Text Generation on Aggregate NLP Tasks (GEC, Smart Reply, Summarization, Tone Adjustment, QA) (test)
32.9Average ScoreSeparate single-task LoRAs
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Separate single-task LoRAsModel=Llama 3.2 3B2026.01 | 32.9 | 100 | |
| Separate single-task LoRAsModel=Qwen 2.5 1.5B2026.01 | 31.7 | 100 | |
| Separate single-task LoRAsModel=StableLM 2 1.6B2026.01 | 29 | 100 | |
| D2CModel=Llama 3.2 3B2026.01 | 26 | 12.5 | |
| D2CModel=Qwen 2.5 1.5B2026.01 | 24.6 | 12.5 | |
| Random clusteringModel=Qwen 2.5 1.5B2026.01 | 22.5 | 12.5 | |
| K-Means w/ SVD clusteringModel=Qwen 2.5 1.5B2026.01 | 22.4 | 12.5 | |
| K-Means w/ SVD clusteringModel=Llama 3.2 3B2026.01 | 22.3 | 12.5 | |
| K-Means clusteringModel=Qwen 2.5 1.5B2026.01 | 21.9 | 12.5 | |
| Random clusteringModel=Llama 3.2 3B2026.01 | 21.5 | 12.5 | |
| K-Means clusteringModel=Llama 3.2 3B2026.01 | 20.4 | 12.5 | |
| D2CModel=StableLM 2 1.6B2026.01 | 17.3 | 12.5 | |
| K-Means w/ SVD clusteringModel=StableLM 2 1.6B2026.01 | 16.1 | 12.5 | |
| K-Means clusteringModel=StableLM 2 1.6B2026.01 | 15.5 | 12.5 | |
| Zero-shotModel=Qwen 2.5 1.5B2026.01 | 15.2 | 0 | |
| Random clusteringModel=StableLM 2 1.6B2026.01 | 15.2 | 12.5 | |
| Zero-shotModel=Llama 3.2 3B2026.01 | 14.7 | 0 | |
| Zero-shotModel=StableLM 2 1.6B2026.01 | 6.5 | 0 |