Math on GSM8K (test)
94.04Mean@4TextGrad-M
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| TextGrad-MOptimization Strategy=Promptwise, LMforward=GPT-4o-mini, LMbackward=GPT-4o2025.05 | 94.04 | — | 74.12 | 1.73 | |
| TextGrad-MOptimization Strategy=Blockwise, LMforward=GPT-4o-mini, LMbackward=GPT-4o2025.05 | 93.98 | — | 74.59 | 2.2 | |
| TextGradValidation Protocol=w/ val revert, LMforward=GPT-4o-mini, LMbackward=GPT-4o2025.05 | 93.74 | — | 72.39 | — | |
| TextGradValidation Protocol=w/o val revert, LMforward=GPT-4o-mini, LMbackward=GPT-4o2025.05 | 93.15 | — | 72.96 | — | |
| GEPALMforward=GPT-4o-mini, LMbackward=GPT-4o2025.05 | 91.55 | — | 72.11 | — | |
| ReSyn#Params=7B, zero-shot=true, temperature=0.8, top-p=0.952026.02 | 91.4 | — | — | — | |
| AdalFlow-MOptimization Strategy=Promptwise, LMforward=GPT-4o-mini, LMbackward=GPT-4o2025.05 | 90.78 | — | 72.63 | 0.8 | |
| AdalFlowLMforward=GPT-4o-mini, LMbackward=GPT-4o2025.05 | 89 | — | 71.83 | — | |
| COPRO-MOptimization Strategy=Promptwise, LMforward=GPT-4o-mini, LMbackward=GPT-4o2025.05 | 88.74 | — | 68.79 | 0.94 | |
| COPROLMforward=GPT-4o-mini, LMbackward=GPT-4o2025.05 | 87.99 | — | 67.84 | — | |
| COPRO-MOptimization Strategy=Blockwise, LMforward=GPT-4o-mini, LMbackward=GPT-4o2025.05 | 86.9 | — | 68.9 | 1.05 | |
| Qwen-2.5-Instruct#Params=7B, zero-shot=true, temperature=0.8, top-p=0.952026.02 | 82.3 | — | — | — | |
| Llama-3.1-Instruct#Params=8B, zero-shot=true, temperature=0.8, top-p=0.952026.02 | 70.7 | — | — | — | |
| MoBERatio=30%, # Params=0.96B2026.03 | 54.3 | 53.8 | — | — | |
| MoBERatio=40%, # Params=0.96B2026.03 | 50.9 | 49 | — | — | |
| MC-SMoERatio=30%, # Params=0.96B2026.03 | 48.4 | 49.3 | — | — | |
| MC-SMoERatio=40%, # Params=0.96B2026.03 | 47.6 | 48.4 | — | — | |
| HC-SMoERatio=40%, # Params=0.96B2026.03 | 43 | 48.9 | — | — | |
| MC-SMoERatio=50%, # Params=0.96B2026.03 | 42.2 | 40.4 | — | — | |
| MoBERatio=50%, # Params=0.96B2026.03 | 39.3 | 38.6 | — | — | |
| LoRARatio=0%, # Params=0.96B2026.03 | 38.9 | 52 | — | — | |
| HC-SMoERatio=30%, # Params=0.96B2026.03 | 36.8 | 49.8 | — | — | |
| HC-SMoERatio=50%, # Params=0.96B2026.03 | 34.3 | 45.8 | — | — | |
| LightMoERatio=40%, # Params=0.96B2026.03 | 33.7 | 50.9 | — | — | |
| LightMoERatio=30%, # Params=0.96B2026.03 | 33.3 | 51.4 | — | — | |
| LightMoERatio=50%, # Params=0.96B2026.03 | 32.8 | 48.4 | — | — | |
| Mistral-Instruct#Params=7B, zero-shot=true, temperature=0.8, top-p=0.952026.02 | 22.3 | — | — | — | |
| OriginalRatio=0%, # Params=-2026.03 | 3 | 26.2 | — | — |