List your company

Language Modeling on DCLM (val)

2.589LossIsoLoCo
2.548522.821763.0953.36824Jun 3, 2026Jun 8, 2026Jun 13, 2026Jun 18, 2026Jun 23, 2026Jun 28, 2026Jul 3, 2026

Evaluation results

MethodLinks
IsoLoCoModel size=512M, Number of workers (R)=12026.072.589————
IsoLoCoModel size=512M, Number of workers (R)=22026.072.593————
IsoLoCoModel size=512M, Number of workers (R)=42026.072.599————
IsoLoCoModel size=512M, Number of workers (R)=82026.072.61————
IsoLoCoModel size=512M, Number of workers (R)=162026.072.628————
DiLoCoModel size=512M, Number of workers (R)=12026.072.63————
AdamW DPModel size=512M2026.072.633————
DiLoCoModel size=512M, Number of workers (R)=22026.072.639————
DiLoCoModel size=512M, Number of workers (R)=42026.072.653————
DiLoCoModel size=512M, Number of workers (R)=82026.072.675————
IsoLoCoModel size=512M, Number of workers (R)=322026.072.676————
UncompressedScale=1B, Pipeline stages=P=42026.062.68—4,096——
UncompressedScale=1B, Pipeline stages=P=82026.062.68—4,096——
DiLoCoModel size=512M, Number of workers (R)=162026.072.706————
IsoLoCoModel size=512M, Number of workers (R)=642026.072.716————
MAPLScale=1B, Pipeline stages=P=42026.062.721.385128—
MAPLScale=1B, Pipeline stages=P=82026.062.732.025128—
MAPL+ VQScale=1B, Pipeline stages=P=82026.062.742.325616—
MAPL+ VQScale=1B, Pipeline stages=P=42026.062.763.0125616—
DiLoCoModel size=512M, Number of workers (R)=322026.072.767————
MAPLScale=500M, Pipeline stages=P=42026.062.79-1.95126—
UncompressedScale=500M, Pipeline stages=P=42026.062.84—3,072——
UncompressedScale=500M, Pipeline stages=P=82026.062.84—3,072——
MAPLScale=500M, Pipeline stages=P=82026.062.8405126—
IsoLoCoModel size=178M, Number of workers (R)=12026.072.842————
IsoLoCoModel size=178M, Number of workers (R)=22026.072.852————
DiLoCo SGD + Iso-CModel size=178M, Number of workers (R)=12026.072.855————
DiLoCoModel size=512M, Number of workers (R)=642026.072.863————
IsoLoCoModel size=178M, Number of workers (R)=42026.072.864————
DiLoCoModel size=178M, Number of workers (R)=12026.072.869————
DiLoCo SGD + Iso-CModel size=178M, Number of workers (R)=22026.072.872————
MAPL+ VQScale=500M, Pipeline stages=P=82026.062.881.4925612—
AdamW DPModel size=178M2026.072.881————
DiLoCoModel size=178M, Number of workers (R)=22026.072.882————
DiLoCo SGDModel size=178M, Number of workers (R)=12026.072.887————
IsoLoCoModel size=178M, Number of workers (R)=82026.072.889————
DiLoCo SGD + Iso-CModel size=178M, Number of workers (R)=42026.072.891————
DiLoCo SGDModel size=178M, Number of workers (R)=22026.072.905————
DiLoCoModel size=178M, Number of workers (R)=42026.072.911————
MAPL+ VQScale=500M, Pipeline stages=P=42026.062.922.7525612—
DiLoCo SGD + Iso-CModel size=178M, Number of workers (R)=82026.072.929————
IsoLoCoModel size=178M, Number of workers (R)=162026.072.93————
DiLoCo SGDModel size=178M, Number of workers (R)=42026.072.937————
DiLoCoModel size=178M, Number of workers (R)=82026.072.95————
DiLoCo SGDModel size=178M, Number of workers (R)=82026.072.99————
DiLoCo SGD + Iso-CModel size=178M, Number of workers (R)=162026.072.992————
IsoLoCoModel size=178M, Number of workers (R)=322026.072.992————
DiLoCoModel size=178M, Number of workers (R)=162026.073.005————
SSNScale=1B, Pipeline stages=P=42026.063.0513.935128—
IsoLoCoModel size=178M, Number of workers (R)=642026.073.062————
DiLoCo SGDModel size=178M, Number of workers (R)=162026.073.064————
DiLoCo SGD + Iso-CModel size=178M, Number of workers (R)=322026.073.07————
DiLoCoModel size=178M, Number of workers (R)=322026.073.077————
SSNScale=1B, Pipeline stages=P=82026.063.0815.055128—
SSNScale=500M, Pipeline stages=P=42026.063.098.925126—
SSNScale=500M, Pipeline stages=P=82026.063.129.95126—
UncompressedScale=150M, Pipeline stages=P=42026.063.13—2,048——
UncompressedScale=150M, Pipeline stages=P=82026.063.13—2,048——
MAPLScale=150M, Pipeline stages=P=42026.063.1560.845124—
IsoLoCoModel size=178M, Number of workers (R)=1282026.073.16————
DiLoCo SGD + Iso-CModel size=178M, Number of workers (R)=642026.073.162————
MAPLScale=150M, Pipeline stages=P=82026.063.1651.115124—
MAPL+ VQScale=150M, Pipeline stages=P=42026.063.1651.112568—
MAPL+ VQScale=150M, Pipeline stages=P=82026.063.171.282568—
DiLoCo SGDModel size=178M, Number of workers (R)=322026.073.178————
DiLoCoModel size=178M, Number of workers (R)=642026.073.195————
SSN (AdamW version)Scale=500M, Pipeline stages=P=42026.063.2514.345126—
DiLoCo SGD + Iso-CModel size=178M, Number of workers (R)=1282026.073.28————
SSN (AdamW version)Scale=500M, Pipeline stages=P=82026.063.2915.95126—
DiLoCoModel size=178M, Number of workers (R)=1282026.073.355————
DiLoCo SGDModel size=178M, Number of workers (R)=642026.073.357————
SSN (AdamW version)Scale=1B, Pipeline stages=P=42026.063.3826.345128—
SSNScale=150M, Pipeline stages=P=42026.063.398.375124—
SSN (AdamW version)Scale=1B, Pipeline stages=P=82026.063.3926.425128—
SSNScale=150M, Pipeline stages=P=82026.063.48.635124—
SSN (AdamW version)Scale=150M, Pipeline stages=P=42026.063.4911.495124—
SSN (AdamW version)Scale=150M, Pipeline stages=P=82026.063.5212.465124—
DiLoCo SGDModel size=178M, Number of workers (R)=1282026.073.601————
Arch-WarmupModel Parameters=1B, Model Layers=16, Architecture=Transformer2026.06————22.64
BaselineModel Parameters=1B, Model Layers=16, Architecture=Transformer2026.06————165.62
QK-NormModel Parameters=1B, Model Layers=16, Architecture=Transformer2026.06————61.57
SoftcapModel Parameters=1B, Model Layers=16, Architecture=Transformer2026.06————43.38