37313432835821,
-      "a2v": 2.985074626865672,
+      "a2v": 20.895522388059703,
       "t2a": 11.940298507462687,
-      "t2v": 1.492537313432836,
-      "v2a": 19.402985074626866,
-      "v2t": 40.298507462686565
+      "t2v": 23.880597014925375,
+      "v2a": 22.388059701492537,
+      "v2t": 62.6865671641791
     },
     "temporal/order": {
       "a2t": 46.96969696969697,
-      "a2v": 0.0,
+      "a2v": 22.727272727272727,
       "t2a": 43.93939393939394,
-      "t2v": 0.0,
-      "v2a": 0.0,
-      "v2t": 10.606060606060606
+      "t2v": 48.484848484848484,
+      "v2a": 25.757575757575758,
+      "v2t": 66.66666666666667
     }
   }
 }

eval_logs/minicpm_o_interleave/lite/xmod_bench_lite_t2a.jsonl

@@ -433,7 +433,7 @@
 {"doc_id": 432, "target": "A", "filtered_resps": "A", "token_counts": [null], "doc_hash": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b", "xmod_bench_score": {"family": "linguistic", "subtask": "recognition", "config": "t2a", "correct": 1}, "input": "Context: scored for the second occasion on a four-yard run in the fourth quarter\n\nWhich audio most likely contains the same content as this text? Answer the question with A, B, C, or D\nA: <media_0>\nB: <media_1>\nC: <media_2>\nD: <media_3>\nAnswer with the option's letter (A, B, C, or D) directly."}
 {"doc_id": 433, "target": "A", "filtered_resps": "C", "token_counts": [null], "doc_hash": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b", "xmod_bench_score": {"family": "temporal", "subtask": "order", "config": "t2a", "correct": 0}, "input": "Context: 1. baby laughter; 2. goat bleating; 3. playing djembe; 4. opening or closing drawers; 5. electric shaver\n\nThe text below describes the temporal sequence in which several events occur over time (from first to last). Listen carefully to each audio clip and pay attention to the chronological order of the sounds. Which audio clip best matches this temporal sequence? Choose A, B, C, or D.\nA: <media_0>\nB: <media_1>\nC: <media_2>\nD: <media_3>\nAnswer with the option's letter (A, B, C, or D) directly."}
 {"doc_id": 434, "target": "A", "filtered_resps": "A", "token_counts": [null], "doc_hash": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b", "xmod_bench_score": {"family": "perception", "subtask": "natures", "config": "t2a", "correct": 1}, "input": "Context: Thunder\n\nBased on this description: 'thunder', which audio clip best matches the natural scenes described? Answer with A, B, C, or D\nA: <media_0>\nB: <media_1>\nC: <media_2>\nD: <media_3>\nAnswer with the option's letter (A, B, C, or D) directly."}
-{"doc_id": 435, "target": "D", "filtered_resps": "A", "token_counts": [null], "doc_hash": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b", "xmod_bench_score": {"family": "temporal", "subtask": "count", "config": "t2a", "correct": 0}, "input": "Context: 3 times\n\nListen carefully to each audio clip—focus on how many repetitions of the action are in the audio. Which audio clip best matches this number of repetitions? Choose A, B, C, or D.\nA: <media_0>\nB: <media_1>\nC: <media_2>\nD: <media_3>\nAnswer with the option's letter (A, B, C, or D) directly."}
+{"doc_id": 435, "target": "D", "filtered_resps": "C", "token_counts": [null], "doc_hash": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b", "xmod_bench_score": {"family": "temporal", "subtask": "count", "config": "t2a", "correct": 0}, "input": "Context: 3 times\n\nListen carefully to each audio clip—focus on how many repetitions of the action are in the audio. Which audio clip best matches this number of repetitions? Choose A, B, C, or D.\nA: <media_0>\nB: <media_1>\nC: <media_2>\nD: <media_3>\nAnswer with the option's letter (A, B, C, or D) directly."}
 {"doc_id": 436, "target": "B", "filtered_resps": "D", "token_counts": [null], "doc_hash": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b", "xmod_bench_score": {"family": "temporal", "subtask": "calculation", "config": "t2a", "correct": 0}, "input": "Context: 2 times\n\nApply the following formula to this number of repetitions: ((repetition count - 1) * 3). Then listen carefully to each audio clip—focus on how many repetitions of the action are in the audio. Which audio clip best matches the resulting number? Choose A, B, C, or D.\nA: <media_0>\nB: <media_1>\nC: <media_2>\nD: <media_3>\nAnswer with the option's letter (A, B, C, or D) directly."}
 {"doc_id": 437, "target": "D", "filtered_resps": "D", "token_counts": [null], "doc_hash": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b", "xmod_bench_score": {"family": "spatial", "subtask": "panaroma", "config": "t2a", "correct": 1}, "input": "Context: A male speech sound is toward the front-right (-77°). \n\nSelect the spatial audio clip that best matches the text-described sound direction. (Angles range from -180° to 180°, increasing counter-clockwise: 0° = front, +90° = left, −90° = right, ±180° = back)\nA: <media_0>\nB: <media_1>\nC: <media_2>\nD: <media_3>\nAnswer with the option's letter (A, B, C, or D) directly."}
 {"doc_id": 438, "target": "A", "filtered_resps": "A", "token_counts": [null], "doc_hash": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b", "xmod_bench_score": {"family": "perception", "subtask": "finegrained", "config": "t2a", "correct": 1}, "input": "Context: squishing water\n\nBased on this description: 'squishing water', which audio clip best matches what is described? Answer with A, B, C, or D\nA: <media_0>\nB: <media_1>\nC: <media_2>\nD: <media_3>\nAnswer with the option's letter (A, B, C, or D) directly."}
@@ -828,7 +828,7 @@
 {"doc_id": 827, "target": "B", "filtered_resps": "A", "token_counts": [null], "doc_hash": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b", "xmod_bench_score": {"family": "spatial", "subtask": "arrangements", "config": "t2a", "correct": 0}, "input": "Context: From left to right: Viola, Violin, Violin\n\nThe text below describes how several instruments are arranged from left to right. Listen carefully to each audio clip—focus on how the sounds are distributed across the left and right channels. Which audio clip best matches this spatial arrangement? Choose A, B, C, or D.\nA: <media_0>\nB: <media_1>\nC: <media_2>\nD: <media_3>\nAnswer with the option's letter (A, B, C, or D) directly."}
 {"doc_id": 828, "target": "A", "filtered_resps": "A", "token_counts": [null], "doc_hash": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b", "xmod_bench_score": {"family": "perception", "subtask": "general_activities", "config": "t2a", "correct": 1}, "input": "Context: lions growling\n\nBased on this description: 'lions growling', which audio clip best matches what is described? Answer with A, B, C, or D\nA: <media_0>\nB: <media_1>\nC: <media_2>\nD: <media_3>\nAnswer with the option's letter (A, B, C, or D) directly."}
 {"doc_id": 829, "target": "A", "filtered_resps": "A", "token_counts": [null], "doc_hash": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b", "xmod_bench_score": {"family": "perception", "subtask": "general_activities", "config": "t2a", "correct": 1}, "input": "Context: telephone bell ringing\n\nBased on this description: 'telephone bell ringing', which audio clip best matches what is described? Answer with A, B, C, or D\nA: <media_0>\nB: <media_1>\nC: <media_2>\nD: <media_3>\nAnswer with the option's letter (A, B, C, or D) directly."}
-{"doc_id": 830, "target": "A", "filtered_resps": "C", "token_counts": [null], "doc_hash": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b", "xmod_bench_score": {"family": "temporal", "subtask": "calculation", "config": "t2a", "correct": 0}, "input": "Context: 2 times\n\nApply the following formula to this number of repetitions: (repetition count * 3). Then listen carefully to each audio clip—focus on how many repetitions of the action are in the audio. Which audio clip best matches the resulting number? Choose A, B, C, or D.\nA: <media_0>\nB: <media_1>\nC: <media_2>\nD: <media_3>\nAnswer with the option's letter (A, B, C, or D) directly."}
+{"doc_id": 830, "target": "A", "filtered_resps": "", "token_counts": [null], "doc_hash": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b", "xmod_bench_score": {"family": "temporal", "subtask": "calculation", "config": "t2a", "correct": 0}, "input": "Context: 2 times\n\nApply the following formula to this number of repetitions: (repetition count * 3). Then listen carefully to each audio clip—focus on how many repetitions of the action are in the audio. Which audio clip best matches the resulting number? Choose A, B, C, or D.\nA: <media_0>\nB: <media_1>\nC: <media_2>\nD: <media_3>\nAnswer with the option's letter (A, B, C, or D) directly."}
 {"doc_id": 831, "target": "D", "filtered_resps": "A", "token_counts": [null], "doc_hash": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b", "xmod_bench_score": {"family": "temporal", "subtask": "calculation", "config": "t2a", "correct": 0}, "input": "Context: 8 times\n\nApply the following formula to this number of repetitions: (repetition count * 4). Then listen carefully to each audio clip—focus on how many repetitions of the action are in the audio. Which audio clip best matches the resulting number? Choose A, B, C, or D.\nA: <media_0>\nB: <media_1>\nC: <media_2>\nD: <media_3>\nAnswer with the option's letter (A, B, C, or D) directly."}
 {"doc_id": 832, "target": "C", "filtered_resps": "", "token_counts": [null], "doc_hash": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b", "xmod_bench_score": {"family": "knowledge", "subtask": "movie_matching", "config": "t2a", "correct": 0}, "input": "Context: The Demon Slayer Corps are drawn into the Infinity Castle, where Tanjiro, Nezuko, and the Hashira face terrifying Upper Rank demons in a desperate fight as the final battle against Muzan Kibutsuji begins.\n\nRead this plot overview. Which trailer audio belongs to the film described? Answer with A, B, C, or D\nA: <media_0>\nB: <media_1>\nC: <media_2>\nD: <media_3>\nAnswer with the option's letter (A, B, C, or D) directly."}
 {"doc_id": 833, "target": "A", "filtered_resps": "C", "token_counts": [null], "doc_hash": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b", "xmod_bench_score": {"family": "temporal", "subtask": "count", "config": "t2a", "correct": 0}, "input": "Context: 2 times\n\nListen carefully to each audio clip—focus on how many repetitions of the action are in the audio. Which audio clip best matches this number of repetitions? Choose A, B, C, or D.\nA: <media_0>\nB: <media_1>\nC: <media_2>\nD: <media_3>\nAnswer with the option's letter (A, B, C, or D) directly."}