List your company

Question Answering on TriQA

83.43AccuracyCDKC
-2.089220.112942.31564.5171Sep 26, 2025Oct 19, 2025Nov 11, 2025Dec 5, 2025Dec 28, 2025Jan 20, 2026Feb 13, 2026

Evaluation results

MethodLinks
CDKCBackbone=Qwen2.5-14B-Instruct2026.0283.43
GRPOBackbone=Qwen2.5-14B-Instruct2026.0282.63
CoTBackbone=Qwen2.5-14B-Instruct2026.0269.6
RAGBackbone=Qwen2.5-14B-Instruct2026.0269.1
CDKCBackbone=Qwen2.5-3B-Instruct2026.0268.23
GRPOBackbone=Qwen2.5-3B-Instruct2026.0266.7
CGKEBackbone=Qwen2.5-14B-Instruct2026.0266.47
Vanilla SFTBackbone=Qwen2.5-14B-Instruct2026.0264.83
Vanilla LLMBackbone=Qwen2.5-14B-Instruct2026.0263
RAGBackbone=Qwen2.5-3B-Instruct2026.0261.7
EMoEBase Model=ERNIE-4.5-21B-A3B, Trained Activated Experts=6+2, MoE Routing Method=EMoE, Inference Budget (k')=6 + 22025.0961.32
Top-kBase Model=ERNIE-4.5-21B-A3B, Trained Activated Experts=6+2, MoE Routing Method=Top-k, Inference Budget (k')=6 + 22025.0961.1
Top-kBase Model=ERNIE-4.5-21B-A3B, Trained Activated Experts=6+2, MoE Routing Method=Top-k, Inference Budget (k')=12 + 22025.0960.24
EMoEBase Model=ERNIE-4.5-21B-A3B, Trained Activated Experts=6+2, MoE Routing Method=EMoE, Inference Budget (k')=12 + 22025.0960.13
EMoEBase Model=LoRAMoE, Trained Activated Experts=2, MoE Routing Method=EMoE, Inference Budget (k')=62025.0959.77
Top-kBase Model=LoRAMoE, Trained Activated Experts=2, MoE Routing Method=Top-k, Inference Budget (k')=62025.0959.53
EMoEBase Model=LoRAMoE, Trained Activated Experts=2, MoE Routing Method=EMoE, Inference Budget (k')=42025.0959.24
EMoEBase Model=ERNIE-4.5-21B-A3B, Trained Activated Experts=6+2, MoE Routing Method=EMoE, Inference Budget (k')=3 + 22025.0959.22
Top-kBase Model=ERNIE-4.5-21B-A3B, Trained Activated Experts=6+2, MoE Routing Method=Top-k, Inference Budget (k')=3 + 22025.0959.1
Top-kBase Model=LoRAMoE, Trained Activated Experts=2, MoE Routing Method=Top-k, Inference Budget (k')=42025.0959.06
EMoEBase Model=LoRAMoE, Trained Activated Experts=2, MoE Routing Method=EMoE, Inference Budget (k')=22025.0958.2
Top-kBase Model=LoRAMoE, Trained Activated Experts=2, MoE Routing Method=Top-k, Inference Budget (k')=22025.0957.62
EMoEBase Model=LoRAMoE, Trained Activated Experts=2, MoE Routing Method=EMoE, Inference Budget (k')=12025.0954.79
Top-kBase Model=LoRAMoE, Trained Activated Experts=2, MoE Routing Method=Top-k, Inference Budget (k')=12025.0954.32
Llama-3.1Size=8B, Type=AR2026.0152.1
EMoEBase Model=DeepSeek-V2-Lite, Trained Activated Experts=6+2, MoE Routing Method=EMoE, Inference Budget (k')=12 + 22025.0952.07
EMoEBase Model=DeepSeek-V2-Lite, Trained Activated Experts=6+2, MoE Routing Method=EMoE, Inference Budget (k')=6 + 22025.0951.58
CoTBackbone=Qwen2.5-3B-Instruct2026.0249.47
CGKEBackbone=Qwen2.5-3B-Instruct2026.0248.47
EMoEBase Model=DeepSeek-V2-Lite, Trained Activated Experts=6+2, MoE Routing Method=EMoE, Inference Budget (k')=3 + 22025.0947.81
Vanilla SFTBackbone=Qwen2.5-3B-Instruct2026.0247.37
Top-kBase Model=DeepSeek-V2-Lite, Trained Activated Experts=6+2, MoE Routing Method=Top-k, Inference Budget (k')=6 + 22025.0946.86
Top-kBase Model=DeepSeek-V2-Lite, Trained Activated Experts=6+2, MoE Routing Method=Top-k, Inference Budget (k')=12 + 22025.0946.38
Vanilla LLMBackbone=Qwen2.5-3B-Instruct2026.0245.07
Top-kBase Model=DeepSeek-V2-Lite, Trained Activated Experts=6+2, MoE Routing Method=Top-k, Inference Budget (k')=3 + 22025.0941.36
EMoEBase Model=OLMoE-1B-7B-0924, Trained Activated Experts=8, MoE Routing Method=EMoE, Inference Budget (k')=162025.0941.27
EMoEBase Model=OLMoE-1B-7B-0924, Trained Activated Experts=8, MoE Routing Method=EMoE, Inference Budget (k')=82025.0939.64
Top-kBase Model=OLMoE-1B-7B-0924, Trained Activated Experts=8, MoE Routing Method=Top-k, Inference Budget (k')=162025.0939.1
Top-kBase Model=OLMoE-1B-7B-0924, Trained Activated Experts=8, MoE Routing Method=Top-k, Inference Budget (k')=82025.0938.98
EMoEBase Model=OLMoE-1B-7B-0924, Trained Activated Experts=8, MoE Routing Method=EMoE, Inference Budget (k')=42025.0934.84
Top-kBase Model=OLMoE-1B-7B-0924, Trained Activated Experts=8, MoE Routing Method=Top-k, Inference Budget (k')=42025.0932.04
A3Size=8B, Type=A32026.0119.4
DiffuLlamaSize=7B, Type=DD2026.0118.5
DreamSize=7B, Type=DD2026.0118.3
A3Size=3B, Type=A32026.0115.9
A3Size=1B, Type=A32026.0110.2
PlaidSize=1B, Type=CD2026.011.2