data: add condition-3-zh-5k and condition-4-zh-5k short-name configs

#15
Files changed (1) hide show
  1. README.md +69 -1
README.md CHANGED
@@ -98,6 +98,18 @@ configs:
98
  path: data/condition-2-zh-5k/train-*
99
  - split: validation
100
  path: data/condition-2-zh-5k/validation-*
 
 
 
 
 
 
 
 
 
 
 
 
101
  - config_name: condition-5-es-5k-c4ai-aya-expanse-32b
102
  data_files:
103
  - split: train
@@ -596,6 +608,62 @@ dataset_info:
596
  num_examples: 500
597
  download_size: 24686270
598
  dataset_size: 65409803
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
599
  - config_name: condition-5-es-5k-c4ai-aya-expanse-32b
600
  features:
601
  - name: code
@@ -917,7 +985,7 @@ dataset_info:
917
 
918
  > **Experiment and proposed paper title:** _Language Decoded: Exploring the Impact of Native Code on Multilingual Models_
919
 
920
- > **Note (2026-05-18):** Current Phase 3 configs use the short `condition-*` namespace and include `103k`, `20k`, and `5k` sizes for Conditions 1--2. Phase 2 configs remain available under the `phase-2-the-stack-v1-*` namespace for reproducibility.
921
 
922
  Multilingual Python code datasets for the **Language Decoded** project (part of [Cohere's Tiny Aya Expedition](https://aya.for.ai)). See [legesher/language-decoded-experiments](https://huggingface.co/datasets/legesher/language-decoded-experiments) for the canonical project description, the full experimental ladder, and the paper-grade evaluation results.
923
 
 
98
  path: data/condition-2-zh-5k/train-*
99
  - split: validation
100
  path: data/condition-2-zh-5k/validation-*
101
+ - config_name: condition-3-zh-5k
102
+ data_files:
103
+ - split: train
104
+ path: data/phase-2-the-stack-v1/condition-3-zh-5k/train-*
105
+ - split: validation
106
+ path: data/phase-2-the-stack-v1/condition-3-zh-5k/validation-*
107
+ - config_name: condition-4-zh-5k
108
+ data_files:
109
+ - split: train
110
+ path: data/phase-2-the-stack-v1/condition-4-zh-5k/train-*
111
+ - split: validation
112
+ path: data/phase-2-the-stack-v1/condition-4-zh-5k/validation-*
113
  - config_name: condition-5-es-5k-c4ai-aya-expanse-32b
114
  data_files:
115
  - split: train
 
608
  num_examples: 500
609
  download_size: 24686270
610
  dataset_size: 65409803
611
+ - config_name: condition-3-zh-5k
612
+ features:
613
+ - name: file_path
614
+ dtype: large_string
615
+ - name: code
616
+ dtype: large_string
617
+ - name: code_en
618
+ dtype: string
619
+ - name: language
620
+ dtype: large_string
621
+ - name: license
622
+ dtype: large_string
623
+ - name: token_count
624
+ dtype: int64
625
+ - name: source_type
626
+ dtype: large_string
627
+ splits:
628
+ - name: train
629
+ num_bytes: 40782466
630
+ num_examples: 4500
631
+ - name: validation
632
+ num_bytes: 4531385
633
+ num_examples: 500
634
+ download_size: 17299185
635
+ dataset_size: 45313851
636
+ - config_name: condition-4-zh-5k
637
+ features:
638
+ - name: filename
639
+ dtype: string
640
+ - name: content
641
+ dtype: string
642
+ - name: extension
643
+ dtype: string
644
+ - name: source
645
+ dtype: string
646
+ - name: quality_tier
647
+ dtype: string
648
+ - name: sha256
649
+ dtype: string
650
+ - name: byte_size
651
+ dtype: int64
652
+ - name: total_lines
653
+ dtype: int64
654
+ - name: cjk_ratio
655
+ dtype: float64
656
+ - name: has_cjk
657
+ dtype: bool
658
+ splits:
659
+ - name: train
660
+ num_bytes: 44246508
661
+ num_examples: 6553
662
+ - name: validation
663
+ num_bytes: 7522476
664
+ num_examples: 729
665
+ download_size: 18300000
666
+ dataset_size: 51768984
667
  - config_name: condition-5-es-5k-c4ai-aya-expanse-32b
668
  features:
669
  - name: code
 
985
 
986
  > **Experiment and proposed paper title:** _Language Decoded: Exploring the Impact of Native Code on Multilingual Models_
987
 
988
+ > **Note (2026-05-18, updated 2026-10-13):** Current Phase 3 configs use the short `condition-*` namespace and include `103k`, `20k`, and `5k` sizes for Conditions 1--2. Phase 2 configs remain available under the `phase-2-the-stack-v1-*` namespace for reproducibility. `condition-3-zh-5k` and `condition-4-zh-5k` are the Phase 3 names for the community-collected native-code corpora; they read the same files as `phase-2-the-stack-v1-condition-{3,4}-zh-5k` because those corpora are unchanged across phases.
989
 
990
  Multilingual Python code datasets for the **Language Decoded** project (part of [Cohere's Tiny Aya Expedition](https://aya.for.ai)). See [legesher/language-decoded-experiments](https://huggingface.co/datasets/legesher/language-decoded-experiments) for the canonical project description, the full experimental ladder, and the paper-grade evaluation results.
991