Qwen-72B vs MiniCPM5-2B

Try Try

Qwen-72B has 72.3B parameters (71.0B active per token) and takes 145 GB on disk; MiniCPM5-2B has 2.52B (2.25B active) and takes 5.03 GB. Compare them layer by layer.

At a glance

Qwen/Qwen-72Bopenbmb/MiniCPM5-2B
Repo createdNov 26, 2023updated Oct 9, 2024Sep 6, 2026updated Sep 12, 2026
Model typeDenseFull attention (MHA)DenseFull attention (GQA)
InputsTextText
Total parameters72.3B29× more2.52B29× less
Active per token71.0B32× more98% of the model2.25B32× less89% of the model
ExpertsNone (dense)None (dense)
Max context (from config)32K tokens4.0× less128K tokens4.0× more
Layers801.9× more421.9× less
On disk145 GB29× more82 files5.03 GB29× less1 file
PrecisionBF16 (100%)BF16 (100%)
QuantizationNoneoriginal precision (BF16)Noneoriginal precision (BF16)
Fewest GPUs1× NVIDIA B200a single 180 GB card · weights only1× RTX 4060a single 8 GB card · weights only
Licensetongyi-qianwen-license-agreement (custom)apache-2.0
GitHubQwenLM/QwenOpenBMB/MiniCPM