List your company
ResearchFormerly WizWand

Machine Translation on En-De (COMET, BLEU)

85.14COMETNAIT
43.321654.178365.03575.8917Mar 13, 2026Mar 15, 2026Mar 17, 2026Mar 19, 2026Mar 21, 2026Mar 23, 2026Mar 26, 2026
Updated 7mo ago

Evaluation results

MethodLinks
NAITBase Model=ALMA-7b, Data Selection=Top 30% via NAIT2026.0385.1429.02
CPLBase Model=Qwen3-8B, Input Type=Context-aware2026.0384.69—
CPLBase Model=Qwen3-8B, Input Type=Sentence-level2026.0384.63—
CPLBase Model=Llama-3-8B-Instruct, Input Type=Context-aware2026.0384.59—
CrossLossBase Model=Llama-3-8B-Instruct, Input Type=Context-aware2026.0384.56—
FullBase Model=ALMA-7b, Data Selection=Full dataset2026.0384.5428.24
IntraLossBase Model=Llama-3-8B-Instruct, Input Type=Context-aware2026.0384.45—
CrossLossBase Model=Qwen3-8B, Input Type=Context-aware2026.0384.44—
CPLBase Model=Llama-3-8B-Instruct, Input Type=Sentence-level2026.0384.44—
CrossLossBase Model=Llama-3-8B-Instruct, Input Type=Sentence-level2026.0384.42—
IntraLossBase Model=Llama-3-8B-Instruct, Input Type=Sentence-level2026.0384.41—
IntraLossBase Model=Qwen3-8B, Input Type=Context-aware2026.0384.32—
CrossLossBase Model=Qwen3-8B, Input Type=Sentence-level2026.0384.29—
IntraLossBase Model=Qwen3-8B, Input Type=Sentence-level2026.0384.23—
Ctx-Aware_CPOBase Model=Llama-3-8B-Instruct, Input Type=Context-aware2026.0384.14—
Ctx-Aware_CPOBase Model=Qwen3-8B, Input Type=Context-aware2026.0383.95—
Sent-Level_CPOBase Model=Llama-3-8B-Instruct, Input Type=Sentence-level2026.0383.86—
Sent+Ctx_tunedBase Model=Llama-3-8B-Instruct, Input Type=Context-aware2026.0383.84—
Sent+Ctx_tuned+Base Model=Llama-3-8B-Instruct, Input Type=Context-aware2026.0383.81—
Sent+Ctx_tunedBase Model=Llama-3-8B-Instruct, Input Type=Sentence-level2026.0383.69—
Sent+Ctx_tuned+Base Model=Qwen3-8B, Input Type=Context-aware2026.0383.68—
Sent+Ctx_tuned+Base Model=Llama-3-8B-Instruct, Input Type=Sentence-level2026.0383.67—
Sent-Level_CPOBase Model=Qwen3-8B, Input Type=Sentence-level2026.0383.65—
Sent+Ctx_tunedBase Model=Qwen3-8B, Input Type=Context-aware2026.0383.58—
Sent+Ctx_tuned+Base Model=Qwen3-8B, Input Type=Sentence-level2026.0383.49—
Ctx-Aware_tunedBase Model=Qwen3-8B, Input Type=Context-aware2026.0383.43—
Sent+Ctx_tunedBase Model=Qwen3-8B, Input Type=Sentence-level2026.0383.39—
Ctx-Aware_tunedBase Model=Llama-3-8B-Instruct, Input Type=Context-aware2026.0383.39—
Sent-Level_tunedBase Model=Qwen3-8B, Input Type=Sentence-level2026.0383.37—
Sent-Level_tunedBase Model=Llama-3-8B-Instruct, Input Type=Sentence-level2026.0383.26—
CPLBase Model=Qwen3-4B, Input Type=Context-aware2026.0383.25—
CPLBase Model=Qwen3-4B, Input Type=Sentence-level2026.0383.23—
CrossLossBase Model=Qwen3-4B, Input Type=Context-aware2026.0383.02—
CrossLossBase Model=Qwen3-4B, Input Type=Sentence-level2026.0382.86—
IntraLossBase Model=Qwen3-4B, Input Type=Context-aware2026.0382.83—
IntraLossBase Model=Qwen3-4B, Input Type=Sentence-level2026.0382.82—
Ctx-Aware_CPOBase Model=Qwen3-4B, Input Type=Context-aware2026.0382.74—
Rand.Base Model=ALMA-7b, Data Selection=Random 30%2026.0382.4428.29
Sent-Level_CPOBase Model=Qwen3-4B, Input Type=Sentence-level2026.0382.4—
Ctx-AwareBase Model=Qwen3-8B, Input Type=Context-aware2026.0382.29—
Sent-LevelBase Model=Qwen3-8B, Input Type=Sentence-level2026.0382.15—
Sent+Ctx_tunedBase Model=Qwen3-4B, Input Type=Context-aware2026.0382—
Sent+Ctx_tuned+Base Model=Qwen3-4B, Input Type=Context-aware2026.0381.85—
Sent+Ctx_tunedBase Model=Qwen3-4B, Input Type=Sentence-level2026.0381.78—
Sent+Ctx_tuned+Base Model=Qwen3-4B, Input Type=Sentence-level2026.0381.72—
Sent-Level_tunedBase Model=Qwen3-4B, Input Type=Sentence-level2026.0381.55—
Ctx-Aware_tunedBase Model=Qwen3-4B, Input Type=Context-aware2026.0381.52—
Ctx-AwareBase Model=Qwen3-4B, Input Type=Context-aware2026.0380.26—
Sent-LevelBase Model=Qwen3-4B, Input Type=Sentence-level2026.0379—
Sent-LevelBase Model=Llama-3-8B-Instruct, Input Type=Sentence-level2026.0378.32—
Ctx-AwareBase Model=Llama-3-8B-Instruct, Input Type=Context-aware2026.0344.93—