Add files using upload-large-folder tool

datasets/internvl_format/stage1/stage1_withstandard_analysisbehind.json

@@ -0,0 +1,10 @@
+{
+  "stage1_withstandard_analysisbehind": {
+    "root": "/cpfs/user/chenziyang/Libra-V2/datasets/librav_trainset_stage1_images",
+    "annotation": "/cpfs/user/chenziyang/Libra-V2/datasets/internvl/stage1/train.jsonl",
+    "data_augment": false,
+    "max_dynamic_patch": 12,
+    "repeat_time": 1,
+    "length": 216572
+  }
+}

datasets/internvl_format/stage2/stage2_withstandard_analysisbehind.json

@@ -0,0 +1,10 @@
+{
+  "stage2_withstandard_analysisbehind": {
+    "root": "/cpfs/user/chenziyang/Libra-V2/datasets/librav_trainset_stage2_images",
+    "annotation": "/cpfs/user/chenziyang/Libra-V2/datasets/internvl/stage2/train.jsonl",
+    "data_augment": false,
+    "max_dynamic_patch": 12,
+    "repeat_time": 1,
+    "length": 19972
+  }
+}

datasets/llamafactory_format/stage1/dataset_info.json

@@ -0,0 +1,92 @@
+{
+  "librav_stage1_withstandard_analysisbehind": {
+    "file_name": "datas/stage1_withstandard_analysisbehind.json",
+    "formatting": "sharegpt",
+    "columns": {
+      "messages": "messages",
+      "images": "images"
+    },
+    "tags": {
+      "role_tag": "role",
+      "content_tag": "content",
+      "user_tag": "user",
+      "assistant_tag": "assistant",
+      "system_tag": "system"
+    }
+  },
+  "librav_stage1_withstandard_analysisfront": {
+    "file_name": "datas/stage1_withstandard_analysisfront.json",
+    "formatting": "sharegpt",
+    "columns": {
+      "messages": "messages",
+      "images": "images"
+    },
+    "tags": {
+      "role_tag": "role",
+      "content_tag": "content",
+      "user_tag": "user",
+      "assistant_tag": "assistant",
+      "system_tag": "system"
+    }
+  },
+  "librav_stage1_withstandard_noanalysis": {
+    "file_name": "datas/stage1_withstandard_noanalysis.json",
+    "formatting": "sharegpt",
+    "columns": {
+      "messages": "messages",
+      "images": "images"
+    },
+    "tags": {
+      "role_tag": "role",
+      "content_tag": "content",
+      "user_tag": "user",
+      "assistant_tag": "assistant",
+      "system_tag": "system"
+    }
+  },
+  "librav_stage1_nostandard_analysisbehind": {
+    "file_name": "datas/stage1_nostandard_analysisbehind.json",
+    "formatting": "sharegpt",
+    "columns": {
+      "messages": "messages",
+      "images": "images"
+    },
+    "tags": {
+      "role_tag": "role",
+      "content_tag": "content",
+      "user_tag": "user",
+      "assistant_tag": "assistant",
+      "system_tag": "system"
+    }
+  },
+  "librav_stage1_nostandard_analysisfront": {
+    "file_name": "datas/stage1_nostandard_analysisfront.json",
+    "formatting": "sharegpt",
+    "columns": {
+      "messages": "messages",
+      "images": "images"
+    },
+    "tags": {
+      "role_tag": "role",
+      "content_tag": "content",
+      "user_tag": "user",
+      "assistant_tag": "assistant",
+      "system_tag": "system"
+    }
+  },
+  "librav_stage1_nostandard_noanalysis": {
+    "file_name": "datas/stage1_nostandard_noanalysis.json",
+    "formatting": "sharegpt",
+    "columns": {
+      "messages": "messages",
+      "images": "images"
+    },
+    "tags": {
+      "role_tag": "role",
+      "content_tag": "content",
+      "user_tag": "user",
+      "assistant_tag": "assistant",
+      "system_tag": "system"
+    }
+  }
+}

datasets/llamafactory_format/stage2/dataset_info.json

@@ -0,0 +1,75 @@
+{
+  "librav_stage2_withstandard_analysisbehind": {
+    "file_name": "datas/stage2_withstandard_analysisbehind.json",
+    "formatting": "sharegpt",
+    "ranking": true,
+    "columns": {
+      "messages": "conversations",
+      "chosen": "chosen",
+      "rejected": "rejected",
+      "images": "images",
+      "system": "system"
+    }
+  },
+  "librav_stage2_withstandard_analysisfront": {
+    "file_name": "datas/stage2_withstandard_analysisfront.json",
+    "formatting": "sharegpt",
+    "ranking": true,
+    "columns": {
+      "messages": "conversations",
+      "chosen": "chosen",
+      "rejected": "rejected",
+      "images": "images",
+      "system": "system"
+    }
+  },
+  "librav_stage2_withstandard_noanalysis": {
+    "file_name": "datas/stage2_withstandard_noanalysis.json",
+    "formatting": "sharegpt",
+    "ranking": true,
+    "columns": {
+      "messages": "conversations",
+      "chosen": "chosen",
+      "rejected": "rejected",
+      "images": "images",
+      "system": "system"
+    }
+  },
+  "librav_stage2_nostandard_analysisbehind": {
+    "file_name": "datas/stage2_nostandard_analysisbehind.json",
+    "formatting": "sharegpt",
+    "ranking": true,
+    "columns": {
+      "messages": "conversations",
+      "chosen": "chosen",
+      "rejected": "rejected",
+      "images": "images",
+      "system": "system"
+    }
+  },
+  "librav_stage2_nostandard_analysisfront": {
+    "file_name": "datas/stage2_nostandard_analysisfront.json",
+    "formatting": "sharegpt",
+    "ranking": true,
+    "columns": {
+      "messages": "conversations",
+      "chosen": "chosen",
+      "rejected": "rejected",
+      "images": "images",
+      "system": "system"
+    }
+  },
+  "librav_stage2_nostandard_noanalysis": {
+    "file_name": "datas/stage2_nostandard_noanalysis.json",
+    "formatting": "sharegpt",
+    "ranking": true,
+    "columns": {
+      "messages": "conversations",
+      "chosen": "chosen",
+      "rejected": "rejected",
+      "images": "images",
+      "system": "system"
+    }
+  }
+}
+

eval_results/ours/librav-bench-id/stage1_nostandard_analysisbehind_3b_seed3407/stage1_nostandard_analysisbehind_3b_seed3407.txt

@@ -0,0 +1,65 @@
+------------------------------------------------------------------------------------------------------------------------------------------------------
+Prompt Source: figstep
+              precision    recall  f1-score   support
+
+          安全     0.6279    0.8182    0.7105        66
+          风险     0.5641    0.3860    0.4583        57
+         不安全     0.6897    0.6452    0.6667        31
+
+    accuracy                         0.6234       154
+   macro avg     0.6272    0.6164    0.6118       154
+weighted avg     0.6167    0.6234    0.6084       154
+
+------------------------------------------------------------------------------------------------------------------------------------------------------
+------------------------------------------------------------------------------------------------------------------------------------------------------
+Prompt Source: jailbreakv
+              precision    recall  f1-score   support
+
+          安全     0.7955    0.9211    0.8537        76
+          风险     0.6538    0.5152    0.5763        33
+         不安全     0.8077    0.6774    0.7368        31
+
+    accuracy                         0.7714       140
+   macro avg     0.7523    0.7045    0.7223       140
+weighted avg     0.7648    0.7714    0.7624       140
+
+------------------------------------------------------------------------------------------------------------------------------------------------------
+------------------------------------------------------------------------------------------------------------------------------------------------------
+Prompt Source: mmsafetybench
+              precision    recall  f1-score   support
+
+          安全     0.8125    0.9286    0.8667        70
+          风险     0.7759    0.7627    0.7692        59
+         不安全     0.9091    0.6452    0.7547        31
+
+    accuracy                         0.8125       160
+   macro avg     0.8325    0.7788    0.7969       160
+weighted avg     0.8177    0.8125    0.8090       160
+
+------------------------------------------------------------------------------------------------------------------------------------------------------
+------------------------------------------------------------------------------------------------------------------------------------------------------
+Prompt Source: vlsbench
+              precision    recall  f1-score   support
+
+          安全     0.7403    0.9048    0.8143        63
+          风险     0.6136    0.5294    0.5684        51
+         不安全     0.8571    0.7347    0.7912        49
+
+    accuracy                         0.7362       163
+   macro avg     0.7370    0.7230    0.7246       163
+weighted avg     0.7358    0.7362    0.7304       163
+
+------------------------------------------------------------------------------------------------------------------------------------------------------
+------------------------------------------------------------------------------------------------------------------------------------------------------
+Prompt Source: all
+              precision    recall  f1-score   support
+
+          安全     0.7432    0.8945    0.8119       275
+          风险     0.6647    0.5550    0.6049       200
+         不安全     0.8151    0.6831    0.7433       142
+
+    accuracy                         0.7358       617
+   macro avg     0.7410    0.7109    0.7200       617
+weighted avg     0.7343    0.7358    0.7290       617
+
+------------------------------------------------------------------------------------------------------------------------------------------------------

eval_results/ours/librav-bench-id/stage1_nostandard_analysisfront_3b_seed3407/stage1_nostandard_analysisfront_3b_seed3407.txt

@@ -0,0 +1,65 @@
+------------------------------------------------------------------------------------------------------------------------------------------------------
+Prompt Source: figstep
+              precision    recall  f1-score   support
+
+          安全     0.6250    0.7576    0.6849        66
+          风险     0.6122    0.5263    0.5660        57
+         不安全     0.7200    0.5806    0.6429        31
+
+    accuracy                         0.6364       154
+   macro avg     0.6524    0.6215    0.6313       154
+weighted avg     0.6394    0.6364    0.6325       154
+
+------------------------------------------------------------------------------------------------------------------------------------------------------
+------------------------------------------------------------------------------------------------------------------------------------------------------
+Prompt Source: jailbreakv
+              precision    recall  f1-score   support
+
+          安全     0.8140    0.9211    0.8642        76
+          风险     0.5667    0.5152    0.5397        33
+         不安全     0.7500    0.5806    0.6545        31
+
+    accuracy                         0.7500       140
+   macro avg     0.7102    0.6723    0.6861       140
+weighted avg     0.7415    0.7500    0.7413       140
+
+------------------------------------------------------------------------------------------------------------------------------------------------------
+------------------------------------------------------------------------------------------------------------------------------------------------------
+Prompt Source: mmsafetybench
+              precision    recall  f1-score   support
+
+          安全     0.8125    0.9286    0.8667        70
+          风险     0.7500    0.7627    0.7563        59
+         不安全     0.9500    0.6129    0.7451        31
+
+    accuracy                         0.8063       160
+   macro avg     0.8375    0.7681    0.7894       160
+weighted avg     0.8161    0.8063    0.8024       160
+
+------------------------------------------------------------------------------------------------------------------------------------------------------
+------------------------------------------------------------------------------------------------------------------------------------------------------
+Prompt Source: vlsbench
+              precision    recall  f1-score   support
+
+          安全     0.6988    0.9206    0.7945        63
+          风险     0.5714    0.4706    0.5161        51
+         不安全     0.9211    0.7143    0.8046        49
+
+    accuracy                         0.7178       163
+   macro avg     0.7304    0.7018    0.7051       163
+weighted avg     0.7258    0.7178    0.7104       163
+
+------------------------------------------------------------------------------------------------------------------------------------------------------
+------------------------------------------------------------------------------------------------------------------------------------------------------
+Prompt Source: all
+              precision    recall  f1-score   support
+
+          安全     0.7386    0.8836    0.8046       275
+          风险     0.6409    0.5800    0.6089       200
+         不安全     0.8411    0.6338    0.7229       142
+
+    accuracy                         0.7277       617
+   macro avg     0.7402    0.6991    0.7122       617
+weighted avg     0.7305    0.7277    0.7224       617
+
+------------------------------------------------------------------------------------------------------------------------------------------------------