README.md

---
dataset_info:
- config_name: bm25
  features:
  - name: query-id
    dtype: string
  - name: corpus-ids
    list: string
  splits:
  - name: go
    num_bytes: 578229
    num_examples: 50
  - name: java
    num_bytes: 865210
    num_examples: 50
  - name: javascript
    num_bytes: 631841
    num_examples: 50
  - name: php
    num_bytes: 683511
    num_examples: 50
  - name: python
    num_bytes: 673168
    num_examples: 50
  - name: ruby
    num_bytes: 635258
    num_examples: 50
  download_size: 669346
  dataset_size: 4067217
- config_name: corpus
  features:
  - name: _id
    dtype: string
  - name: text
    dtype: string
  splits:
  - name: go
    num_bytes: 5866161
    num_examples: 10000
  - name: java
    num_bytes: 8383266
    num_examples: 10000
  - name: javascript
    num_bytes: 6817497
    num_examples: 6483
  - name: php
    num_bytes: 8308232
    num_examples: 10000
  - name: python
    num_bytes: 12057318
    num_examples: 10000
  - name: ruby
    num_bytes: 1456896
    num_examples: 2279
  download_size: 17919374
  dataset_size: 42889370
- config_name: qrels
  features:
  - name: query-id
    dtype: string
  - name: corpus-id
    dtype: string
  splits:
  - name: go
    num_bytes: 11666
    num_examples: 50
  - name: java
    num_bytes: 17154
    num_examples: 50
  - name: javascript
    num_bytes: 12376
    num_examples: 50
  - name: php
    num_bytes: 13456
    num_examples: 50
  - name: python
    num_bytes: 13454
    num_examples: 50
  - name: ruby
    num_bytes: 12948
    num_examples: 50
  download_size: 61488
  dataset_size: 81054
- config_name: queries
  features:
  - name: _id
    dtype: string
  - name: text
    dtype: string
  splits:
  - name: go
    num_bytes: 11803
    num_examples: 50
  - name: java
    num_bytes: 19923
    num_examples: 50
  - name: javascript
    num_bytes: 15444
    num_examples: 50
  - name: php
    num_bytes: 19306
    num_examples: 50
  - name: python
    num_bytes: 22227
    num_examples: 50
  - name: ruby
    num_bytes: 30739
    num_examples: 50
  download_size: 82419
  dataset_size: 119442
configs:
- config_name: bm25
  data_files:
  - split: go
    path: bm25/go-*
  - split: java
    path: bm25/java-*
  - split: javascript
    path: bm25/javascript-*
  - split: php
    path: bm25/php-*
  - split: python
    path: bm25/python-*
  - split: ruby
    path: bm25/ruby-*
- config_name: corpus
  data_files:
  - split: go
    path: corpus/go-*
  - split: java
    path: corpus/java-*
  - split: javascript
    path: corpus/javascript-*
  - split: php
    path: corpus/php-*
  - split: python
    path: corpus/python-*
  - split: ruby
    path: corpus/ruby-*
- config_name: qrels
  data_files:
  - split: go
    path: qrels/go-*
  - split: java
    path: qrels/java-*
  - split: javascript
    path: qrels/javascript-*
  - split: php
    path: qrels/php-*
  - split: python
    path: qrels/python-*
  - split: ruby
    path: qrels/ruby-*
- config_name: queries
  data_files:
  - split: go
    path: queries/go-*
  - split: java
    path: queries/java-*
  - split: javascript
    path: queries/javascript-*
  - split: php
    path: queries/php-*
  - split: python
    path: queries/python-*
  - split: ruby
    path: queries/ruby-*
---

# NanoCodeSearchNet (with bm25 subset)

A lightweight, evaluation-ready subset of [CodeSearchNet](https://huggingface.co/datasets/code-search-net/code_search_net), designed for fast code-retrieval benchmarking.

## What this dataset is

- A code-retrieval benchmark with 6 programming-language splits: `go`, `java`, `javascript`, `php`, `python`, `ruby`.
- Each split contains 50 queries and up to 10,000 code snippets.
- Queries are function docstrings and positives are corresponding function bodies.

## Data structure

- `corpus`: `_id`, `text`
- `queries`: `_id`, `text`
- `qrels`: `query-id`, `corpus-id`, `score`
- `bm25`: `query-id`, `corpus-ids` (A bm25 subset is included for first-stage retrieval and reranking experiments.)

## Performance Comparison Across Models


| Model | avg | Go | Java | JavaScript | PHP | Python | Ruby |
|---|---:|---:|---:|---:|---:|---:|---:|
| bm25 | 0.4295 | 0.3644 | 0.3952 | 0.3308 | 0.2311 | 0.9552 | 0.3005 |
| e5-small | 0.7268 | 0.6805 | 0.7884 | 0.6386 | 0.6603 | 0.9142 | 0.6788 |
| e5-large | 0.7836 | 0.7595 | 0.8287 | 0.6942 | 0.7291 | 0.9513 | 0.7391 |
| bge-m3 | 0.7107 | 0.6643 | 0.6934 | 0.6138 | 0.6229 | 0.9852 | 0.6844 |

## Subset names

- Split names: `go`, `java`, `javascript`, `php`, `python`, `ruby`
- Config names: `corpus`, `queries`, `qrels`, `bm25`

## BM25 tokenization strategy

- Auto selection chooses the best strategy per split using retrieval metrics.
- Strategy types:
  - `transformer`: `Qwen/Qwen3-0.6B` tokenizer
  - `wordseg`: language-specific word segmentation (`ja`, `zh`, `th`, `ko`)
  - `stemmer`: `PyStemmer`
  - `whitespace`: `str.split()`

Selected strategy by split:

| Split | Strategy | Details |
|---|---|---|
| `go` | `transformer` | tokenizer: `Qwen/Qwen3-0.6B` |
| `java` | `transformer` | tokenizer: `Qwen/Qwen3-0.6B` |
| `javascript` | `transformer` | tokenizer: `Qwen/Qwen3-0.6B` |
| `php` | `transformer` | tokenizer: `Qwen/Qwen3-0.6B` |
| `python` | `transformer` | tokenizer: `Qwen/Qwen3-0.6B` |
| `ruby` | `transformer` | tokenizer: `Qwen/Qwen3-0.6B` |

## Upstream source

- Original dataset: [code-search-net/code_search_net](https://huggingface.co/datasets/code-search-net/code_search_net)
- Paper: [CodeSearchNet Challenge (1909.09436)](https://huggingface.co/papers/1909.09436)

## License

Other. This dataset is derived from CodeSearchNet and follows upstream repository licensing and attribution requirements.