General Capability on All Benchmarks Overall
52.04Overall Average ScoreUltraMix
Evaluation Results
| Method | Links | |
|---|---|---|
| UltraMixTraining Method=DPO, Dataset=UM-190k2025.11 | 52.04 | |
| UltraMixTraining Method=DPO, Dataset=UM-187k2025.11 | 51.74 | |
| TuluDPOTraining Method=DPO, Dataset=TuluDPO2025.11 | 50.87 | |
| UltraMixTraining Method=DPO, Dataset=UM-170k2025.11 | 50.55 | |
| ORPOTraining Method=ORPO, Dataset=ORPO2025.11 | 49.55 | |
| DataAgentRLModel Backbone=LLaMA-DCLM, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 46.84 | |
| SmolLM-3-3B-SFTTraining Method=SFT2025.11 | 46.61 | |
| HelpSteerTraining Method=DPO, Dataset=HelpSteer2025.11 | 46.51 | |
| CodePrefTraining Method=DPO, Dataset=CodePref2025.11 | 45.92 | |
| DataAgentRLModel Backbone=LLaMA-Nemo, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 45.9 | |
| DataAgentSFTModel Backbone=LLaMA-DCLM, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 45.75 | |
| UltraFBTraining Method=DPO, Dataset=UltraFB2025.11 | 45.69 | |
| DataAgentRLModel Backbone=LLaMA-FWE, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 45.55 | |
| DataAgentSFTModel Backbone=LLaMA-FWE, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 45.21 | |
| DataAgentSFTModel Backbone=LLaMA-Nemo, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 45.16 | |
| RegMixModel Backbone=LLaMA-Nemo, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 44.86 | |
| RegMixModel Backbone=LLaMA-DCLM, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 44.67 | |
| RegMixModel Backbone=LLaMA-FWE, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 43.78 | |
| Naive TrainingModel Backbone=LLaMA-DCLM, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 38.51 | |
| Naive TrainingModel Backbone=LLaMA-FWE, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 38.51 | |
| Base ModelModel Backbone=LLaMA-Nemo, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 38.22 | |
| Base ModelModel Backbone=LLaMA-DCLM, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 38.15 | |
| Naive TrainingModel Backbone=LLaMA-Nemo, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 37.86 | |
| Base ModelModel Backbone=LLaMA-FWE, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 37.65 | |
| DataAgentRLModel Backbone=Pythia-1.4B, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 35.25 | |
| RegMixModel Backbone=Pythia-1.4B, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 34.64 | |
| DataAgentSFTModel Backbone=Pythia-1.4B, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 34.01 | |
| Base ModelModel Backbone=Pythia-1.4B, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 33.47 | |
| Naive TrainingModel Backbone=Pythia-1.4B, Data Reweighting Space=52-dimensional (Nvidia domain classifier)2025.07 | 31.06 |