README.md

metadata

license: mit
tags:
  - code
  - assertion-completion
  - repository-level

code2lora-data-commits — per-commit dataset (GRU)

One parquet row per (repository, kept commit), carrying:

• commit_sha, commit_index, commit_timestamp
• production_code_diff -- diff vs the previous kept commit
• diff_embedding -- Qwen3-Embedding of the diff (fp16, [2048])
• repo_state_embedding -- Qwen3-Embedding of the full repo state at this commit (fp16, [2048])
• cross_repo_split, in_repo_split -- the two split labels (see below)

Layout

commits/
  cr_train.parquet    commits from the 400 train repos
  cr_val.parquet      commits from held-out cr-val repos
  cr_test.parquet     commits from held-out cr-test repos
  ood_test.parquet    commits from the post-cutoff OOD repos

Split semantics

Two independent split axes:

• cross_repo_split ∈ {train, cr_val, cr_test} -- which repositories the model has seen.
• in_repo_split ∈ {train, val, test} -- chronological 80/10/10 split of each repo's commits.

The four evaluation suites used in the paper map onto these axes as follows:

Paper suite	File	Selection
Train (loss)	qna/cr_train.parquet	in_repo_split == "train"
IR val	qna/cr_train.parquet	in_repo_split == "val"
IR test	qna/cr_train.parquet	in_repo_split == "test"
CR val	qna/cr_val.parquet	all rows
CR test	qna/cr_test.parquet	all rows

All evaluation suites use the full uncapped assertion set. Only the in-repo train loss path is smart-capped (see code2lora/code2lora-data-smartcap).

Companion datasets

• code2lora/code2lora-data-smartcap smart-capped per-commit QnAs for training
• code2lora/code2lora-data-snapshots static-snapshot variant
• code2lora/code2lora-data-ood OOD repos in the same schema