add internvl-2.5-8B

constants.py

@@ -115,6 +115,7 @@ MODEL_NAME_MAP = {
     "POINTS_15_7B": "POINTS-1.5-8B",
     "InternVL2_5_78B": "InternVL2.5-78B",
     "InternVL2_5_2B": "InternVL2.5-2B",
+    "InternVL2_5_8B": "InternVL2.5-8B",
 }
 
 DIMENSION_NAME_MAP = {
@@ -201,15 +202,16 @@ MODEL_URLS = {
     "Mammoth_VL": "https://huggingface.co/MAmmoTH-VL/MAmmoTH-VL-8B",
     "InternVL2_5_78B": "https://huggingface.co/OpenGVLab/InternVL2_5-78B",
     "InternVL2_5_2B": "https://huggingface.co/OpenGVLab/InternVL2_5-2B",
+    "InternVL2_5_8B": "https://huggingface.co/OpenGVLab/InternVL2_5-8B",
 }
 
 # Define the base MODEL_GROUPS structure
 BASE_MODEL_GROUPS = {
     "All": list(MODEL_NAME_MAP.keys()),
     "Flagship Models": ['Claude_3.5_new', 'GPT_4o', 'Claude_3.5', 'Gemini_1.5_pro_002', 'Qwen2_VL_72B', 'InternVL2_76B', 'llava_onevision_72B', 'NVLM', 'Molmo_72B', 'InternVL2_5_78B'],
-    "Efficiency Models": ['Gemini_1.5_flash_002', 'GPT_4o_mini', 'Qwen2_VL_7B', 'Pixtral_12B', 'Aria', 'InternVL2_8B', 'Phi-3.5-vision', 'MiniCPM_v2.6', 'llava_onevision_7B', 'Llama_3_2_11B', 'Idefics3', 'Molmo_7B_D', "Aquila_VL_2B", "POINTS_7B", "Qwen2_VL_2B", "InternVL2_2B", "InternVL2_5_2B"],
+    "Efficiency Models": ['Gemini_1.5_flash_002', 'GPT_4o_mini', 'Qwen2_VL_7B', 'Pixtral_12B', 'Aria', 'InternVL2_8B', 'Phi-3.5-vision', 'MiniCPM_v2.6', 'llava_onevision_7B', 'Llama_3_2_11B', 'Idefics3', 'Molmo_7B_D', "Aquila_VL_2B", "POINTS_7B", "Qwen2_VL_2B", "InternVL2_2B", "InternVL2_5_2B", "InternVL2_5_8B"],
     "Proprietary Flagship models": ['Claude_3.5_new', 'GPT_4o', 'Claude_3.5', 'Gemini_1.5_pro_002'],
     "Proprietary Efficiency Models": ['Gemini_1.5_flash_002', 'GPT_4o_mini'],
     "Open-source Flagship Models": ['Qwen2_VL_72B', 'InternVL2_76B', 'llava_onevision_72B', 'NVLM', "Molmo_72B", "InternVL2_5_78B"],
-    "Open-source Efficiency Models": ['Qwen2_VL_7B', 'Pixtral_12B', 'Aria', 'InternVL2_8B', 'Phi-3.5-vision', 'MiniCPM_v2.6', 'llava_onevision_7B', 'Llama_3_2_11B', 'Idefics3', 'Molmo_7B_D', "Aquila_VL_2B", "POINTS_7B", "Qwen2_VL_2B", "InternVL2_2B", "InternVL2_5_2B"]
-}
+    "Open-source Efficiency Models": ['Qwen2_VL_7B', 'Pixtral_12B', 'Aria', 'InternVL2_8B', 'Phi-3.5-vision', 'MiniCPM_v2.6', 'llava_onevision_7B', 'Llama_3_2_11B', 'Idefics3', 'Molmo_7B_D', "Aquila_VL_2B", "POINTS_7B", "Qwen2_VL_2B", "InternVL2_2B", "InternVL2_5_2B", "InternVL2_5_8B"]
+}

static/eval_results/Default/InternVL2_5_8B/summary_results.json

@@ -0,0 +1,249 @@
+{
+    "model_summary": {
+        "core": {
+            "num_eval_tasks": 440,
+            "num_eval_samples": 6539,
+            "macro_mean_score": 0.28341178736010597
+        },
+        "open": {
+            "num_eval_tasks": 65,
+            "num_eval_samples": 1163,
+            "macro_mean_score": 0.4427412542642981
+        },
+        "overall_score": 0.3039195405259921
+    },
+    "keyword_stats": {
+        "skills": {
+            "Object Recognition and Classification": {
+                "count": 303,
+                "num_samples": 4755,
+                "tasks": [],
+                "average_score": 0.30632267145420167
+            },
+            "Text Recognition (OCR)": {
+                "count": 137,
+                "num_samples": 2239,
+                "tasks": [],
+                "average_score": 0.32780231152819705
+            },
+            "Language Understanding and Generation": {
+                "count": 154,
+                "num_samples": 2509,
+                "tasks": [],
+                "average_score": 0.3474357934495928
+            },
+            "Scene and Event Understanding": {
+                "count": 154,
+                "num_samples": 2467,
+                "tasks": [],
+                "average_score": 0.3581812555811988
+            },
+            "Mathematical and Logical Reasoning": {
+                "count": 109,
+                "num_samples": 1910,
+                "tasks": [],
+                "average_score": 0.24957753120558315
+            },
+            "Commonsense and Social Reasoning": {
+                "count": 51,
+                "num_samples": 855,
+                "tasks": [],
+                "average_score": 0.4216233765938271
+            },
+            "Ethical and Safety Reasoning": {
+                "count": 15,
+                "num_samples": 245,
+                "tasks": [],
+                "average_score": 0.5161152882205514
+            },
+            "Domain-Specific Knowledge and Skills": {
+                "count": 77,
+                "num_samples": 1386,
+                "tasks": [],
+                "average_score": 0.2832458606328222
+            },
+            "Spatial and Temporal Reasoning": {
+                "count": 152,
+                "num_samples": 2437,
+                "tasks": [],
+                "average_score": 0.23652758853070416
+            },
+            "Planning and Decision Making": {
+                "count": 37,
+                "num_samples": 577,
+                "tasks": [],
+                "average_score": 0.1316894515451977
+            }
+        },
+        "input_format": {
+            "User Interface Screenshots": {
+                "count": 93,
+                "num_samples": 1517,
+                "tasks": [],
+                "average_score": 0.3170058923831629
+            },
+            "Text-Based Images and Documents": {
+                "count": 82,
+                "num_samples": 1294,
+                "tasks": [],
+                "average_score": 0.22476699307920894
+            },
+            "Diagrams and Data Visualizations": {
+                "count": 101,
+                "num_samples": 1718,
+                "tasks": [],
+                "average_score": 0.27470636143635613
+            },
+            "Videos": {
+                "count": 43,
+                "num_samples": 698,
+                "tasks": [],
+                "average_score": 0.358656962172874
+            },
+            "Artistic and Creative Content": {
+                "count": 32,
+                "num_samples": 541,
+                "tasks": [],
+                "average_score": 0.3109129114251568
+            },
+            "Photographs": {
+                "count": 143,
+                "num_samples": 2248,
+                "tasks": [],
+                "average_score": 0.3559644169537599
+            },
+            "3D Models and Aerial Imagery": {
+                "count": 11,
+                "num_samples": 169,
+                "tasks": [],
+                "average_score": 0.14065544299986515
+            }
+        },
+        "output_format": {
+            "contextual_formatted_text": {
+                "count": 98,
+                "num_samples": 1514,
+                "tasks": [],
+                "average_score": 0.2593999929737164
+            },
+            "structured_output": {
+                "count": 110,
+                "num_samples": 1714,
+                "tasks": [],
+                "average_score": 0.24115694869183088
+            },
+            "exact_text": {
+                "count": 83,
+                "num_samples": 1278,
+                "tasks": [],
+                "average_score": 0.32751507985720435
+            },
+            "numerical_data": {
+                "count": 49,
+                "num_samples": 862,
+                "tasks": [],
+                "average_score": 0.28350382022275183
+            },
+            "open_ended_output": {
+                "count": 80,
+                "num_samples": 1454,
+                "tasks": [],
+                "average_score": 0.39498520559767875
+            },
+            "multiple_choice": {
+                "count": 85,
+                "num_samples": 1363,
+                "tasks": [],
+                "average_score": 0.3394899886026274
+            }
+        },
+        "input_num": {
+            "6-8 images": {
+                "count": 21,
+                "num_samples": 314,
+                "tasks": [],
+                "average_score": 0.12507018680488066
+            },
+            "9-image or more": {
+                "count": 41,
+                "num_samples": 623,
+                "tasks": [],
+                "average_score": 0.2869727946384576
+            },
+            "1-image": {
+                "count": 315,
+                "num_samples": 5228,
+                "tasks": [],
+                "average_score": 0.3332167660932209
+            },
+            "video": {
+                "count": 43,
+                "num_samples": 698,
+                "tasks": [],
+                "average_score": 0.358656962172874
+            },
+            "4-5 images": {
+                "count": 34,
+                "num_samples": 520,
+                "tasks": [],
+                "average_score": 0.17503971457329898
+            },
+            "2-3 images": {
+                "count": 51,
+                "num_samples": 802,
+                "tasks": [],
+                "average_score": 0.25000251910306803
+            }
+        },
+        "app": {
+            "Information_Extraction": {
+                "count": 72,
+                "num_samples": 1124,
+                "tasks": [],
+                "average_score": 0.3510200305843745
+            },
+            "Planning": {
+                "count": 78,
+                "num_samples": 1239,
+                "tasks": [],
+                "average_score": 0.1596951437265508
+            },
+            "Coding": {
+                "count": 31,
+                "num_samples": 474,
+                "tasks": [],
+                "average_score": 0.25454014939309055
+            },
+            "Perception": {
+                "count": 145,
+                "num_samples": 2313,
+                "tasks": [],
+                "average_score": 0.332705158221202
+            },
+            "Metrics": {
+                "count": 20,
+                "num_samples": 309,
+                "tasks": [],
+                "average_score": 0.4496016958712894
+            },
+            "Science": {
+                "count": 29,
+                "num_samples": 574,
+                "tasks": [],
+                "average_score": 0.28828525298916796
+            },
+            "Knowledge": {
+                "count": 97,
+                "num_samples": 1605,
+                "tasks": [],
+                "average_score": 0.3477512139656071
+            },
+            "Mathematics": {
+                "count": 33,
+                "num_samples": 547,
+                "tasks": [],
+                "average_score": 0.25856175669225717
+            }
+        }
+    }
+}