{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/census/cos-similarity","entry":"cos_similarity","source":"Syntology differential census (groundwork/55, run v2_2026-09-22), per sample; not an archive number","census_date":"2026-09-22","battery_sha256":["b986f7e04d794a0d88ad4c5f32cf63ec3590b5deff0192150737bc5f1c0b4677"],"runner_sha256":["5a452d0e7c0da5b80771d1be2afe3572e253568cf5c5d59a0d08ebd663d00808"],"bucket_key":"positional (rank, kind, dtype) of each array argument; the argument name is not part of the key because the harness draws the shared array from (rank, kind) and casts it to the dtype, whatever the name","bucket_fields":["rank","kind","dtype"],"claim":"Implementations sharing this entry name were each run on one shared input fixed by the positional (rank, kind, dtype) of their array arguments (the bucket). A cluster is the set whose recorded output digest (sha256 of the output rounded to 6 decimals) is identical. Identical values to six decimals on the shared input are agreement on those inputs, not a statement about the whole domain and not a substitution claim.","n_implementations_compared":5,"n_papers":7,"n_buckets":3,"n_distinct_outputs":5,"n_class_bearing":0,"not_compared":{"not_run_on_shared_input":{"n":0,"by_error":{}},"no_array_argument_ran_on_own_fixture_arguments_only":{"n":1,"n_papers":1,"recorded_shared_digest_equals_own_fixture_digest":1},"output_not_digested_non_numeric":{"n":0,"by_type":{}}},"withdrawn_excluded":0,"code_page":"/code/cos-similarity","buckets":[{"bucket":[[2,"float","float32"],[2,"float","float32"]],"n_implementations_compared":3,"n_papers":5,"n_not_digested":0,"not_digested_by_type":{},"clusters":[{"output_sha":"6f54f8d378485875","size":1,"n_papers":3,"shape":[8,8],"dtype":"float32","type":"Tensor","finite":true,"max_difference_among_recorded_values":null,"members_with_recorded_values":1,"torch_reference_conventions_with_this_digest":[],"values":[1.0,-0.5404032468795776,-0.028821606189012527,-0.6983117461204529,-0.2543237805366516,-0.2780967950820923,-0.1319955587387085,0.651767373085022,-0.5404032468795776,1.0000001192092896,-0.46969205141067505,0.13053511083126068,0.1603958159685135,0.8260827660560608,0.2009304016828537,-0.6391847133636475,-0.028821606189012527,-0.46969205141067505,0.9999999403953552,0.4388355314731598,-0.7910568714141846,-0.1930801123380661,-0.7949833869934082,-0.048688579350709915,-0.6983117461204529,0.13053511083126068,0.4388355314731598,1.0,-0.05132824927568436,-0.1843065321445465,0.040962204337120056,-0.7964414358139038,-0.2543237805366516,0.1603958159685135,-0.7910568714141846,-0.05132824927568436,1.0,-0.2673105001449585,0.9276159405708313,0.013831527903676033,-0.2780967950820923,0.8260827660560608,-0.1930801123380661,-0.1843065321445465,-0.2673105001449585,1.0,-0.3068235516548157,-0.28617602586746216,-0.1319955587387085,0.2009304016828537,-0.7949833869934082,0.040962204337120056,0.9276159405708313,-0.3068235516548157,1.0,-0.1753852218389511,0.651767373085022,-0.6391847133636475,-0.048688579350709915,-0.7964414358139038,0.013831527903676033,-0.28617602586746216,-0.1753852218389511,0.9999999403953552],"values_recorded":64,"members":[{"code_sha256_prefix":"cc642412d4347cce","path":"similarity.py","papers":["2204.10965","2408.01432","2604.14519"],"paper_pages":[{"arxiv_id":"2204.10965","page":"/paper/clip-dissect-automatic-description-of-neuron"},{"arxiv_id":"2408.01432","page":"/paper/vlg-cbm-training-concept-bottleneck-models"},{"arxiv_id":"2604.14519","page":"/paper/arxiv-2604-14519"}],"arg_sig_recorded":[["clip_feats",2,"float","float32"],["target_feats",2,"float","float32"]],"scalar_args":{"device":"'cpu'"},"class_bearing":false}]},{"output_sha":"97c4d5a90e4c64b5","size":1,"n_papers":1,"shape":[4,4],"dtype":"float32","type":"Tensor","finite":true,"max_difference_among_recorded_values":null,"members_with_recorded_values":1,"torch_reference_conventions_with_this_digest":[],"values":[2.9891891479492188,-0.352142333984375,2.2116928100585938,-0.7149543762207031,-0.3734607696533203,3.140777587890625,-1.258409023284912,0.6062393188476562,1.7532298564910889,-0.9438650608062744,2.3765981197357178,-0.7479329109191895,-0.6722259521484375,0.5377655029296875,-0.8910980224609375,2.821239471435547],"values_recorded":16,"members":[{"code_sha256_prefix":"c06cc6143cfaf205","path":"tau_norm.py","papers":["1910.09217"],"paper_pages":[{"arxiv_id":"1910.09217","page":"/paper/decoupling-representation-and-classifier-for"}],"arg_sig_recorded":[["A",2,"float","float32"],["B",2,"float","float32"]],"scalar_args":{},"class_bearing":false}]},{"output_sha":"adc71adfbba8ca30","size":1,"n_papers":1,"shape":[4,4],"dtype":"float32","type":"Tensor","finite":true,"max_difference_among_recorded_values":null,"members_with_recorded_values":1,"torch_reference_conventions_with_this_digest":[],"values":[0.9977779388427734,-0.11758613586425781,0.7382721900939941,-0.2374095916748047,-0.11758613586425781,0.9974908828735352,-0.398309588432312,0.19110107421875,0.7382721900939941,-0.398309588432312,1.0007596015930176,-0.31519222259521484,-0.2374095916748047,0.19110107421875,-0.31519222259521484,0.9986743927001953],"values_recorded":16,"members":[{"code_sha256_prefix":"aa98a2606d29ec50","path":"fsl_head/pcia.py","papers":["2304.00022"],"paper_pages":[{"arxiv_id":"2304.00022","page":"/paper/what-makes-for-effective-few-shot-point-cloud"}],"arg_sig_recorded":[["x",2,"float","float32"],["y",2,"float","float32"]],"scalar_args":{},"class_bearing":false}]}]},{"bucket":[[1,"float","float64"],[1,"float","float64"]],"n_implementations_compared":1,"n_papers":1,"n_not_digested":0,"not_digested_by_type":{},"clusters":[{"output_sha":"6c3c396ed6b5c36d","size":1,"n_papers":1,"shape":[],"dtype":"float64","type":"float64","finite":true,"max_difference_among_recorded_values":null,"members_with_recorded_values":1,"torch_reference_conventions_with_this_digest":[],"values":[0.9999999999999998],"values_recorded":1,"members":[{"code_sha256_prefix":"8542e4e462554ea2","path":"get_cosine_similarities.py","papers":["2504.12939"],"paper_pages":[{"arxiv_id":"2504.12939","page":"/paper/disentangling-polysemantic-channels-in"}],"arg_sig_recorded":[["a",1,"float","float64"],["b",1,"float","float64"]],"scalar_args":{},"class_bearing":false}]}]},{"bucket":[[3,"float","float32"],[3,"float","float32"]],"n_implementations_compared":1,"n_papers":1,"n_not_digested":0,"not_digested_by_type":{},"clusters":[{"output_sha":"5f07eef034c5a21f","size":1,"n_papers":1,"shape":[2],"dtype":"float32","type":"Tensor","finite":true,"max_difference_among_recorded_values":null,"members_with_recorded_values":1,"torch_reference_conventions_with_this_digest":[],"values":[0.9999999403953552,0.9999998807907104],"values_recorded":2,"members":[{"code_sha256_prefix":"60f2dfd4e656f6b5","path":"interpretability/robustness.py","papers":["2304.06715"],"paper_pages":[{"arxiv_id":"2304.06715","page":"/paper/evaluating-the-robustness-of-interpretability"}],"arg_sig_recorded":[["x1",3,"float","float32"],["x2",3,"float","float32"]],"scalar_args":{"reduce":"False"},"class_bearing":false}]}]}]}