repo_path stringlengths 6 100 | repo_id int64 5.96k 1.02B | commit_id stringlengths 40 40 | github_metadata dict | num_files int64 6 8.31k | files listlengths 6 8.31k | estimated_tokens float64 16.7k 1.05M | selection_metadata dict | chunk_embeddings large listlengths 3 327 | chunk_metadata large listlengths 3 327 | embedding_model_tokens int64 17k 2.67M | embedding_reused_prefix bool 1
class | embedding list | embedding_tokens int32 16.4k 2.58M | embedding_chunks int32 3 327 | embedding_source_file stringclasses 137
values | embedding_source_row int64 0 745 | selection_row_id int64 0 92.9k | selection_cluster int32 0 1.02k | selection_max_cosine float32 -1 0.96 | selection_nearest_row_id int64 -1 135k |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Bahn-X/swift-composable-navigator | 308,578,832 | 69dbc2ab36dffa1e48d1b7f87a03db4767162538 | {"branch":"main","commit_count":118,"repo_created_at":"2020-10-30T09:12:04.000Z","is_fork":false,"is(...TRUNCATED) | 117 | [{"content_id":"43cfd29199c99e7370a37764e94328b0cb3fdae6","content":"import ComposableNavigator\nimp(...TRUNCATED) | 65,559.09726 | {"input_num_files":120,"input_estimated_tokens":65906.096626576,"source_file_count":105,"source_toke(...TRUNCATED) | [[0.02291288785636425,0.01695799082517624,0.07878769934177399,-0.04336201027035713,-0.00338349305093(...TRUNCATED) | [{"token_start":0,"token_end":8136,"new_token_start":0,"new_tokens":8136,"overlap_tokens":0,"input_t(...TRUNCATED) | 82,125 | false | [-0.00024687027325853705,0.035528890788555145,0.06846614181995392,-0.038115132600069046,-0.015031894(...TRUNCATED) | 79,389 | 11 | data/train-00000-of-00200.parquet | 0 | 0 | 5 | 0.88419 | 11,087 |
HewlettPackard/hpe-oneview-hubot | 73,227,722 | a5a714d1f9c67eb623945b86b636480e9317868a | {"branch":"master","commit_count":204,"repo_created_at":"2016-11-08T21:16:51.000Z","is_fork":false,"(...TRUNCATED) | 93 | [{"content_id":"650479b4cc1ab5ff5fd810de267e6a455e515f21","content":"/*\n(c) Copyright 2016-2017 Hew(...TRUNCATED) | 125,336.037694 | {"input_num_files":94,"input_estimated_tokens":125369.87511417833,"source_file_count":89,"source_tok(...TRUNCATED) | [[-0.045008640736341476,0.0642203614115715,0.01500230748206377,-0.02813253551721573,-0.0296754539012(...TRUNCATED) | [{"token_start":0,"token_end":8146,"new_token_start":0,"new_tokens":8146,"overlap_tokens":0,"input_t(...TRUNCATED) | 149,425 | false | [-0.03731190413236618,0.052258655428886414,0.020600896328687668,-0.027187423780560493,-0.01254727132(...TRUNCATED) | 144,475 | 19 | data/train-00000-of-00200.parquet | 3 | 3 | 822 | 0.884345 | 126,808 |
belaalb/G2DM | 245,548,355 | e48523722bc77304c204bba78893fb58605e3453 | {"branch":"master","commit_count":47,"repo_created_at":"2020-03-07T01:32:23.000Z","is_fork":false,"i(...TRUNCATED) | 32 | [{"content_id":"d8f6a886c860ff075a5675c1158cee02dc5032e6","content":"import numpy as np\nimport scip(...TRUNCATED) | 41,495.343668 | {"input_num_files":33,"input_estimated_tokens":41637.03786542266,"source_file_count":28,"source_toke(...TRUNCATED) | [[-0.04875975847244263,-0.005389716010540724,0.03163716197013855,0.032286133617162704,-0.02680053375(...TRUNCATED) | [{"token_start":0,"token_end":8110,"new_token_start":0,"new_tokens":8110,"overlap_tokens":0,"input_t(...TRUNCATED) | 58,222 | false | [-0.03650451824069023,-0.010780413635075092,0.029376106336712837,0.031430140137672424,-0.04002301767(...TRUNCATED) | 56,318 | 8 | data/train-00000-of-00200.parquet | 46 | 46 | 792 | 0.917476 | 134,339 |
d-Rickyy-b/pastepwn | 147,591,439 | 6832d70cb1eb9e141e49a776b69cf2844906025b | {"branch":"master","commit_count":810,"repo_created_at":"2018-09-05T23:35:41.000Z","is_fork":false,"(...TRUNCATED) | 162 | [{"content_id":"bee36df831f8aaacb6e047f5665f398cc775be0c","content":"import logging\nimport sys\nfro(...TRUNCATED) | 80,834.786584 | {"input_num_files":165,"input_estimated_tokens":81163.39340892251,"source_file_count":155,"source_to(...TRUNCATED) | [[-0.04318351671099663,0.02517392300069332,0.05946945399045944,-0.045882195234298706,-0.026454385370(...TRUNCATED) | [{"token_start":0,"token_end":8167,"new_token_start":0,"new_tokens":8167,"overlap_tokens":0,"input_t(...TRUNCATED) | 100,115 | false | [-0.017080681398510933,0.026442788541316986,0.06343312561511993,-0.04922420158982277,-0.009905349463(...TRUNCATED) | 96,822 | 13 | data/train-00000-of-00200.parquet | 51 | 51 | 341 | 0.878721 | 102,414 |
konsultaner/jsonOdm | 31,983,550 | 6853a6393ce041280d3c3fa982859bc728877c8f | {"branch":"master","commit_count":131,"repo_created_at":"2015-03-10T21:46:54.000Z","is_fork":false,"(...TRUNCATED) | 16 | [{"content_id":"2e758e08f47090894ed1cb01679bbd92fc25d19f","content":"\"use strict\";\n\n// for code (...TRUNCATED) | 128,470.99762 | {"input_num_files":17,"input_estimated_tokens":128494.26084637817,"source_file_count":12,"source_tok(...TRUNCATED) | [[0.006390456110239029,0.019929466769099236,0.03792339563369751,-0.039919789880514145,-0.01433195918(...TRUNCATED) | [{"token_start":0,"token_end":7431,"new_token_start":0,"new_tokens":7431,"overlap_tokens":0,"input_t(...TRUNCATED) | 154,477 | false | [-0.00430414080619812,0.022296026349067688,0.029044901952147484,-0.044725190848112106,-0.01936873979(...TRUNCATED) | 149,063 | 21 | data/train-00000-of-00200.parquet | 64 | 64 | 1,004 | 0.892265 | 1,088 |
run-as-root/magento2-message-queue-retry | 606,031,797 | 4d60d493577784b23f3f68a29dcba300e1973781 | {"branch":"develop","commit_count":76,"repo_created_at":"2023-02-24T12:53:24.000Z","is_fork":false,"(...TRUNCATED) | 88 | [{"content_id":"fc19713c2aea0cc1156342bbb75f5e42542e7ee1","content":"<?php\n\ndeclare(strict_types=1(...TRUNCATED) | 43,054.886529 | {"input_num_files":90,"input_estimated_tokens":43162.965865476595,"source_file_count":68,"source_tok(...TRUNCATED) | [[-0.06595748662948608,0.04341103881597519,0.02331424318253994,-0.031358398497104645,-0.030132133513(...TRUNCATED) | [{"token_start":0,"token_end":7556,"new_token_start":0,"new_tokens":7556,"overlap_tokens":0,"input_t(...TRUNCATED) | 41,190 | false | [-0.05240950360894203,0.04788064956665039,0.03834393247961998,-0.030305489897727966,-0.0272387098520(...TRUNCATED) | 39,778 | 6 | data/train-00000-of-00200.parquet | 74 | 74 | 926 | 0.893579 | 103,368 |
guluc3m/gul-zoe | 8,519,502 | bbcf99bc182b25bc309c3d662b70c172ab0b8908 | {"branch":"master","commit_count":269,"repo_created_at":"2013-03-02T11:16:14.000Z","is_fork":false,"(...TRUNCATED) | 96 | [{"content_id":"da0e0b2e9688c3e8201399ae4f3ad84dbede42c9","content":"JAR=zoe.jar\n\nall: test jar\n\(...TRUNCATED) | 72,112.700326 | {"input_num_files":97,"input_estimated_tokens":72179.84645709133,"source_file_count":91,"source_toke(...TRUNCATED) | [[-0.020255006849765778,-0.005541468504816294,0.011883432976901531,-0.05996834486722946,0.0160293020(...TRUNCATED) | [{"token_start":0,"token_end":8029,"new_token_start":0,"new_tokens":8029,"overlap_tokens":0,"input_t(...TRUNCATED) | 78,665 | false | [-0.02500687912106514,0.008052959106862545,0.03281586617231369,-0.04578282684087753,0.00216408399865(...TRUNCATED) | 76,201 | 10 | data/train-00000-of-00200.parquet | 99 | 99 | 164 | 0.874835 | 129,528 |
leggedrobotics/elevation_mapping_cupy | 473,697,880 | 20a8a26b67a995b43eb44c23568854d1fed82a52 | {"branch":"main","commit_count":955,"repo_created_at":"2022-03-24T17:00:25.000Z","is_fork":false,"is(...TRUNCATED) | 188 | [{"content_id":"68565bb91e07c17cbf188684423e1da2215b14ea","content":"//\n// Created by rgrandia on 0(...TRUNCATED) | 204,738.224879 | {"input_num_files":189,"input_estimated_tokens":204774.44149324775,"source_file_count":136,"source_t(...TRUNCATED) | [[-0.008552966639399529,-0.01746056042611599,0.03217947855591774,-0.0271991565823555,-0.045946862548(...TRUNCATED) | [{"token_start":0,"token_end":8046,"new_token_start":0,"new_tokens":8046,"overlap_tokens":0,"input_t(...TRUNCATED) | 201,655 | false | [-0.028865711763501167,-0.029522785916924477,0.040684863924980164,-0.026122475042939186,-0.051425732(...TRUNCATED) | 195,061 | 25 | data/train-00000-of-00200.parquet | 108 | 108 | 654 | 0.919026 | 113,614 |
nglauber/tdcapp | 140,335,726 | 5d7672c5cc79b21559d02a9452c84660cf348e29 | {"branch":"master","commit_count":23,"repo_created_at":"2018-07-09T19:58:48.000Z","is_fork":false,"i(...TRUNCATED) | 124 | [{"content_id":"d4afdb8a46f08b92f3f68a22a00cf7c0dc19d326","content":"package br.com.nglauber.tdcapp.(...TRUNCATED) | 36,727.558593 | {"input_num_files":124,"input_estimated_tokens":36727.55859320245,"source_file_count":106,"source_to(...TRUNCATED) | [[-0.06605115532875061,0.017438098788261414,0.008409930393099785,-0.017062591388821602,0.00271336268(...TRUNCATED) | [{"token_start":0,"token_end":8124,"new_token_start":0,"new_tokens":8124,"overlap_tokens":0,"input_t(...TRUNCATED) | 48,280 | false | [-0.07357839494943619,0.016708236187696457,0.014408113434910774,-0.018073027953505516,0.021045511588(...TRUNCATED) | 46,916 | 6 | data/train-00000-of-00200.parquet | 110 | 110 | 425 | 0.887454 | 50,047 |
willwerscheid/flashier | 157,906,610 | 45363f64c9ff01f0b63c49d5f3854fedef6ac1aa | {"branch":"master","commit_count":622,"repo_created_at":"2018-11-16T18:31:51.000Z","is_fork":false,"(...TRUNCATED) | 56 | [{"content_id":"f6f061d74346527e484419480be2854c78548c13","content":"#' Nullcheck flash factors\n#'\(...TRUNCATED) | 84,303.985426 | {"input_num_files":57,"input_estimated_tokens":84325.92687836557,"source_file_count":50,"source_toke(...TRUNCATED) | [[-0.019793344661593437,0.0001758851285558194,0.09015226364135742,0.002110024681314826,-0.0028730127(...TRUNCATED) | [{"token_start":0,"token_end":8123,"new_token_start":0,"new_tokens":8123,"overlap_tokens":0,"input_t(...TRUNCATED) | 100,791 | false | [-0.022988902404904366,0.005499590653926134,0.07981710135936737,-0.017293652519583702,0.028537994250(...TRUNCATED) | 97,537 | 13 | data/train-00000-of-00200.parquet | 118 | 118 | 854 | 0.906935 | 65,883 |
End of preview. Expand in Data Studio
Stack v3.1 Part2: semantic diversity and length selection
Status: complete. Selected 12,813 repositories, containing 1,923,925,329 stored canonical content tokens.
Source: storagestuffnobody/stack-v3.1-subsampled-part2-embeddings. All original fields, including text, nested metadata, embeddings and provenance, are preserved for retained rows.
Normalized existing 768-dimensional aggregate embeddings were grouped into 1,024 spherical k-means buckets. Within each bucket, overlap is the maximum FP32 cosine similarity to any other member, excluding self, evaluated exhaustively. Keep the 30% with lowest overlap, then the longest 30% of those survivors by embedding_tokens, separately within every bucket. Both quotas round upward; source row IDs and overlap scores break ties deterministically.
Selection is global across the source's default splits; every retained row keeps its original split when split preservation is enabled. There is no additional per-split quota. A source split may have zero retained rows.
| Source split | Input rows | Selected rows | Selected content tokens |
|---|---|---|---|
| train | 135,049 | 12,813 | 1,923,925,329 |
This is fixed-rate semantic redundancy selection. It does not guarantee elimination of every duplicate or measure literal passage overlap. Similarity is assessed within buckets. No additional topic, language, NSFW or content-quality filters were introduced.
The selection_* columns identify the source row, bucket, closest-neighbour score and neighbour source row. Full decisions, bucket counts, code, pinned inputs and verification are under selection/ and job-inputs/. Source licensing and attribution remain applicable.
Verified token budget and scope
Part2 was selected independently; this is not joint clustering or cross-part deduplication with Part1.
| Part2 stage | Repositories | Stored canonical content tokens |
|---|---|---|
| Input | 135,049 | 19,660,497,066 |
| Least-overlap 30% per bucket | 41,013 | 2,916,182,935 |
| Longest 30% of those survivors | 12,813 | 1,923,925,329 |
The existing Part1 selection contains 3,027,545,995 tokens. Adding this Part2 selection gives 4,951,471,324 tokens, below the approximately 10B target. These are stored canonical counts from the embedding tokenizer, not the older size-based token estimates. The two parts remain separate datasets.
Full clustering reached its 30-iteration limit without convergence. The similarity score can favor shorter repositories and does not establish literal code duplication or downstream quality. No extra length bands or filters were added. The independent token-budget audit checked every original Part2 length-column hash and reconstructed both masks exactly. Default data paths exclude pilot artifacts.
- Downloads last month
- 238