Ubuntu
commited on
Commit
Β·
97df98a
1
Parent(s):
1dc83ef
Update README.md
Browse files
README.md
CHANGED
@@ -31,41 +31,37 @@ This directory contains data for fine-tuning models and building RAGs for the do
|
|
31 |
## Overview
|
32 |
|
33 |
- **data/**: Contains all datasets and indexes.
|
34 |
-
- **
|
35 |
-
- **raw_data.tar**:
|
36 |
-
- **samples/**:
|
37 |
-
- **processed/**: Parquet
|
38 |
- **faiss/**: Vector indexes for RAG workflows.
|
39 |
-
- **scripts/**: Python and
|
40 |
|
41 |
## Data Structure
|
42 |
|
43 |
```
|
44 |
data/
|
45 |
-
βββ
|
46 |
-
βββ
|
47 |
-
β βββ prs/
|
48 |
-
β βββ diffs/
|
49 |
-
β βββ
|
50 |
βββ processed/
|
51 |
β βββ train.parquet
|
52 |
β βββ test.parquet
|
53 |
βββ faiss/
|
54 |
-
βββ index
|
|
|
55 |
```
|
56 |
|
57 |
-
- **raw_data/**: Contains the full, unprocessed PR and diff data as collected from GitHub.
|
58 |
-
- **samples/**: Contains JSONL files formatted for use with HuggingFace Datasets.
|
59 |
-
- **processed/**: Contains files for model fine-tuning.
|
60 |
-
- **faiss/**: Contains vector indexes for fast retrieval in RAG setups.
|
61 |
-
|
62 |
## Generated dataset
|
63 |
|
64 |
PR is considered as a timeline with events. Input is PR metadata (title, description, label) and commit n-1, with all events between n-1 and n. Completion is n. It is possible to filter by time, label, authors, etc.
|
65 |
|
66 |
## Scripts
|
67 |
|
68 |
-
See [scripts/README.md](scripts/README.md) for details on running the crawler, dataset generation,
|
69 |
|
70 |
## PyTorch Dataset Example
|
71 |
|
|
|
31 |
## Overview
|
32 |
|
33 |
- **data/**: Contains all datasets and indexes.
|
34 |
+
- **raw/sample/**: Sample PRs and diffs collected from GitHub.
|
35 |
+
- **raw_data.tar**: Archive of collected PRs and diffs from GitHub.
|
36 |
+
- **samples/**: Json files with processed samples suitable for dataset generation.
|
37 |
+
- **processed/**: Parquet files for fine-tuning (e.g., `train.parquet`, `test.parquet`).
|
38 |
- **faiss/**: Vector indexes for RAG workflows.
|
39 |
+
- **scripts/**: Python and nodejs scripts for crawling, processing, and indexing.
|
40 |
|
41 |
## Data Structure
|
42 |
|
43 |
```
|
44 |
data/
|
45 |
+
βββ raw/
|
46 |
+
| βββ sample/
|
47 |
+
β β βββ prs/
|
48 |
+
β β βββ diffs/
|
49 |
+
β βββ raw_data.tar
|
50 |
βββ processed/
|
51 |
β βββ train.parquet
|
52 |
β βββ test.parquet
|
53 |
βββ faiss/
|
54 |
+
βββ index.faiss
|
55 |
+
βββ index.pkl
|
56 |
```
|
57 |
|
|
|
|
|
|
|
|
|
|
|
58 |
## Generated dataset
|
59 |
|
60 |
PR is considered as a timeline with events. Input is PR metadata (title, description, label) and commit n-1, with all events between n-1 and n. Completion is n. It is possible to filter by time, label, authors, etc.
|
61 |
|
62 |
## Scripts
|
63 |
|
64 |
+
See [scripts/README.md](scripts/README.md) for details on running the crawler, dataset generation, and RAG indexing.
|
65 |
|
66 |
## PyTorch Dataset Example
|
67 |
|