Language Decoded: Multilingual Python Code Datasets for Model Training

Language Decoded Data is a multilingual code dataset for the Language Decoded project, part of Cohere's research. The dataset includes configurations for Phase 3 with sizes of 103k, 20k, and 5k rows for Conditions 1 and 2, and Phase 2 configurations remain available for reproducibility. It was last updated by user 'legesher' on Hugging Face on 2026-05-31.

Use Cases

• Training multilingual code generation models based on the described multilingual Python code data.
• Experimenting with the impact of native code on model performance based on the project's research focus.
• Reproducing results from the Language Decoded project using the available Phase 2 configurations.
• Benchmarking model performance across different dataset sizes (e.g., 103k, 20k, 5k) as mentioned in the description.

Strengths

• Includes specific dataset sizes (103k, 20k, 5k) for different experimental conditions.
• Maintains Phase 2 configurations for reproducibility of prior research.
• Last update timestamp (2026-05-31) is explicitly provided.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown for the overall dataset, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

legesher on Hugging Face, part of Cohere's Language Decoded project.

Freshness

Last updated 2026-05-31 11:05:25; freshness should be verified.