List your company

Inference Throughput on SQuAD (Hallucination Detection)

5,303Inference Throughput (Samples/sec)Max Pooling
-212.02641,219.75932,651.5454,083.3307May 9, 2026

Evaluation results

MethodLinks
Max PoolingModel=LLaMA-3.1-8B2026.055,303
Mean PoolingModel=LLaMA-3.1-8B2026.055,274
Max PoolingModel=Mistral-Nemo-Instruct (12B)2026.054,288
Mean PoolingModel=Mistral-Nemo-Instruct (12B)2026.054,237
HaMI (Original)Model=LLaMA-3.1-8B2026.053,402
HaMI (Original)Model=Mistral-Nemo-Instruct (12B)2026.053,092
Mean PoolingModel=LLaMA-3.3-Instruct-70B2026.052,637
Max PoolingModel=LLaMA-3.3-Instruct-70B2026.052,585
HaMI (Original)Model=LLaMA-3.3-Instruct-70B2026.052,436
HaMI (SP)Model=Mistral-Nemo-Instruct (12B)2026.050.12
HaMI (SP)Model=LLaMA-3.3-Instruct-70B2026.050.1
HaMI (SP)Model=LLaMA-3.1-8B2026.050.09