Translation
Transformers
Safetensors
qwen3
text-generation
text-generation-inference

The Turkish-English translation model that achieved the highest Bleu score with the Flores test dataset.

#3
by NovaYear - opened

BLEU Skore: 41.45
Detailed Scores: [70.87941236482351, 47.636874122601775, 34.42178684081923, 25.410255218260424]
Brevity Penalty (BP): 1.000
Reference Length: 24360
Hypothesis Length: 24505

However, despite the excellent results, the translation process is very slow. Translating 1000 Flores test data points took almost 30 minutes. Alternatively, using QuickMT models, which translate 1000 data points in just 2 minutes, might be much more time-efficient. I should also mention that the QuickMT model has a SacreBleu score of 39.44.

Sign up or log in to comment