Sentiment Analysis on SST (Accuracy, F1-score)
100AccuracyLlama-3.2-1B-Instruct
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Llama-3.2-1B-InstructInput Type=Clean, Execution Mode=Eager, Backend=Inductor2026.05 | 100 | — | — | |
| Llama-3.2-1B-InstructInput Type=Clean, Execution Mode=Compiled, Backend=Inductor2026.05 | 100 | — | — | |
| Llama-3.2-1B-InstructInput Type=Trigger, Execution Mode=Compiled, Backend=Inductor2026.05 | 100 | — | — | |
| Llama-3.2-3B-InstructInput Type=Clean, Execution Mode=Eager, Backend=Inductor2026.05 | 100 | — | — | |
| Llama-3.2-3B-InstructInput Type=Clean, Execution Mode=Compiled, Backend=Inductor2026.05 | 100 | — | — | |
| Llama-3.2-3B-InstructInput Type=Trigger, Execution Mode=Compiled, Backend=Inductor2026.05 | 100 | — | — | |
| Qwen2.5-1.5B-InstructInput Type=Clean, Execution Mode=Eager, Backend=Inductor2026.05 | 100 | — | — | |
| Qwen2.5-1.5B-InstructInput Type=Clean, Execution Mode=Compiled, Backend=Inductor2026.05 | 100 | — | — | |
| Qwen2.5-1.5B-InstructInput Type=Trigger, Execution Mode=Eager, Backend=Inductor2026.05 | 100 | — | — | |
| Qwen2.5-3B-InstructInput Type=Clean, Execution Mode=Eager, Backend=Inductor2026.05 | 100 | — | — | |
| Qwen2.5-3B-InstructInput Type=Clean, Execution Mode=Compiled, Backend=Inductor2026.05 | 100 | — | — | |
| Qwen2.5-3B-InstructInput Type=Trigger, Execution Mode=Compiled, Backend=Inductor2026.05 | 97.5 | — | — | |
| LaMini-LLaMA# of params=7B2023.04 | 93.6 | — | — | |
| LaMini-GPT-J# of params=6B2023.04 | 93 | — | — | |
| Llama-3.2-3B-InstructInput Type=Trigger, Execution Mode=Eager, Backend=Inductor2026.05 | 88.7 | — | — | |
| Order LossBackbone=CNN2021.06 | 86.2 | 86.2 | 0.041 | |
| Gate + OrderBackbone=CNN2021.06 | 86.1 | 86.2 | 0.047 | |
| Alpaca# of params=7B2023.04 | 85.8 | — | — | |
| Gate LossBackbone=CNN2021.06 | 85.2 | 85.4 | — | |
| Base LossBackbone=CNN2021.06 | 85.1 | 85.4 | — | |
| SL (Saliency Learning)Backbone=CNN2021.06 | 84.9 | 85.1 | — | |
| IGA (IG Attribution)Backbone=CNN2021.06 | 84.8 | 85.2 | — | |
| BaselineBackbone=CNN2021.06 | 84.7 | 85.1 | — | |
| Llama-3.2-1B-InstructInput Type=Trigger, Execution Mode=Eager, Backend=Inductor2026.05 | 80 | — | — | |
| DOPA*Backbone=Qwen3-8B2026.04 | 78.35 | — | — | |
| DrICLBackbone=Qwen3-8B2026.04 | 78.26 | — | — | |
| RandomBackbone=Qwen3-8B2026.04 | 76.85 | — | — | |
| DICLBackbone=Qwen3-8B2026.04 | 76.1 | — | — | |
| KNNBackbone=Qwen3-8B2026.04 | 76.01 | — | — | |
| InfICLBackbone=Qwen3-8B2026.04 | 75.35 | — | — | |
| Qwen2.5-1.5B-InstructInput Type=Trigger, Execution Mode=Compiled, Backend=Inductor2026.05 | 75 | — | — | |
| RewriteBackbone=Qwen3-8B2026.04 | 74.98 | — | — | |
| Qwen2.5-3B-InstructInput Type=Trigger, Execution Mode=Eager, Backend=Inductor2026.05 | 73.8 | — | — | |
| DOPA*Backbone=LLaMA3.1-8B2026.04 | 72.16 | — | — | |
| DOPA*Backbone=Qwen3-1.7B2026.04 | 71.79 | — | — | |
| InfICLBackbone=LLaMA3.1-8B2026.04 | 70.76 | — | — | |
| KNNBackbone=Qwen3-1.7B2026.04 | 70.67 | — | — | |
| KNNBackbone=LLaMA3.1-8B2026.04 | 70.48 | — | — | |
| DrICLBackbone=Qwen3-1.7B2026.04 | 70.38 | — | — | |
| RewriteBackbone=Qwen3-1.7B2026.04 | 70.29 | — | — | |
| DICLBackbone=LLaMA3.1-8B2026.04 | 70.1 | — | — | |
| InfICLBackbone=Qwen3-1.7B2026.04 | 69.92 | — | — | |
| DICLBackbone=Qwen3-1.7B2026.04 | 69.82 | — | — | |
| DrICLBackbone=LLaMA3.1-8B2026.04 | 69.73 | — | — | |
| RandomBackbone=LLaMA3.1-8B2026.04 | 69.63 | — | — | |
| RandomBackbone=Qwen3-1.7B2026.04 | 69.17 | — | — | |
| RewriteBackbone=LLaMA3.1-8B2026.04 | 69.07 | — | — | |
| DOPA*Backbone=LLaMA3.2-3B2026.04 | 68.88 | — | — | |
| DICLBackbone=LLaMA3.2-3B2026.04 | 68.42 | — | — | |
| DOPA*Backbone=Gemma2-2B2026.04 | 68.13 | — | — | |
| DICLBackbone=Gemma2-2B2026.04 | 68.1 | — | — | |
| CLIPPre-training Data=WIT-400M, Backbone=ViT-L/14, Evaluation Protocol=Zero-shot classification2024.07 | 67.3 | — | — | |
| DrICLBackbone=LLaMA3.2-3B2026.04 | 67.1 | — | — | |
| DrICLBackbone=Gemma2-2B2026.04 | 67.1 | — | — | |
| RewriteBackbone=Gemma2-2B2026.04 | 67.01 | — | — | |
| RandomBackbone=Gemma2-2B2026.04 | 66.45 | — | — | |
| RandomBackbone=LLaMA3.2-3B2026.04 | 66.26 | — | — | |
| KNNBackbone=Gemma2-2B2026.04 | 66.26 | — | — | |
| KNNBackbone=LLaMA3.2-3B2026.04 | 65.42 | — | — | |
| InfICLBackbone=Gemma2-2B2026.04 | 64.57 | — | — | |
| InfICLBackbone=LLaMA3.2-3B2026.04 | 64.39 | — | — | |
| RewriteBackbone=LLaMA3.2-3B2026.04 | 64.29 | — | — | |
| CLIP (reported)Pre-training Data=WIT-400M, Backbone=ViT-L/14, Evaluation Protocol=Zero-shot classification2024.07 | 64 | — | — | |
| DOPA*Backbone=GPT2-xl2026.04 | 59.61 | — | — | |
| FLIPPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Zero-shot classification2024.07 | 58.5 | — | — | |
| MLCDPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Zero-shot classification2024.07 | 57.9 | — | — | |
| OpenCLIPPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Zero-shot classification2024.07 | 57.1 | — | — | |
| DICLBackbone=GPT2-xl2026.04 | 56.92 | — | — | |
| LLaMA# of params=7B2023.04 | 53 | — | — | |
| DrICLBackbone=GPT2-xl2026.04 | 52.76 | — | — | |
| KNNBackbone=GPT2-xl2026.04 | 51.17 | — | — | |
| RewriteBackbone=GPT2-xl2026.04 | 50.61 | — | — | |
| GPT-J# of params=6B2023.04 | 49.3 | — | — | |
| RandomBackbone=GPT2-xl2026.04 | 47.7 | — | — | |
| InfICLBackbone=GPT2-xl2026.04 | 46.95 | — | — |