List your company
ResearchFormerly WizWand

Knowledge Retrieval on TriviaQA

62.11Exact Match (EM)W/O (FP16)
-1.725214.847431.4247.9926Nov 9, 2025
Updated 5mo ago

Evaluation results

MethodLinks
W/O (FP16)Backbone=Gemma-2-9B2025.1162.11—68.9217.21
SliM-LLMBackbone=Gemma-2-9B2025.1161.23—68.878.31
TAQ-KLBackbone=Gemma-2-9B2025.1161.04—68.586.76
TAQ-ISBackbone=Gemma-2-9B2025.1159.67—68.086.76
TAQ-OBackbone=Gemma-2-9B2025.1159.33—67.679.02
GPTQBackbone=Gemma-2-9B2025.1152.64—61.045.75
AWQBackbone=Gemma-2-9B2025.1151.51—60.145.74
W/O (FP16)Backbone=Qwen2.5-3B2025.1117.24—24.695.75
W/O (FP16)Backbone=Qwen2.5-7B2025.1116.94—25.6214.19
GPTQBackbone=Qwen2.5-7B2025.1115.97—26.575.19
SliM-LLMBackbone=Qwen2.5-7B2025.1115.77—26.25.83
AWQBackbone=Qwen2-7B2025.1114.79—25.165.19
TAQ-KLBackbone=Qwen2.5-7B2025.1114.06—23.835.95
TAQ-ISBackbone=Qwen2.5-7B2025.1112.55—23.955.95
TAQ-OBackbone=Qwen2.5-7B2025.1112.26—23.239.05
AWQBackbone=Qwen2.5-3B2025.119.72—17.662.5
TAQ-OBackbone=Qwen2.5-3B2025.118.45—20.343.2
GPTQBackbone=Qwen2.5-3B2025.117.23—15.541.93
AWQBackbone=Qwen2.5-7B2025.116.98—18.155.19
GPTQBackbone=Qwen2-7B2025.115.71—17.15.19
TAQ-ISBackbone=Qwen2.5-3B2025.113.91—17.142.25
SliM-LLMBackbone=Qwen2.5-3B2025.113.22—17.482.77
SliM-LLMBackbone=Qwen2-7B2025.112.44—21.095.83
TAQ-ISBackbone=Qwen2-7B2025.112.25—20.795.95
TAQ-KLBackbone=Qwen2-7B2025.112.25—21.175.95
TAQ-OBackbone=Qwen2-7B2025.111.86—20.589.05
TAQ-KLBackbone=Qwen2.5-3B2025.111.32—13.692.25
W/O (FP16)Backbone=Qwen2-7B2025.110.73—12.9414.19
MTPLoop strategy=1 to 8, Training protocol=Continued pre-training, Training data size=70B tokens2026.03—3.353——
MTPLoop strategy=4 to 8, Training protocol=Continued pre-training, Training data size=70B tokens2026.03—3.324——
MTP-DLoop strategy=1, Training protocol=Continued pre-training, Training data size=70B tokens2026.03—1.114——
MTP-DLoop strategy=4, Training protocol=Continued pre-training, Training data size=70B tokens2026.03—2.782——
MTP-DLoop strategy=1 to 8, Training protocol=Training Free, Training data size=70B tokens2026.03—2.263——
MTP-DLoop strategy=1 to 8, Training protocol=Continued pre-training, Training data size=70B tokens2026.03—3.455——
MTP-DLoop strategy=4 to 8, Training protocol=Training Free, Training data size=70B tokens2026.03—3.192——
MTP-DLoop strategy=4 to 8, Training protocol=Continued pre-training, Training data size=70B tokens2026.03—3.985——
MTP-DLoop strategy=4 to 8, Training protocol=Continued pre-training, Training data size=350B tokens2026.03—3.972——
MTP-DLoop strategy=1 to 16, Training protocol=Training Free, Training data size=70B tokens2026.03—2.687——
MTP-DLoop strategy=1 to 16, Training protocol=Continued pre-training, Training data size=70B tokens2026.03—3.157——
MTP-DLoop strategy=4 to 16, Training protocol=Training Free, Training data size=70B tokens2026.03—3.741——
MTP-DLoop strategy=4 to 16, Training protocol=Continued pre-training, Training data size=70B tokens2026.03—4.765——
MTP-DLoop strategy=4 to 8 to 16, Training protocol=Continued pre-training, Training data size=70B tokens2026.03—4.323——
MTP-D ensembleLoop strategy=4 to 8, Training protocol=Continued pre-training, Training data size=70B tokens2026.03—3.959——