Nutrient Estimation on FoodBench-QA (val)
55.46SugarTF-IDF + Gemini 2.5 Flash
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TF-IDF + Gemini 2.5 FlashModel Architecture=LLM-based Refinement2026.04 | 55.46 | 67.79 | 52.12 | 58.39 | 1 | |
| Gemini 2.5 FlashModel Architecture=LLM-based Direct Inference2026.04 | 55.36 | 67.17 | 54.12 | 59.73 | 1 | |
| TF-IDF + Ridge RegressionModel Architecture=Traditional Machine Learning2026.04 | 50.26 | 64.62 | 40.58 | 50.33 | 1 | |
| Gemma-3-27BModel Architecture=LLM-based Direct Inference, Number of parameters=27B, Quantization=Q4_K_M2026.04 | 41.51 | 60.21 | 41.44 | 48.02 | 1.4 | |
| GPT-OSS-20BModel Architecture=LLM-based Direct Inference, Number of parameters=20B, Quantization=MXFP42026.04 | 32.07 | 41.38 | 29.93 | 37 | 9.9 | |
| Nemotron-3-Nano-30BModel Architecture=LLM-based Direct Inference, Number of parameters=30B2026.04 | 22.05 | 30.58 | 19.73 | 26.23 | 23.7 | |
| DeBERTa-v3Model Architecture=Encoder-based model2026.04 | 8.87 | 23.86 | 8.27 | 8.01 | 3.58 |