List your company
ResearchFormerly WizWand

Knowledge Base Question Answering on GraphQ (test)

69.47F1DARA (w/Mistral-7B)
19.581232.533145.48558.4369Jun 11, 2024Oct 8, 2024Feb 4, 2025Jun 4, 2025Oct 1, 2025Jan 28, 2026May 28, 2026
Updated 5mo ago

Evaluation results

MethodLinks
DARA (w/Mistral-7B)Learning Mode=Fine-tuned LLM Agent, Backbone=Mistral-7B, Zero-shot=true, Entity Linker=Golden2024.0669.4756.96
DARA (w/Llama-2-13B)Learning Mode=Fine-tuned LLM Agent, Backbone=Llama-2-13B, Zero-shot=true, Entity Linker=Golden2024.0667.3455.57
Pangu (T5-base)Learning Mode=Bottom-up Parser, Data=full train data, Backbone=T5-base, Zero-shot=true, Entity Linker=Golden2024.0666.756.06
AgentBench (GPT-4)Learning Mode=Off-the-shelf (in-context learning), Backbone=GPT-4, Zero-shot=true, Entity Linker=Golden2024.0664.4853.86
DARA (w/CodeLlama-7B)Learning Mode=Fine-tuned LLM Agent, Backbone=CodeLlama-7B, Zero-shot=true, Entity Linker=Golden2024.0663.1551.83
DARA (w/Llama-2-7B)Learning Mode=Fine-tuned LLM Agent, Backbone=Llama-2-7B, Zero-shot=true, Entity Linker=Golden2024.0662.7451.51
GAPDLLM=Llama-3.1-8B, Category=Fine-tune-based Methods2026.0555.1—
AgentLM-13BLearning Mode=Fine-tuned LLM Agent, Backbone=13B, Zero-shot=true, Entity Linker=Golden2024.0654.3544.34
KBQA-R1LLM=Llama-3.1-8B, Evaluation Protocol=Fine-tune-based2025.1253.8—
KBQA-R1LLM=Llama-3.1-8B, Category=Fine-tune-based Methods2026.0553.8—
AgentBench-7BLearning Mode=Fine-tuned LLM Agent, Backbone=7B, Zero-shot=true, Entity Linker=Golden2024.0653.3742.72
KBQA-R1 HarnessLLM=Kimi-K2.5, Category=Prompting Methods2026.0553.3—
KBQA-R1 HarnessLLM=Claude 4.6 Sonnet, Category=Prompting Methods2026.0551.8—
KBQA-R1 HarnessLLM=gpt-5.3-codex, Category=Prompting Methods2026.0551.5—
KBQA-o1LLM=Llama-3.1-8B, Evaluation Protocol=Fine-tune-based2025.1248.7—
KBQA-o1LLM=Llama-3.1-8B, Category=Fine-tune-based Methods2026.0548.7—
KBQA-R1 HarnessLLM=GLM-5, Category=Prompting Methods2026.0548.5—
ArcaneQALearning Mode=Bottom-up Parser, Data=full train data, Zero-shot=true, Entity Linker=Golden2024.0647.537
CoTKRLLM=Llama-3-8B, Evaluation Protocol=Fine-tune-based2025.1247.5—
CoTKRLLM=Llama-3-8B, Category=Fine-tune-based Methods2026.0547.5—
AgentLM-7BLearning Mode=Fine-tuned LLM Agent, Backbone=7B, Zero-shot=true, Entity Linker=Golden2024.0643.9236.21
ArcaneQALLM=BERT-base, Evaluation Protocol=Fine-tune-based2025.1231.8—
ArcaneQALLM=BERT-base, Category=Fine-tune-based Methods2026.0531.8—
KB-CoderLLM=GPT-3.5-turbo, Evaluation Protocol=Prompting2025.1231.1—
KB-CoderLLM=GPT-3.5-turbo, Category=Prompting Methods2026.0531.1—
AgentBench (Llama-2-chat-70B)Learning Mode=Off-the-shelf (in-context learning), Backbone=Llama-2-chat-70B, Zero-shot=true, Entity Linker=Golden2024.0630.3325.63
KB-BINDERLLM=Codex-davinci-002, Evaluation Protocol=Prompting2025.1227.1—
KB-BINDERLLM=Codex-davinci-002, Category=Prompting Methods2026.0527.1—
BERT+RankingLLM=BERT-base, Evaluation Protocol=Fine-tune-based2025.1225—
BERT+RankingLLM=BERT-base, Category=Fine-tune-based Methods2026.0525—
SPARQALLM=BERT-base, Evaluation Protocol=Fine-tune-based2025.1221.5—
SPARQALLM=BERT-base, Category=Fine-tune-based Methods2026.0521.5—
Pangu (T5-Large)Learning Mode=Bottom-up Parser, Data=full train data, Backbone=T5-Large, Zero-shot=true, Entity Linker=Golden2024.06—55.57