List your company

Downstream Task on 11 Downstream Tasks Aggregate

64.6Average AccuracyLLaMA2-7B
32.67240.96149.2557.539Oct 10, 2023Jan 19, 2024Apr 30, 2024Aug 10, 2024Nov 19, 2024Mar 1, 2025Jun 11, 2025

Evaluation results

MethodLinks
LLaMA2-7B# tokens for training=2T, Training data source=Different from RedPajama2023.1064.6———————————
Sheared-LLaMA-2.7B# tokens for training=50B, Training data source=RedPajama2023.1056.7———————————
Sheared-LLaMA-1.3B# tokens for training=50B, Training data source=RedPajama2023.1051———————————
TinyLlama-1.1BFFN Size=100%2025.0647.18———————————
DIVE 1/8retraining budget=15B tokens, ARC-c shots=25, HellaSwag shots=10, MMLU shots=5, active experts=1/82025.0642.528368.0154.752.5324.422.5537.1719.3559.8820.225.88
DIVE 1/8FFN Size=50% × 12025.0642.17———————————
FLAPFFN Size=50%2025.0641.42———————————
DIVE 2/8retraining budget=15B tokens, ARC-c shots=25, HellaSwag shots=10, MMLU shots=5, active experts=2/82025.0641.1281.564.9654.7847.123.1222.5833.2721.9758.4718.226.33
DIVE 2/8FFN Size=25% × 22025.0640.92———————————
LLaMA-MoE 2/8retraining budget=15B tokens, ARC-c shots=25, HellaSwag shots=10, MMLU shots=5, active experts=2/82025.0639.7777.962.8453.5142.8922.0123.0832.2321.5161.7714.625.14
LLaMA-MoE 1/8retraining budget=15B tokens, ARC-c shots=25, HellaSwag shots=10, MMLU shots=5, active experts=1/82025.0639.6577.461.1552.8143.5622.121.0733.7320.5861.7116.625.47
LLM-PrunerFFN Size=50%2025.0639.57———————————
LLaMA-MoE 1/8FFN Size=50% × 12025.0639.34———————————
LLaMA-MoE 2/8FFN Size=25% × 22025.0639.24———————————
Coupled AdamDataset Size (D)=20B, Model Size (N)=760M, Evaluation Protocol=Zero-shot2025.0239.2———————————
Standard AdamDataset Size (D)=20B, Model Size (N)=760M, Evaluation Protocol=Zero-shot2025.0238.5———————————
Standard AdamDataset Size (D)=10B, Model Size (N)=760M, Evaluation Protocol=Zero-shot2025.0237.5———————————
Coupled AdamDataset Size (D)=10B, Model Size (N)=760M, Evaluation Protocol=Zero-shot2025.0237.2———————————
Coupled AdamDataset Size (D)=20B, Model Size (N)=355M, Evaluation Protocol=Zero-shot2025.0237.2———————————
Standard AdamDataset Size (D)=20B, Model Size (N)=355M, Evaluation Protocol=Zero-shot2025.0236.6———————————
Coupled AdamDataset Size (D)=10B, Model Size (N)=355M, Evaluation Protocol=Zero-shot2025.0236.5———————————
Standard AdamDataset Size (D)=5B, Model Size (N)=760M, Evaluation Protocol=Zero-shot2025.0236———————————
Standard AdamDataset Size (D)=10B, Model Size (N)=355M, Evaluation Protocol=Zero-shot2025.0235.9———————————
Coupled AdamDataset Size (D)=5B, Model Size (N)=760M, Evaluation Protocol=Zero-shot2025.0235.7———————————
Standard AdamDataset Size (D)=5B, Model Size (N)=355M, Evaluation Protocol=Zero-shot2025.0235.2———————————
Coupled AdamDataset Size (D)=5B, Model Size (N)=355M, Evaluation Protocol=Zero-shot2025.0235———————————
Coupled AdamDataset Size (D)=20B, Model Size (N)=125M, Evaluation Protocol=Zero-shot2025.0235———————————
Standard AdamDataset Size (D)=20B, Model Size (N)=125M, Evaluation Protocol=Zero-shot2025.0234.6———————————
Standard AdamDataset Size (D)=10B, Model Size (N)=125M, Evaluation Protocol=Zero-shot2025.0234.3———————————
Coupled AdamDataset Size (D)=10B, Model Size (N)=125M, Evaluation Protocol=Zero-shot2025.0234.3———————————
Standard AdamDataset Size (D)=5B, Model Size (N)=125M, Evaluation Protocol=Zero-shot2025.0234———————————
Coupled AdamDataset Size (D)=5B, Model Size (N)=125M, Evaluation Protocol=Zero-shot2025.0233.9———————————