diff --git a/.gitattributes b/.gitattributes index dafdfd192f82399957ed9ed8ade0162a8a06ae20..c28b69282f8f30e9a1ffe70e71c4a640948fd04f 100644 --- a/.gitattributes +++ b/.gitattributes @@ -47,3 +47,19 @@ llava_phi_c09/llava_phi_c09_SEEDBench_IMG_openai_result.xlsx filter=lfs diff=lfs llava_phi_c10/llava_phi_c10_SEEDBench_IMG_openai_result.xlsx filter=lfs diff=lfs merge=lfs -text vicuna7b_moe_llava/vicuna7b_moe_llava_SEEDBench_IMG.xlsx filter=lfs diff=lfs merge=lfs -text vicuna7b_moe_llava/vicuna7b_moe_llava_SEEDBench_IMG_openai_result.xlsx filter=lfs diff=lfs merge=lfs -text +llava-phi-smoe/llava-phi-smoe_A-OKVQA_openai_result.xlsx filter=lfs diff=lfs merge=lfs -text +llava-phi-smoe/llava-phi-smoe_TextVQA_VAL.xlsx filter=lfs diff=lfs merge=lfs -text +llava_phi3_5/llava_phi3_5_A-OKVQA.xlsx filter=lfs diff=lfs merge=lfs -text +llava_phi3_5/llava_phi3_5_A-OKVQA_openai_result.xlsx filter=lfs diff=lfs merge=lfs -text +llava_phi3_5/llava_phi3_5_AI2D_TEST.xlsx filter=lfs diff=lfs merge=lfs -text +llava_phi3_5/llava_phi3_5_AI2D_TEST_openai_result.xlsx filter=lfs diff=lfs merge=lfs -text +llava_phi3_5/llava_phi3_5_MME.xlsx filter=lfs diff=lfs merge=lfs -text +llava_phi3_5/llava_phi3_5_MME_auxmatch.xlsx filter=lfs diff=lfs merge=lfs -text +llava_phi3_5/llava_phi3_5_MMMU_DEV_VAL.xlsx filter=lfs diff=lfs merge=lfs -text +llava_phi3_5/llava_phi3_5_MMMU_DEV_VAL_openai_result.xlsx filter=lfs diff=lfs merge=lfs -text +llava_phi3_5/llava_phi3_5_POPE.xlsx filter=lfs diff=lfs merge=lfs -text +llava_phi3_5/llava_phi3_5_POPE_auxmatch.xlsx filter=lfs diff=lfs merge=lfs -text +llava_phi3_5/llava_phi3_5_Q-Bench1_VAL_openai_result.xlsx filter=lfs diff=lfs merge=lfs -text +llava_phi3_5/llava_phi3_5_ScienceQA_VAL.xlsx filter=lfs diff=lfs merge=lfs -text +llava_phi3_5/llava_phi3_5_ScienceQA_VAL_openai_result.xlsx filter=lfs diff=lfs merge=lfs -text +llava_phi3_5/llava_phi3_5_TextVQA_VAL.xlsx filter=lfs diff=lfs merge=lfs -text diff --git a/llava-phi-smoe/01_SEEDBench_IMG.pkl b/llava-phi-smoe/01_SEEDBench_IMG.pkl new file mode 100644 index 0000000000000000000000000000000000000000..67b85630ce2320d3adac787c50f29a3fc1d90552 --- /dev/null +++ b/llava-phi-smoe/01_SEEDBench_IMG.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3a881656a2e630e9b511336af746b517f155f69eef6e7eb9fe3708daad77178f +size 399293 diff --git a/llava-phi-smoe/01_ScienceQA_VAL.pkl b/llava-phi-smoe/01_ScienceQA_VAL.pkl new file mode 100644 index 0000000000000000000000000000000000000000..b559337b97d92dbcebb372cd3e029cdac1b42b81 --- /dev/null +++ b/llava-phi-smoe/01_ScienceQA_VAL.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aad2ccd66f337b9e38a3dafba3a8ef2cc238c3ca23755dcb39b76f34a441a3ed +size 330895 diff --git a/llava-phi-smoe/llava-phi-smoe_A-OKVQA.xlsx b/llava-phi-smoe/llava-phi-smoe_A-OKVQA.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..92906683a0374381541e60b407a7a060469ee1a6 Binary files /dev/null and b/llava-phi-smoe/llava-phi-smoe_A-OKVQA.xlsx differ diff --git a/llava-phi-smoe/llava-phi-smoe_A-OKVQA_acc.csv b/llava-phi-smoe/llava-phi-smoe_A-OKVQA_acc.csv new file mode 100644 index 0000000000000000000000000000000000000000..13ba5fdc515ea39aba96f352c8357d87f708dc7b --- /dev/null +++ b/llava-phi-smoe/llava-phi-smoe_A-OKVQA_acc.csv @@ -0,0 +1,2 @@ +"split","Overall","ALL" +"val","0.2576419213973799","0.2576419213973799" diff --git a/llava-phi-smoe/llava-phi-smoe_A-OKVQA_openai_result.pkl b/llava-phi-smoe/llava-phi-smoe_A-OKVQA_openai_result.pkl new file mode 100644 index 0000000000000000000000000000000000000000..77b7cda2f7979a6e413af458f1fd072c52957cf8 --- /dev/null +++ b/llava-phi-smoe/llava-phi-smoe_A-OKVQA_openai_result.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:37f0c74611011a0587784a129fd6c55c05959fceb83409e37bb2734fe7b48727 +size 71102 diff --git a/llava-phi-smoe/llava-phi-smoe_A-OKVQA_openai_result.xlsx b/llava-phi-smoe/llava-phi-smoe_A-OKVQA_openai_result.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..7ab6dd3b53f2b62010d54746dfb2e816ecdb3fc5 --- /dev/null +++ b/llava-phi-smoe/llava-phi-smoe_A-OKVQA_openai_result.xlsx @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5704fd2fbc4dc80a61a50eb50c10d624fd856b48f4a5f3b8215606bdbe15faaa +size 102691 diff --git a/llava-phi-smoe/llava-phi-smoe_HallusionBench.xlsx b/llava-phi-smoe/llava-phi-smoe_HallusionBench.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..4b700555f045d56a72a92f22d9722b4cd17dfdeb Binary files /dev/null and b/llava-phi-smoe/llava-phi-smoe_HallusionBench.xlsx differ diff --git a/llava-phi-smoe/llava-phi-smoe_HallusionBench_auxmatch.xlsx b/llava-phi-smoe/llava-phi-smoe_HallusionBench_auxmatch.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..c9509482ce3f19742caa7ea4ad1619d38992d502 Binary files /dev/null and b/llava-phi-smoe/llava-phi-smoe_HallusionBench_auxmatch.xlsx differ diff --git a/llava-phi-smoe/llava-phi-smoe_HallusionBench_score.csv b/llava-phi-smoe/llava-phi-smoe_HallusionBench_score.csv new file mode 100644 index 0000000000000000000000000000000000000000..8a6e9f78b9d668fbd1ac478569b0cdefef8f6c8e --- /dev/null +++ b/llava-phi-smoe/llava-phi-smoe_HallusionBench_score.csv @@ -0,0 +1,13 @@ +"split","aAcc","fAcc","qAcc" +"Overall","0.0","0.0","0.0" +"VS","0.0","0.0","0.0" +"VD","0.0","0.0","0.0" +"VD_video","0.0","0.0","0.0" +"VS_chart","0.0","0.0","0.0" +"VD_math","0.0","0.0","0.0" +"VS_table","0.0","0.0","0.0" +"VD_figure","0.0","0.0","0.0" +"VS_map","0.0","0.0","0.0" +"VS_ocr","0.0","0.0","0.0" +"VD_ocr","0.0","0.0","0.0" +"VD_illusion","0.0","0.0","0.0" diff --git a/llava-phi-smoe/llava-phi-smoe_RealWorldQA.xlsx b/llava-phi-smoe/llava-phi-smoe_RealWorldQA.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..014b7ca9683c500dd9b11b67637b4df0962a82ca Binary files /dev/null and b/llava-phi-smoe/llava-phi-smoe_RealWorldQA.xlsx differ diff --git a/llava-phi-smoe/llava-phi-smoe_RealWorldQA_acc.csv b/llava-phi-smoe/llava-phi-smoe_RealWorldQA_acc.csv new file mode 100644 index 0000000000000000000000000000000000000000..07d753d5d4ea06a7bff03dc9fd485dc22ba48929 --- /dev/null +++ b/llava-phi-smoe/llava-phi-smoe_RealWorldQA_acc.csv @@ -0,0 +1,2 @@ +"split","Overall" +"none","0.16470588235294117" diff --git a/llava-phi-smoe/llava-phi-smoe_RealWorldQA_openai_result.pkl b/llava-phi-smoe/llava-phi-smoe_RealWorldQA_openai_result.pkl new file mode 100644 index 0000000000000000000000000000000000000000..c040bc9606cbcdda24eff572cf7c6b127a9152d7 --- /dev/null +++ b/llava-phi-smoe/llava-phi-smoe_RealWorldQA_openai_result.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cb8894801480ad9d5e676e4ac015a45694a7e5d8d1cb724319d0732cc9b63a0d +size 73580 diff --git a/llava-phi-smoe/llava-phi-smoe_RealWorldQA_openai_result.xlsx b/llava-phi-smoe/llava-phi-smoe_RealWorldQA_openai_result.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..4a83a0491a758ba45709c1cad4856cff44f1f7b8 Binary files /dev/null and b/llava-phi-smoe/llava-phi-smoe_RealWorldQA_openai_result.xlsx differ diff --git a/llava-phi-smoe/llava-phi-smoe_TextVQA_VAL.xlsx b/llava-phi-smoe/llava-phi-smoe_TextVQA_VAL.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..693b9bfb1a880a128bbd0901495265027bd90d0c --- /dev/null +++ b/llava-phi-smoe/llava-phi-smoe_TextVQA_VAL.xlsx @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c3d9024043cf6c0677f05887175d5f1b52ba88d080d5764044d36e52496578e +size 321174 diff --git a/llava-phi-smoe/llava-phi-smoe_TextVQA_VAL_acc.csv b/llava-phi-smoe/llava-phi-smoe_TextVQA_VAL_acc.csv new file mode 100644 index 0000000000000000000000000000000000000000..4a6a442415a21f787ed8e2fc831a0e7c51ac7aa2 --- /dev/null +++ b/llava-phi-smoe/llava-phi-smoe_TextVQA_VAL_acc.csv @@ -0,0 +1,2 @@ +"Overall" +"0.0" diff --git a/llava_phi3_5/01_SEEDBench_IMG.pkl b/llava_phi3_5/01_SEEDBench_IMG.pkl new file mode 100644 index 0000000000000000000000000000000000000000..da7036453f9a5ac024f0fada0b1efc16e441ebe9 --- /dev/null +++ b/llava_phi3_5/01_SEEDBench_IMG.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4ab39c503e4ed74ab9bb9289c676f60095a5d0f26d988c4a0dbfdce98e726398 +size 519529 diff --git a/llava_phi3_5/llava_phi3_5_A-OKVQA.xlsx b/llava_phi3_5/llava_phi3_5_A-OKVQA.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..ede6f5fe139014badced9a337033985ca9638d9d --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_A-OKVQA.xlsx @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:50d49b6fe9dedae2756cbd1e8d32faf7ccf5285b3e5d342d45ddc8975ea988b9 +size 102619 diff --git a/llava_phi3_5/llava_phi3_5_A-OKVQA_acc.csv b/llava_phi3_5/llava_phi3_5_A-OKVQA_acc.csv new file mode 100644 index 0000000000000000000000000000000000000000..6fdf359f1543aa137c3b6ea4bd8e1a1cb632429a --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_A-OKVQA_acc.csv @@ -0,0 +1,2 @@ +"split","Overall","ALL" +"val","0.7502183406113537","0.7502183406113537" diff --git a/llava_phi3_5/llava_phi3_5_A-OKVQA_openai_result.pkl b/llava_phi3_5/llava_phi3_5_A-OKVQA_openai_result.pkl new file mode 100644 index 0000000000000000000000000000000000000000..8c5053c7ab5a314724b727c4062193a32f8ac0cd --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_A-OKVQA_openai_result.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:039c08df0a86b19f52a3e5647a408143f12fe68d7366bce5b0b70719e210e186 +size 81753 diff --git a/llava_phi3_5/llava_phi3_5_A-OKVQA_openai_result.xlsx b/llava_phi3_5/llava_phi3_5_A-OKVQA_openai_result.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..0909bf1dac6f6793c2464a8d7ccb2947ae83513b --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_A-OKVQA_openai_result.xlsx @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f3a3fd70e5e9e59b58d462a56b4c73cfeb1e53d1a22b4d16414dd1b39ce65f81 +size 121307 diff --git a/llava_phi3_5/llava_phi3_5_AI2D_TEST.xlsx b/llava_phi3_5/llava_phi3_5_AI2D_TEST.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..65b3403e571abf6fdf552680d8571e2ccc7f5925 --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_AI2D_TEST.xlsx @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aa6e061382313340b43b5c0c5f6824f05b5a4a2443db9d47a020512953c74d2f +size 262388 diff --git a/llava_phi3_5/llava_phi3_5_AI2D_TEST_acc.csv b/llava_phi3_5/llava_phi3_5_AI2D_TEST_acc.csv new file mode 100644 index 0000000000000000000000000000000000000000..2e261886dceea0fa296d1df965167512bec947a2 --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_AI2D_TEST_acc.csv @@ -0,0 +1,2 @@ +"split","Overall","atomStructure","eclipses","faultsEarthquakes","foodChainsWebs","lifeCycles","moonPhaseEquinox","partsOfA","partsOfTheEarth","photosynthesisRespiration","rockCycle","rockStrata","solarSystem","typesOf","volcano","waterCNPCycle" +"none","0.5952072538860104","0.75","0.8064516129032258","0.7142857142857143","0.7271944922547332","0.4460431654676259","0.4259927797833935","0.6160164271047228","0.5576923076923077","0.43037974683544306","0.47761194029850745","0.5121951219512195","0.8055555555555556","0.4839650145772595","0.5625","0.4090909090909091" diff --git a/llava_phi3_5/llava_phi3_5_AI2D_TEST_openai_result.pkl b/llava_phi3_5/llava_phi3_5_AI2D_TEST_openai_result.pkl new file mode 100644 index 0000000000000000000000000000000000000000..ad4ee01b9e9c1b651760faf48bdb342008ebb3d5 --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_AI2D_TEST_openai_result.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:674e89eef6647355f67cb838ed2afa85dd8f776ff4568e43fb6ce103c0f535be +size 235462 diff --git a/llava_phi3_5/llava_phi3_5_AI2D_TEST_openai_result.xlsx b/llava_phi3_5/llava_phi3_5_AI2D_TEST_openai_result.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..6c2ffbcb4dd570be3d15cbdf2728cc570ab2bd56 --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_AI2D_TEST_openai_result.xlsx @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0f1569d2d1d93635e070d1c9184c4d6cec657fdf0703a764baec0a9bbe1c876b +size 300775 diff --git a/llava_phi3_5/llava_phi3_5_HallusionBench.xlsx b/llava_phi3_5/llava_phi3_5_HallusionBench.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..8d33d44b15485d19be69774a29ac8a621b8978a0 Binary files /dev/null and b/llava_phi3_5/llava_phi3_5_HallusionBench.xlsx differ diff --git a/llava_phi3_5/llava_phi3_5_HallusionBench_auxmatch.xlsx b/llava_phi3_5/llava_phi3_5_HallusionBench_auxmatch.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..e3882011d39fca8bea957f30397e3e4bb360fcc2 Binary files /dev/null and b/llava_phi3_5/llava_phi3_5_HallusionBench_auxmatch.xlsx differ diff --git a/llava_phi3_5/llava_phi3_5_HallusionBench_score.csv b/llava_phi3_5/llava_phi3_5_HallusionBench_score.csv new file mode 100644 index 0000000000000000000000000000000000000000..6c3c8a51e2486217e5346957708df607ffed668a --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_HallusionBench_score.csv @@ -0,0 +1,13 @@ +"split","aAcc","fAcc","qAcc" +"Overall","51.10410094637224","21.67630057803468","18.461538461538463" +"VS","42.5","7.758620689655173","15.730337078651685" +"VD","56.34517766497462","28.695652173913043","20.216606498194945" +"VS_ocr","46.2962962962963","19.230769230769234","11.11111111111111" +"VD_ocr","65.1685393258427","30.23255813953488","27.906976744186046" +"VS_chart","37.69230769230769","5.0","22.36842105263158" +"VD_illusion","59.72222222222222","40.32258064516129","23.61111111111111" +"VD_video","53.529411764705884","14.583333333333334","8.695652173913043" +"VD_math","44.44444444444444","19.444444444444446","20.37037037037037" +"VS_map","54.6875","9.090909090909092","15.625" +"VD_figure","62.5","34.146341463414636","25.64102564102564" +"VS_table","39.285714285714285","0.0","6.976744186046512" diff --git a/llava_phi3_5/llava_phi3_5_MME.xlsx b/llava_phi3_5/llava_phi3_5_MME.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..7091f5104f1ae1bafb6697b84179e02ed9bf3dd6 --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_MME.xlsx @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa1f98f7a215c47bc188a0bddb06fb0e962ef957118669041908903e95a4fe20 +size 115518 diff --git a/llava_phi3_5/llava_phi3_5_MME_auxmatch.xlsx b/llava_phi3_5/llava_phi3_5_MME_auxmatch.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..1ca5d8d18461d6f710a700baaf130add8eec472e --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_MME_auxmatch.xlsx @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ce026cf5eca609ba39050fa3d50949573602a07075be58406173b00a3f8b2603 +size 131716 diff --git a/llava_phi3_5/llava_phi3_5_MME_score.csv b/llava_phi3_5/llava_phi3_5_MME_score.csv new file mode 100644 index 0000000000000000000000000000000000000000..4bb0c82525fa75f312e10b97f87f3d85d125113c --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_MME_score.csv @@ -0,0 +1,2 @@ +"perception","reasoning","OCR","artwork","celebrity","code_reasoning","color","commonsense_reasoning","count","existence","landmark","numerical_calculation","position","posters","scene","text_translation" +"1049.1003401360545","256.07142857142856","87.5","63.50000000000001","105.0","55.0","135.0","93.57142857142857","118.33333333333333","150.0","96.25","40.0","55.00000000000001","92.51700680272108","146.0","67.5" diff --git a/llava_phi3_5/llava_phi3_5_MMMU_DEV_VAL.xlsx b/llava_phi3_5/llava_phi3_5_MMMU_DEV_VAL.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..034bd8840fde01bb759c0f8d1c4724f727745e7c --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_MMMU_DEV_VAL.xlsx @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:640e3bed20facc22450a08dc51637a513149ad70ac5b7458882ae33c3920d798 +size 278137 diff --git a/llava_phi3_5/llava_phi3_5_MMMU_DEV_VAL_acc.csv b/llava_phi3_5/llava_phi3_5_MMMU_DEV_VAL_acc.csv new file mode 100644 index 0000000000000000000000000000000000000000..a70c3e9c1121087b2d6b922edb2b72737f97dffc --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_MMMU_DEV_VAL_acc.csv @@ -0,0 +1,3 @@ +"split","Overall","Accounting","Agriculture","Architecture_and_Engineering","Art","Art_Theory","Basic_Medical_Science","Biology","Chemistry","Clinical_Medicine","Computer_Science","Design","Diagnostics_and_Laboratory_Medicine","Economics","Electronics","Energy_and_Power","Finance","Geography","History","Literature","Manage","Marketing","Materials","Math","Mechanical_Engineering","Music","Pharmacy","Physics","Psychology","Public_Health","Sociology","Art & Design","Business","Health & Medicine","Humanities & Social Science","Science","Tech & Engineering" +"dev","0.4","0.4","0.4","0.0","0.4","0.4","1.0","0.4","0.4","0.0","0.2","0.8","0.2","0.8","0.2","0.6","0.0","0.0","0.6","0.6","0.4","0.4","0.2","0.2","0.6","0.2","0.2","0.4","0.8","0.6","0.6","0.45","0.4","0.4","0.65","0.28","0.3142857142857143" +"validation","0.3611111111111111","0.4","0.5666666666666667","0.3333333333333333","0.6","0.5","0.3333333333333333","0.26666666666666666","0.26666666666666666","0.23333333333333334","0.16666666666666666","0.5333333333333333","0.4","0.2","0.26666666666666666","0.3","0.2","0.23333333333333334","0.4666666666666667","0.7","0.2","0.3333333333333333","0.3333333333333333","0.3333333333333333","0.36666666666666664","0.4","0.1","0.3333333333333333","0.4","0.4","0.6666666666666666","0.5083333333333333","0.26666666666666666","0.29333333333333333","0.5583333333333333","0.2866666666666667","0.3333333333333333" diff --git a/llava_phi3_5/llava_phi3_5_MMMU_DEV_VAL_openai_result.pkl b/llava_phi3_5/llava_phi3_5_MMMU_DEV_VAL_openai_result.pkl new file mode 100644 index 0000000000000000000000000000000000000000..0c49aaa6dd0b79d54d51214a8ab4889ae70407b2 --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_MMMU_DEV_VAL_openai_result.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7edd43ee74fbd40f4081505b89557835062bf811f8b308f0ba048cd820fa6fb5 +size 92674 diff --git a/llava_phi3_5/llava_phi3_5_MMMU_DEV_VAL_openai_result.xlsx b/llava_phi3_5/llava_phi3_5_MMMU_DEV_VAL_openai_result.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..9f254e32bf6dae7a419378398edd662c56a7a720 --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_MMMU_DEV_VAL_openai_result.xlsx @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6d5ee4a07cedbf9657dd0403ba228f76105d26fc051afa35c5eb16f92fd6120a +size 301035 diff --git a/llava_phi3_5/llava_phi3_5_POPE.xlsx b/llava_phi3_5/llava_phi3_5_POPE.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..d1ff81bbd92e7c21952eb63c92eabf3908362413 --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_POPE.xlsx @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4b72213c240ec08b0e2e0186a3cc29699ff978191ba7cf71d58e53a67ce070db +size 172435 diff --git a/llava_phi3_5/llava_phi3_5_POPE_auxmatch.xlsx b/llava_phi3_5/llava_phi3_5_POPE_auxmatch.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..5ae83c4e00780f4a73c07e6b81b22e6ba9bf077f --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_POPE_auxmatch.xlsx @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f289ce5f913e308c0d7ffbf145839e8b4d0ce50fe6306cf8a10e76b094e3afe5 +size 207659 diff --git a/llava_phi3_5/llava_phi3_5_POPE_score.csv b/llava_phi3_5/llava_phi3_5_POPE_score.csv new file mode 100644 index 0000000000000000000000000000000000000000..3578bbbe12f66249affe309be2a2a7865961a119 --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_POPE_score.csv @@ -0,0 +1,5 @@ +"split","Overall","acc","precision","recall" +"Overall","76.24280017722641","70.21111111111111","63.40456890198969","95.6" +"popular","76.01378213623111","69.83333333333334","63.088429388473386","95.6" +"adversarial","74.84342379958247","67.86666666666666","61.492281303602056","95.6" +"random","77.93478260869564","72.93333333333332","65.77981651376147","95.6" diff --git a/llava_phi3_5/llava_phi3_5_Q-Bench1_VAL.xlsx b/llava_phi3_5/llava_phi3_5_Q-Bench1_VAL.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..5f2e78e7ef6ba55a69f8c3361571a50f8cffc399 Binary files /dev/null and b/llava_phi3_5/llava_phi3_5_Q-Bench1_VAL.xlsx differ diff --git a/llava_phi3_5/llava_phi3_5_Q-Bench1_VAL_acc.csv b/llava_phi3_5/llava_phi3_5_Q-Bench1_VAL_acc.csv new file mode 100644 index 0000000000000000000000000000000000000000..cf7a06b541e232261b7a064d08fbc243374233ee --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_Q-Bench1_VAL_acc.csv @@ -0,0 +1,2 @@ +"split","Overall","type_0_concern_0","type_0_concern_1","type_0_concern_2","type_0_concern_3","type_1_concern_0","type_1_concern_1","type_1_concern_2","type_1_concern_3","type_2_concern_0","type_2_concern_1","type_2_concern_2","type_2_concern_3" +"val","0.6120401337792643","0.6223404255319149","0.6971428571428572","0.6581196581196581","0.8571428571428571","0.5","0.7678571428571429","0.6","0.7777777777777778","0.39204545454545453","0.6","0.3793103448275862","0.6941176470588235" diff --git a/llava_phi3_5/llava_phi3_5_Q-Bench1_VAL_openai_result.pkl b/llava_phi3_5/llava_phi3_5_Q-Bench1_VAL_openai_result.pkl new file mode 100644 index 0000000000000000000000000000000000000000..ba7a2858ae116e8c38f99ad2565e90f95aa3f7b5 --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_Q-Bench1_VAL_openai_result.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c2a6e678501f856b5dcb1f870fb06cff1ce8d8febbc33c8664fcf96fb15e4032 +size 102413 diff --git a/llava_phi3_5/llava_phi3_5_Q-Bench1_VAL_openai_result.xlsx b/llava_phi3_5/llava_phi3_5_Q-Bench1_VAL_openai_result.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..79094450e045eda316c8d51ed83bd7f83baa4614 --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_Q-Bench1_VAL_openai_result.xlsx @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ae53f57132ea2f8af909d15fe820668524ca09f491c4165437bacc6fc4142d88 +size 104563 diff --git a/llava_phi3_5/llava_phi3_5_RealWorldQA.xlsx b/llava_phi3_5/llava_phi3_5_RealWorldQA.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..e8e83409b1e6b9f436d1d786aa2ab1447caad439 Binary files /dev/null and b/llava_phi3_5/llava_phi3_5_RealWorldQA.xlsx differ diff --git a/llava_phi3_5/llava_phi3_5_RealWorldQA_acc.csv b/llava_phi3_5/llava_phi3_5_RealWorldQA_acc.csv new file mode 100644 index 0000000000000000000000000000000000000000..8a19a5f08793e85ad66710ef6f20127ff7f80d4d --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_RealWorldQA_acc.csv @@ -0,0 +1,2 @@ +"split","Overall" +"none","0.5581699346405229" diff --git a/llava_phi3_5/llava_phi3_5_RealWorldQA_openai_result.pkl b/llava_phi3_5/llava_phi3_5_RealWorldQA_openai_result.pkl new file mode 100644 index 0000000000000000000000000000000000000000..34b74316bbc04882ecc0e2a2f2d6c1ea471aac42 --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_RealWorldQA_openai_result.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5c29eb7fc98f595f669d718917735a76567029248e7a5e9d6da1923205cc3b9e +size 60817 diff --git a/llava_phi3_5/llava_phi3_5_RealWorldQA_openai_result.xlsx b/llava_phi3_5/llava_phi3_5_RealWorldQA_openai_result.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..1fcdeb0e23d7340c055a0f1fd188b477bbcd4248 Binary files /dev/null and b/llava_phi3_5/llava_phi3_5_RealWorldQA_openai_result.xlsx differ diff --git a/llava_phi3_5/llava_phi3_5_ScienceQA_VAL.xlsx b/llava_phi3_5/llava_phi3_5_ScienceQA_VAL.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..31248954d76040965acf20fbea1986bd4326ec78 --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_ScienceQA_VAL.xlsx @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:75e830fe6923c8da6931a4ef6e849c1e7400486f2302d397e31b6050bdf3a314 +size 356715 diff --git a/llava_phi3_5/llava_phi3_5_ScienceQA_VAL_acc.csv b/llava_phi3_5/llava_phi3_5_ScienceQA_VAL_acc.csv new file mode 100644 index 0000000000000000000000000000000000000000..26841bf944bab03fe634d91eae8b1c90aeabbbfe --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_ScienceQA_VAL_acc.csv @@ -0,0 +1,2 @@ +"split","Overall","Adaptations","Adaptations and natural selection","Age of Exploration","Ancient Egypt and Kush","Ancient Mesopotamia","Animals","Astronomy","Atoms and molecules","Basic economic principles","Chemical reactions","Cities","Classification","Classification and scientific names","Climate change","Colonial America","Context clues","Descriptive details","Designing experiments","Domain-specific vocabulary","Early 19th century American history","Early Americas","Earth events","Ecological interactions","Ecosystems","Engineering practices","English colonies in North America","Force and motion","Fossils","Genes to traits","Geography","Government","Independent reading comprehension","Informational texts: level 1","Magnets","Maps","Materials","Medieval Asia","Natural resources and human impacts","Oceania: geography","Oceans and continents","Oceans and continents ","Particle motion and energy","Persuasive strategies","Physical Geography","Plant reproduction","Plants","Plate tectonics","Read-alone texts","Rocks and minerals","Rome and the Byzantine Empire","Scientific names","Solutions","State capitals","States","States of matter","The American Revolution","The Americas: geography","The Antebellum period","The Civil War and Reconstruction","The Silk Road","Thermal energy","Velocity, acceleration, and forces","Visual elements","Water cycle","Weather and climate","World religions" +"val","0.7052932761087267","0.9166666666666666","1.0","1.0","0.0","1.0","1.0","0.6304347826086957","0.6190476190476191","0.59375","1.0","0.125","0.9285714285714286","1.0","1.0","0.5714285714285714","0.6","1.0","0.967741935483871","0.8","1.0","0.5","1.0","0.76","0.835820895522388","1.0","0.5128205128205128","0.72","0.7058823529411765","0.1276595744680851","0.5352112676056338","0.0","1.0","1.0","0.3533834586466165","0.6612903225806451","0.8793103448275862","1.0","0.5","0.3829787234042553","0.9","1.0","0.17647058823529413","0.5555555555555556","0.8372093023255814","0.6","1.0","1.0","1.0","1.0","1.0","0.9761904761904762","0.40298507462686567","0.9811320754716981","0.0","0.9210526315789473","0.6666666666666666","0.3","1.0","1.0","1.0","0.6666666666666666","0.0784313725490196","0.8888888888888888","1.0","0.484375","0.6666666666666666" diff --git a/llava_phi3_5/llava_phi3_5_ScienceQA_VAL_openai_result.pkl b/llava_phi3_5/llava_phi3_5_ScienceQA_VAL_openai_result.pkl new file mode 100644 index 0000000000000000000000000000000000000000..1d81cfe9e7d0d7404623c43440d2670f40ab149d --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_ScienceQA_VAL_openai_result.pkl @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:afd3d042829db41f86536d5cffff40360ecbb2af3b4265513b02a4b983576565 +size 172130 diff --git a/llava_phi3_5/llava_phi3_5_ScienceQA_VAL_openai_result.xlsx b/llava_phi3_5/llava_phi3_5_ScienceQA_VAL_openai_result.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..a8ca1ff87fe3359a74ced8123449a1f798824136 --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_ScienceQA_VAL_openai_result.xlsx @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4b5bbe2108c34c933fb4881b4a716df80a1528a341d2b210bca7fc65f3598c0a +size 387714 diff --git a/llava_phi3_5/llava_phi3_5_TextVQA_VAL.xlsx b/llava_phi3_5/llava_phi3_5_TextVQA_VAL.xlsx new file mode 100644 index 0000000000000000000000000000000000000000..0f72d9113d579e0515e6fe779bb15addc0475f1c --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_TextVQA_VAL.xlsx @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6462e849e0dcf3cafb6ca082a81906223c614f7f866e3b41d792613d0df80b33 +size 355614 diff --git a/llava_phi3_5/llava_phi3_5_TextVQA_VAL_acc.csv b/llava_phi3_5/llava_phi3_5_TextVQA_VAL_acc.csv new file mode 100644 index 0000000000000000000000000000000000000000..4a6a442415a21f787ed8e2fc831a0e7c51ac7aa2 --- /dev/null +++ b/llava_phi3_5/llava_phi3_5_TextVQA_VAL_acc.csv @@ -0,0 +1,2 @@ +"Overall" +"0.0"