LYRICAL Russian to English Machine Translation Model

By Silver Age Poets
Variant 0.1: ORPO-tuned Qwen2.5-3B Model Weights

Fine-tuned leveraging Odds Ratio Preference Optimization (ORPO) on our Russian-to-English song lyrics translation/localization dataset.
Our aim with this work is to foster a translation model capable of adaptively localizing idiomatic, formal/poetic/rhythmic, and performance-catered features of lyrical input texts, whilst retaining adequate accuracy at the level of direct semantic translation.
The training was held for 6 Epochs at Rank 128.

This is an experimental/WIP fine-tune, and probably first of many. We find that its zero-shot translations (without extensive specialized instruction templating/priming) are already generally more reminiscent of input songs' phrasal rhythm/meter and expressive tone, as compared to base model outputs.
However, it makes far more semantic errors/mis-attributions and/or arbitrary substitutions (straying from translation into extrapolative composition).
We would appreciate feedback/reports from anyone else who happens to try out this model, or its other variants (to be released in the near future).

Next, we will attempt to fine-tune a larger or/and an already translation-specialized model, and continue experimenting with hyperparameters/techniques.

Downloads last month
5
Safetensors
Model size
3.09B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for AlekseyCalvin/Lyrical_MT_ru2en_var1_Qwen25_3B_ORPO_r128ft

Base model

Qwen/Qwen2.5-3B
Finetuned
(747)
this model
Quantizations
1 model

Dataset used to train AlekseyCalvin/Lyrical_MT_ru2en_var1_Qwen25_3B_ORPO_r128ft