List your company

Coreference Resolution on XWinograd

79.9AccuracyLANG
47.97256.26164.5572.839Jan 15, 2026Feb 5, 2026Feb 26, 2026Mar 19, 2026Apr 9, 2026Apr 30, 2026May 21, 2026

Evaluation results

MethodLinks
LANGbackbone=Qwen2.5-7B-Instruct2026.0579.979.9
BYOL-nyaParameters=12B, Pre-training Type=CPT2026.0170.37—
HQ seedFiltering Strategy=HQ seed2026.0469.64—
MKC-eFiltering Strategy=MKC-e2026.0469.25—
ML (15%)Filtering Strategy=ML (15%)2026.0468.65—
BYOL-nyaParameters=4B, Pre-training Type=CPT2026.0168.34—
HQFiltering Strategy=HQ2026.0468.25—
No filteringFiltering Strategy=No filtering2026.0468.06—
MLFiltering Strategy=ML2026.0467.66—
ML (Q3)Filtering Strategy=ML (Q3)2026.0466.67—
Qwen2.5-7B-Instruct2026.0565.765.7
Vanilla GRPObackbone=Qwen2.5-7B-Instruct2026.0565.765.7
BYOL-nyaParameters=1B, Pre-training Type=CPT2026.0163.32—
BYOL-mriParameters=12B, Training Stage=CPT2026.0162.67—
LC-GRPObackbone=Qwen2.5-7B-Instruct2026.0562.262.2
Gemma-3Parameters=12B, Pre-training Type=PT2026.0161.39—
BYOL-mriParameters=4B, Training Stage=CPT2026.0159.68—
BYOL-mriParameters=1B, Training Stage=CPT2026.0157.33—
ApertusParameters=8B, Pre-training Type=25092026.0156.15—
Gemma-3Parameters=4B, Pre-training Type=PT2026.0154.76—
Qwen-3Parameters=8B, Pre-training Type=Base2026.0152.09—
Gemma-3Parameters=1B, Pre-training Type=PT2026.0151.34—
Qwen-3Parameters=14B, Pre-training Type=Base2026.0151.12—
Llama-3.2Parameters=1B2026.0150.59—
Llama-3.1Parameters=8B2026.0150.59—
Qwen-3Parameters=1.7B, Pre-training Type=Base2026.0149.2—