Orpo Dpo Mix 40K: Portuguese Translation of a Preference Tuning Dataset

A Portuguese translation of the 'mlabonne/orpo-dpo-mix-40k' dataset, created by user BornSaint and last updated on 2025-05 06. The dataset was translated using a quantized machine translation model over more than a week on a single GPU thread. The original dataset is likely used for preference tuning and reinforcement learning from human feedback.

Use Cases

• Fine-tuning language models for Portuguese text generation based on translated preference pairs.
• Training reward models for Portuguese dialogue systems based on the translated preference data.
• Conducting comparative studies on preference tuning methodologies across different languages.
• Developing instruction-following models for Portuguese based on translated instruction-response pairs.

Strengths

• Dataset is a direct translation of a known preference-tuning dataset ('mlabonne/orpo-dpo-mix-40k').
• Translation process is documented, specifying the model ('google/madlad400-3b-mt') and hardware used.
• Last update timestamp (2025-05-06) is provided, indicating recent activity.

Limitations

• Row count, column definitions, and sample data are unavailable, limiting suitability assessment.
• License information is unknown, which may restrict usage.
• Translation quality and potential biases introduced by the automated process are not evaluated.

Provenance

Source

Hugging Face user BornSaint, derived from 'mlabonne/orpo-dpo-mix-40k'.

Collection Method

Automated translation using a quantized ctranslate2 version of 'google/madlad400-3b-mt'.

Time Range

null

Freshness

Last updated 2025-05-06 00:18:02.

Geography

null