Hallucination Detection on TriviaQA (test)
98.3AUC-ROCMultiHaluDet
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| MultiHaluDetBackbone=Mistral-7B2026.05 | 98.3 | — | — | — | — | — | — | — | — | |
| MultiHaluDetLLM Backbone=Mistral-7B2026.05 | 98.3 | — | — | — | — | — | — | — | — | |
| MultiHaluDetLLM Backbone=LLaMA2-7B2026.05 | 98.26 | — | — | — | — | — | — | — | — | |
| MultiHaluDetBackbone=LLaMA2-7B2026.05 | 98.2 | — | — | — | — | — | — | — | — | |
| HARPModels=LLaMA-3.1-8B, Single=true2025.09 | 92.9 | — | — | — | — | — | — | — | — | |
| HARPModels=Qwen-2.5-7B-Instruct, Single=true2025.09 | 92.8 | — | — | — | — | — | — | — | — | |
| MLP Probe2025.12 | 92.23 | — | 68.91 | 92.23 | 38.86 | 44.77 | — | — | — | |
| TriLensLLM=Gemma-2-9B2026.05 | 91.06 | — | — | — | — | — | — | — | — | |
| Spilled ΔEModel=Gemma-Instruct 4B, Pool=Min2026.02 | 89.01 | — | — | — | — | — | — | — | — | |
| TriLensLLM=Llama-3-8B2026.05 | 88.61 | — | — | — | — | — | — | — | — | |
| ACSELLM Backbone=Mistral-7B-Instruct2026.05 | 88 | 86 | — | — | — | — | 31 | 28 | 85 | |
| SpikeScoreLLM Backbone=Qwen3-14B, Method Category=Cross-domain specialized methods2026.01 | 86.97 | — | — | — | — | — | — | — | — | |
| CDESLLM Backbone=Vicuna-13B2025.05 | 86.3 | — | — | — | — | — | — | — | — | |
| HCPDK=5, Inf. Time (s)=1.1313, GPU Mem. (MiB)=190512026.06 | 86.25 | — | — | — | — | — | — | — | — | |
| ODESLLM Backbone=Vicuna-13B2025.05 | 85.9 | — | — | — | — | — | — | — | — | |
| HaloScopeModels=Qwen-2.5-7B-Instruct, Single=true2025.09 | 85.3 | — | — | — | — | — | — | — | — | |
| HCPDK=1, Inf. Time (s)=0.2349, GPU Mem. (MiB)=185132026.06 | 85.21 | — | — | — | — | — | — | — | — | |
| SEBase Model=Llama 3.1 8B2026.02 | 85.2 | — | — | — | — | — | — | — | — | |
| SDESLLM Backbone=Mistral-7B-v0.32025.05 | 85.1 | — | — | — | — | — | — | — | — | |
| Best BaselineBackbone=Mistral-7B2026.05 | 85.1 | — | — | — | — | — | — | — | — | |
| Neural SDEsLLM Backbone=Mistral-7B2026.05 | 85.1 | — | — | — | — | — | — | — | — | |
| SAPLMALLM Backbone=Vicuna-13B2025.05 | 85 | — | — | — | — | — | — | — | — | |
| CDESLLM Backbone=Gemma-2-9B2025.05 | 85 | — | — | — | — | — | — | — | — | |
| TDGNetgraph_type=dynamic, decoding_gen_length=32, decoding_steps=16, granularity=token-level2026.02 | 85 | — | — | — | — | — | — | — | — | |
| CDESLLM Backbone=Llama-2-13B2025.05 | 84.9 | — | — | — | — | — | — | — | — | |
| ODESLLM Backbone=Gemma-2-9B2025.05 | 84.6 | — | — | — | — | — | — | — | — | |
| MINDLLM Backbone=Mistral-7B-v0.32025.05 | 84.5 | — | — | — | — | — | — | — | — | |
| MINDLLM Backbone=Mistral-7B2026.05 | 84.5 | — | — | — | — | — | — | — | — | |
| Probe@ExactLLM Backbone=Vicuna-13B2025.05 | 84.4 | — | — | — | — | — | — | — | — | |
| SDESLLM Backbone=Vicuna-13B2025.05 | 84.3 | — | — | — | — | — | — | — | — | |
| Marginal EmModel=Gemma-Instruct 4B, Pool=Max2026.02 | 84.14 | — | — | — | — | — | — | — | — | |
| SpikeScoreLLM Backbone=Qwen3-8B, Method Category=Cross-domain specialized methods2026.01 | 84.13 | — | — | — | — | — | — | — | — | |
| SAPLMALLM Backbone=Mistral-7B-v0.32025.05 | 84.1 | — | — | — | — | — | — | — | — | |
| Probe@ExactLLM Backbone=Mistral-7B-v0.32025.05 | 84.1 | — | — | — | — | — | — | — | — | |
| CDESLLM Backbone=Mistral-7B-v0.32025.05 | 84.1 | — | — | — | — | — | — | — | — | |
| SAPLMALLM Backbone=Mistral-7B2026.05 | 84.1 | — | — | — | — | — | — | — | — | |
| Probe@ExactLLM Backbone=Mistral-7B2026.05 | 84.1 | — | — | — | — | — | — | — | — | |
| Neural CDEsLLM Backbone=Mistral-7B2026.05 | 84.1 | — | — | — | — | — | — | — | — | |
| Probe@ExactLLM Backbone=Gemma-2-9B2025.05 | 84 | — | — | — | — | — | — | — | — | |
| ODESLLM Backbone=Mistral-7B-v0.32025.05 | 83.7 | — | — | — | — | — | — | — | — | |
| CDESLLM Backbone=Llama-2-7B2025.05 | 83.7 | — | — | — | — | — | — | — | — | |
| Best BaselineBackbone=LLaMA2-7B2026.05 | 83.7 | — | — | — | — | — | — | — | — | |
| Neural ODEsLLM Backbone=Mistral-7B2026.05 | 83.7 | — | — | — | — | — | — | — | — | |
| Neural CDEsLLM Backbone=LLaMA2-7B2026.05 | 83.7 | — | — | — | — | — | — | — | — | |
| Marginal EmModel=Gemma-Instruct 1B, Pool=Max2026.02 | 83.66 | — | — | — | — | — | — | — | — | |
| ODESLLM Backbone=Llama-2-13B2025.05 | 83.6 | — | — | — | — | — | — | — | — | |
| SAPLMALLM Backbone=Gemma-2-9B2025.05 | 83.4 | — | — | — | — | — | — | — | — | |
| Linear Probe2025.12 | 83.36 | — | 17.3 | 83.36 | 11.78 | 32.55 | — | — | — | |
| SDESLLM Backbone=Llama-2-13B2025.05 | 83.3 | — | — | — | — | — | — | — | — | |
| MINDLLM Backbone=Vicuna-13B2025.05 | 83.2 | — | — | — | — | — | — | — | — | |
| SDESLLM Backbone=Gemma-2-9B2025.05 | 83.2 | — | — | — | — | — | — | — | — | |
| SEBase Model=SmolLM3 3B2026.02 | 83.2 | — | — | — | — | — | — | — | — | |
| PerplexityModels=Qwen-2.5-7B-Instruct, Single=true2025.09 | 83.1 | — | — | — | — | — | — | — | — | |
| CHARMgraph_type=static, decoding_gen_length=32, decoding_steps=16, granularity=token-level2026.02 | 83 | — | — | — | — | — | — | — | — | |
| SAPLMALLM=Qwen2.5-7B2026.05 | 82.97 | — | — | — | — | — | — | — | — | |
| INSIDELLM Backbone=Llama-2-13B2025.05 | 82.6 | — | — | — | — | — | — | — | — | |
| CDESLLM Backbone=Alpaca-13B2025.05 | 82.6 | — | — | — | — | — | — | — | — | |
| Probe@ExactLLM Backbone=Llama-2-13B2025.05 | 82.5 | — | — | — | — | — | — | — | — | |
| MINDLLM Backbone=Llama-2-13B2025.05 | 82.3 | — | — | — | — | — | — | — | — | |
| INSIDELLM Backbone=Gemma-2-9B2025.05 | 82 | — | — | — | — | — | — | — | — | |
| Probe@ExactLLM Backbone=Alpaca-13B2025.05 | 81.9 | — | — | — | — | — | — | — | — | |
| EUBHDLLM Backbone=Gemma-2-9B2025.05 | 81.7 | — | — | — | — | — | — | — | — | |
| INSIDELLM Backbone=Llama-2-7B2025.05 | 81.7 | — | — | — | — | — | — | — | — | |
| ODESLLM Backbone=Llama-2-7B2025.05 | 81.7 | — | — | — | — | — | — | — | — | |
| INSIDELLM Backbone=LLaMA2-7B2026.05 | 81.7 | — | — | — | — | — | — | — | — | |
| Neural ODEsLLM Backbone=LLaMA2-7B2026.05 | 81.7 | — | — | — | — | — | — | — | — | |
| SEBase Model=Qwen3-8B2026.02 | 81.6 | — | — | — | — | — | — | — | — | |
| SDESLLM Backbone=Alpaca-13B2025.05 | 81.5 | — | — | — | — | — | — | — | — | |
| INSIDELLM Backbone=Mistral-7B-v0.32025.05 | 81.3 | — | — | — | — | — | — | — | — | |
| INSIDELLM Backbone=Mistral-7B2026.05 | 81.3 | — | — | — | — | — | — | — | — | |
| Probe@ExactLLM Backbone=LLaMA2-7B2026.05 | 81.3 | — | — | — | — | — | — | — | — | |
| EUBHDLLM Backbone=Llama-2-13B2025.05 | 81.1 | — | — | — | — | — | — | — | — | |
| SAPLMALLM Backbone=Llama-2-13B2025.05 | 81.1 | — | — | — | — | — | — | — | — | |
| MINDLLM Backbone=Alpaca-13B2025.05 | 81.1 | — | — | — | — | — | — | — | — | |
| MINDLLM Backbone=Gemma-2-9B2025.05 | 81.1 | — | — | — | — | — | — | — | — | |
| TriLensLLM=Qwen2.5-7B2026.05 | 81.03 | — | — | — | — | — | — | — | — | |
| SDESLLM Backbone=Llama-2-7B2025.05 | 81 | — | — | — | — | — | — | — | — | |
| Neural SDEsLLM Backbone=LLaMA2-7B2026.05 | 81 | — | — | — | — | — | — | — | — | |
| INSIDELLM Backbone=Vicuna-13B2025.05 | 80.8 | — | — | — | — | — | — | — | — | |
| Spilled ΔEModel=Gemma-Instruct 1B, Pool=Min2026.02 | 80.73 | — | — | — | — | — | — | — | — | |
| SEBase Model=Llama 3.2 3B2026.02 | 80.7 | — | — | — | — | — | — | — | — | |
| PerplexityInf. Time (s)=0.1349, GPU Mem. (MiB)=171852026.06 | 80.62 | — | — | — | — | — | — | — | — | |
| EUBHDLLM Backbone=Mistral-7B-v0.32025.05 | 80.6 | — | — | — | — | — | — | — | — | |
| EUBHDLLM Backbone=Mistral-7B2026.05 | 80.6 | — | — | — | — | — | — | — | — | |
| EUBHDLLM Backbone=Llama-2-7B2025.05 | 80.5 | — | — | — | — | — | — | — | — | |
| ODESLLM Backbone=Alpaca-13B2025.05 | 80.5 | — | — | — | — | — | — | — | — | |
| SEBase Model=Ministral 8B2026.02 | 80.5 | — | — | — | — | — | — | — | — | |
| EUBHDLLM Backbone=LLaMA2-7B2026.05 | 80.5 | — | — | — | — | — | — | — | — | |
| Probe@ExactLLM Backbone=Llama-2-7B2025.05 | 80.3 | — | — | — | — | — | — | — | — | |
| SAPLMALLM Backbone=Alpaca-13B2025.05 | 80.2 | — | — | — | — | — | — | — | — | |
| PCPBase Model=Qwen3-4B2026.02 | 80.2 | — | — | — | — | — | — | — | — | |
| LN-EntropyModels=Qwen-2.5-7B-Instruct, Single=false2025.09 | 80.2 | — | — | — | — | — | — | — | — | |
| ICR ProbeLLM=Gemma-2-9B2026.05 | 80.01 | — | — | — | — | — | — | — | — | |
| SAPLMALLM Backbone=Llama-2-7B2025.05 | 80 | — | — | — | — | — | — | — | — | |
| CAPLLM Backbone=Mistral-7B-Instruct2026.05 | 80 | 79 | — | — | — | — | 48 | 41 | 76 | |
| SAPLMALLM Backbone=LLaMA2-7B2026.05 | 80 | — | — | — | — | — | — | — | — | |
| TSVInf. Time (s)=0.0934, GPU Mem. (MiB)=235922026.06 | 79.78 | — | — | — | — | — | — | — | — | |
| PCPBase Model=SmolLM3 3B2026.02 | 79.7 | — | — | — | — | — | — | — | — | |
| SpikeScoreLLM Backbone=Llama3.1-8B, Method Category=Cross-domain specialized methods2026.01 | 79.47 | — | — | — | — | — | — | — | — | |
| MINDLLM Backbone=Llama-2-7B2025.05 | 79.4 | — | — | — | — | — | — | — | — |