List your company

Machine Translation on FLORES-200 TR-EN (test)

19.1BLEUMix-MoE
9.42811.93914.4516.961May 23, 2026

Evaluation results

MethodLinks
Mix-MoEArchitecture Strategy=Mix-MoE2026.0519.138.873.2
MOE-LPRArchitecture Strategy=MOE-LPR2026.051837.172.1
MoEArchitecture Strategy=MoE2026.0517.233.867.2
Dense PT+SFTTraining Strategy=Single-stage Fine-tuning baseline2026.0516.837.172.5
Full FinetuneTraining Strategy=Full Finetune2026.0516.236.572.1
LLAMA-ProTraining Strategy=LLAMA-Pro2026.051530.566.5
Qwen2.5-1.5BModel Scale=1.5B2026.0514.526.564.5
LoRATraining Strategy=LoRA2026.0514.528.266.1
LLAMA-MOE v2Architecture Strategy=LLAMA-MOE v22026.0513.822.865.1
Qwen2.5-0.5BModel Scale=0.5B2026.0511.22158.9
Llama3.2-1BModel Scale=1B2026.059.818.555.4