List your company

Mathematical Reasoning on GSM8K (Overall, Prune)

22.079Overall AccuracySLM
3.940368.6494313.358518.06757Feb 24, 2025

Evaluation results

MethodLinks
SLMBackbone=Llama3-70B-Instruct, ratio=0.52025.0222.07915
H2OBackbone=Llama3-70B-Instruct, ratio=0.52025.0219.06211.4
Llama3 (Full)Backbone=Llama3-70B-Instruct2025.0217.504—
SnapKVBackbone=Llama3-70B-Instruct, ratio=0.52025.0216.51711.7
ReFreeKVBackbone=Llama3-70B-Instruct, k=12025.0215.97515.4
SLMBackbone=Llama3-8B-Instruct, ratio=0.52025.026.0710.6
SnapKVBackbone=Llama3-8B-Instruct, ratio=0.52025.025.8742.1
Llama3 (Full)Backbone=Llama3-8B-Instruct2025.024.693—
H2OBackbone=Llama3-8B-Instruct, ratio=0.52025.024.6862
ReFreeKVBackbone=Llama3-8B-Instruct, k=12025.024.6383.4