Language Modeling on Tulu3 tydiqa (Evaluation)
1.764PerplexityAdamW-influence (online)
Evaluation Results
| Method | Links | |
|---|---|---|
| AdamW-influence (online)Backbone=Llama-3.2-1B, Fine-tuning regime=LoRA, Selected Ratio=0.52026.05 | 1.764 | |
| SGD-influence (online)Backbone=Llama-3.2-1B, Fine-tuning regime=LoRA, Selected Ratio=0.52026.05 | 1.81 | |
| AdamW-influence (online)Backbone=Llama-3.2-1B, Fine-tuning regime=LoRA, Selected Ratio=0.82026.05 | 1.859 | |
| AdamW-influence (offline)Backbone=Llama-3.2-1B, Fine-tuning regime=LoRA, Selected Ratio=0.52026.05 | 1.868 | |
| SGD-influence (online)Backbone=Llama-3.2-1B, Fine-tuning regime=LoRA, Selected Ratio=0.82026.05 | 1.872 | |
| AdamW-influence (offline)Backbone=Llama-3.2-1B, Fine-tuning regime=LoRA, Selected Ratio=0.82026.05 | 1.872 | |
| AdamW-influence (online)Backbone=Llama-3.2-1B, Fine-tuning regime=Full, Selected Ratio=0.52026.05 | 1.886 | |
| TracIN (offline)Backbone=Llama-3.2-1B, Fine-tuning regime=LoRA, Selected Ratio=0.82026.05 | 1.91 | |
| AdamW-influence (online)Backbone=Llama-3.2-1B, Fine-tuning regime=Full, Selected Ratio=0.82026.05 | 1.933 | |
| AdamW-influence (offline)Backbone=Llama-3.2-1B, Fine-tuning regime=Full, Selected Ratio=0.82026.05 | 1.939 | |
| TracIN (offline)Backbone=Llama-3.2-1B, Fine-tuning regime=LoRA, Selected Ratio=0.52026.05 | 1.955 | |
| AdamW-influence (offline)Backbone=Llama-3.2-1B, Fine-tuning regime=Full, Selected Ratio=0.52026.05 | 1.957 | |
| SGD-influence (online)Backbone=Llama-3.2-1B, Fine-tuning regime=Full, Selected Ratio=0.52026.05 | 2.029 | |
| SGD-influence (online)Backbone=Llama-3.2-1B, Fine-tuning regime=Full, Selected Ratio=0.82026.05 | 2.055 | |
| TracIN (offline)Backbone=Llama-3.2-1B, Fine-tuning regime=Full, Selected Ratio=0.52026.05 | 2.152 | |
| TracIN (offline)Backbone=Llama-3.2-1B, Fine-tuning regime=Full, Selected Ratio=0.82026.05 | 2.175 |