Add per-split selected BM25 tokenization strategy table

README.md

@@ -201,11 +201,23 @@ A lightweight, evaluation-ready subset of [CodeSearchNet](https://huggingface.co
 
 ## BM25 tokenization strategy
 
-- `transformer`: `Qwen/Qwen3-0.6B` tokenizer
-- `word_seg`: language-specific word segmentation for `ja`, `zh`, `th`, `ko`
-- `stemmer`: `PyStemmer`
-- `whitespace`: simple whitespace split (`str.split()`)
-- Auto selection chooses the best strategy per split/language using retrieval metrics.
+- Auto selection chooses the best strategy per split using retrieval metrics.
+- Strategy types:
+  - `transformer`: `Qwen/Qwen3-0.6B` tokenizer
+  - `word_seg`: language-specific word segmentation (`ja`, `zh`, `th`, `ko`)
+  - `stemmer`: `PyStemmer`
+  - `whitespace`: `str.split()`
+
+Selected strategy by split:
+
+| Split | Strategy | Details |
+|---|---|---|
+| `NanoCodeSearchNetGo` | `transformer` | tokenizer: `Qwen/Qwen3-0.6B` |
+| `NanoCodeSearchNetJava` | `transformer` | tokenizer: `Qwen/Qwen3-0.6B` |
+| `NanoCodeSearchNetJavaScript` | `transformer` | tokenizer: `Qwen/Qwen3-0.6B` |
+| `NanoCodeSearchNetPHP` | `transformer` | tokenizer: `Qwen/Qwen3-0.6B` |
+| `NanoCodeSearchNetPython` | `transformer` | tokenizer: `Qwen/Qwen3-0.6B` |
+| `NanoCodeSearchNetRuby` | `transformer` | tokenizer: `Qwen/Qwen3-0.6B` |
 
 ## Upstream source