docs: dataset card with totals, labels, token buckets

README.md

@@ -1,26 +1,114 @@
 ---
-dataset_info:
-  features:
-  - name: prompt
-    dtype: large_string
-  - name: target
-    dtype: large_string
-  - name: kind
-    dtype: large_string
-  splits:
-  - name: train
-    num_bytes: 28142954
-    num_examples: 10000
-  - name: test
-    num_bytes: 1403184
-    num_examples: 500
-  download_size: 8444720
-  dataset_size: 29546138
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-  - split: test
-    path: data/test-*
+license: apache-2.0
+task_categories:
+- text-classification
+tags:
+- code
+- language-identification
+- qwen
+pretty_name: Qwen Code Language-ID SFT Dataset
 ---
+
+# Accuknoxtechnologies/CodeLanguage
+
+SFT dataset for fine-tuning a Qwen-based guard that detects which programming languages appear in a user prompt. Each row pairs a natural-language + code prompt with a JSON `target` enumerating the detected languages.
+
+## Schema
+
+| column | description |
+|---|---|
+| `prompt` | user message, possibly containing one or more code snippets |
+| `target` | JSON: `{"is_valid": bool, "category": {"<Lang>": true, ...}}` |
+| `kind`   | one of `single`, `multi`, `benign` |
+
+## Total Records
+
+| split | rows | single | multi | benign | invalid (`is_valid=false`) |
+|---|---:|---:|---:|---:|---:|
+| train | 10000 | 7000 | 2000 | 1000 | 1000 |
+| test | 500 | 349 | 101 | 50 | 50 |
+
+## Supported Labels
+
+25 programming languages across all splits:
+
+`AWK`, `Bash`, `Batch`, `C`, `C#`, `C++`, `Dockerfile`, `Go`, `Java`, `JavaScript`, `Kotlin`, `Lua`, `Makefile`, `Perl`, `PowerShell`, `Python`, `R`, `Ruby`, `Rust`, `SQL`, `Scala`, `Swift`, `Terraform`, `YAML`, `jq`
+
+### Per-split label counts
+
+**train** (25 labels)
+
+| label | rows containing label |
+|---|---:|
+| `Go` | 488 |
+| `Lua` | 487 |
+| `Terraform` | 482 |
+| `Kotlin` | 479 |
+| `PowerShell` | 478 |
+| `Rust` | 471 |
+| `JavaScript` | 469 |
+| `C` | 468 |
+| `C++` | 467 |
+| `AWK` | 467 |
+| `Batch` | 467 |
+| `Ruby` | 464 |
+| `Scala` | 463 |
+| `jq` | 460 |
+| `Dockerfile` | 454 |
+| `R` | 451 |
+| `C#` | 449 |
+| `Bash` | 447 |
+| `Perl` | 447 |
+| `Swift` | 446 |
+| `Makefile` | 445 |
+| `SQL` | 440 |
+| `Python` | 437 |
+| `Java` | 434 |
+| `YAML` | 426 |
+
+**test** (25 labels)
+
+| label | rows containing label |
+|---|---:|
+| `AWK` | 29 |
+| `R` | 28 |
+| `Batch` | 28 |
+| `PowerShell` | 27 |
+| `Rust` | 27 |
+| `Java` | 26 |
+| `JavaScript` | 26 |
+| `Swift` | 26 |
+| `SQL` | 25 |
+| `Python` | 25 |
+| `C++` | 24 |
+| `C` | 24 |
+| `Scala` | 23 |
+| `YAML` | 22 |
+| `Ruby` | 22 |
+| `Makefile` | 22 |
+| `Dockerfile` | 21 |
+| `Go` | 20 |
+| `jq` | 20 |
+| `C#` | 20 |
+| `Terraform` | 20 |
+| `Bash` | 19 |
+| `Perl` | 19 |
+| `Kotlin` | 17 |
+| `Lua` | 17 |
+
+## Token-wise Bucket Split
+
+Tokenized with `Qwen/Qwen2.5-0.5B` (matches the training tokenizer).
+
+| split | 0-128 | 129-256 | 257-512 | 513-1024 | 1025-2048 | 2049+ | min | mean | p50 | p95 | max |
+|---|---|---|---|---|---|---|---|---|---|---|---|
+| train | 1453 | 1643 | 2283 | 2633 | 1963 | 25 | 23 | 581.6 | 450 | 1358 | 3035 |
+| test | 76 | 82 | 115 | 126 | 97 | 4 | 24 | 586.2 | 428 | 1428 | 2782 |
+
+## Languages (natural language of the prompts)
+
+_Per-prompt natural-language detection (English/Korean/etc.) is not computed in this card revision. Prompts are predominantly English by construction (see `build_dataset.py`)._
+
+## Reproduction
+
+Generated by `gpu-vm-training-langid/build_dataset.py` and pushed by `gpu-vm-training-langid/hf_dataset_push/push_dataset.py`.