Skip to content

InternLM 2.5 20B

by Shanghai AI Lab · internlm-2.5 family

20B

parameters

text-generation code-generation reasoning multilingual tool-use math creative-writing

InternLM 2.5 20B is the larger variant in Shanghai AI Lab's InternLM 2.5 series, delivering substantially improved reasoning and generation quality over the 7B model. It supports an extraordinary 1M token context length and offers strong performance across mathematical reasoning, code generation, multilingual tasks, and creative writing. With its Apache 2.0 license and robust tool-use capabilities, the 20B model is well-suited for complex agentic applications and professional workflows. It provides a compelling balance of capability and efficiency for users with mid-range to high-end hardware.

Quick Start with Ollama

ollama run 20b-q4_K_M
Resources Ollama Hugging Face Official Page
Creator Shanghai AI Lab
Parameters 20B
Architecture transformer-decoder
Context 1024K tokens
Released Jul 3, 2024
License Apache 2.0
Ollama internlm2:20b

Quantization Options

Format File Size VRAM Required Quality Ollama Tag
Q4_K_M rec 10.5 GB 12 GB 20b-q4_K_M
Q8_0 18.5 GB 22 GB 20b-q8_0
F16 37 GB 42 GB 20b-f16

Compatible Hardware

Q4_K_M requires 12 GB VRAM

Compatible Hardware

HardwareVRAMTypeFitEst. Speed
Mac Studio M4 Ultra 512GB512 GBmacRuns~68 tok/s
Mac Pro M2 Ultra 192GB192 GBmacRuns~67 tok/s
Mac Studio M4 Ultra 192GB192 GBmacRuns~68 tok/s
Mac Studio M4 Max 128GB128 GBmacRuns~46 tok/s
MacBook Pro M4 Max 128GB128 GBmacRuns~46 tok/s
MacBook Pro M5 Max 128GB128 GBmacRuns~46 tok/s
NVIDIA RTX PRO 6000 Blackwell96 GBgpuRuns~160 tok/s
MacBook Pro M3 Max 96GB96 GBmacRuns~33 tok/s
Mac mini M4 Pro 64GB64 GBmacRuns~23 tok/s
Mac Studio M4 Max 64GB64 GBmacRuns~46 tok/s
MacBook Pro M4 Max 64GB64 GBmacRuns~46 tok/s
MacBook Pro M5 Max 64GB64 GBmacRuns~46 tok/s
NVIDIA RTX 6000 Ada Generation48 GBgpuRuns~80 tok/s
NVIDIA RTX A600048 GBgpuRuns~64 tok/s
NVIDIA RTX PRO 5000 Blackwell48 GBgpuRuns~80 tok/s
Mac mini M4 Pro 48GB48 GBmacRuns~23 tok/s
MacBook Pro M3 Max 48GB48 GBmacRuns~33 tok/s
MacBook Pro M4 Max 48GB48 GBmacRuns~46 tok/s
MacBook Pro M4 Pro 48GB48 GBmacRuns~23 tok/s
MacBook Pro M5 Max 48GB48 GBmacRuns~34 tok/s
MacBook Pro M5 Pro 48GB48 GBmacRuns~23 tok/s
Mac Studio M4 Max 36GB36 GBmacRuns~46 tok/s
MacBook Pro M3 Pro 36GB36 GBmacRuns~13 tok/s
MacBook Pro M5 Max 36GB36 GBmacRuns~34 tok/s
NVIDIA RTX 5000 Ada Generation32 GBgpuRuns~60 tok/s
NVIDIA GeForce RTX 509032 GBgpuRuns~149 tok/s
iMac M4 32GB32 GBmacRuns~10 tok/s
Mac mini M4 32GB32 GBmacRuns~10 tok/s
MacBook Air M5 32GB32 GBmacRuns~10 tok/s
MacBook Air M4 32GB32 GBmacRuns~10 tok/s
MacBook Pro M5 32GB32 GBmacRuns~10 tok/s
AMD Radeon RX 7900 XTX24 GBgpuRuns~80 tok/s
NVIDIA GeForce RTX 3090 Ti24 GBgpuRuns~84 tok/s
NVIDIA GeForce RTX 309024 GBgpuRuns~78 tok/s
NVIDIA GeForce RTX 409024 GBgpuRuns~84 tok/s
NVIDIA RTX A500024 GBgpuRuns~64 tok/s
iMac M3 24GB24 GBmacRuns~8 tok/s
Mac mini M2 24GB24 GBmacRuns~8 tok/s
Mac mini M4 Pro 24GB24 GBmacRuns~23 tok/s
MacBook Air M2 24GB24 GBmacRuns~8 tok/s
MacBook Air M4 24GB24 GBmacRuns~10 tok/s
MacBook Air M5 24GB24 GBmacRuns~10 tok/s
MacBook Pro M4 Pro 24GB24 GBmacRuns~23 tok/s
MacBook Pro M5 24GB24 GBmacRuns~10 tok/s
MacBook Pro M5 Pro 24GB24 GBmacRuns~23 tok/s
AMD Radeon RX 7900 XT20 GBgpuRuns~67 tok/s
NVIDIA RTX 4000 Ada Generation20 GBgpuRuns~30 tok/s
MacBook Pro M3 Pro 18GB18 GBmacRuns~13 tok/s
AMD Radeon RX 6800 XT16 GBgpuRuns~43 tok/s
AMD Radeon RX 6900 XT16 GBgpuRuns~43 tok/s
AMD Radeon RX 7800 XT16 GBgpuRuns~52 tok/s
AMD Radeon RX 9060 XT 16GB16 GBgpuRuns~45 tok/s
AMD Radeon RX 9070 XT16 GBgpuRuns~54 tok/s
AMD Radeon RX 907016 GBgpuRuns~45 tok/s
Intel Arc A77016 GBgpuRuns~47 tok/s
NVIDIA GeForce RTX 4060 Ti 16GB16 GBgpuRuns~24 tok/s
NVIDIA GeForce RTX 4070 Ti Super16 GBgpuRuns~56 tok/s
NVIDIA GeForce RTX 4080 Super16 GBgpuRuns~61 tok/s
NVIDIA GeForce RTX 408016 GBgpuRuns~60 tok/s
NVIDIA GeForce RTX 5060 Ti 16GB16 GBgpuRuns~37 tok/s
NVIDIA GeForce RTX 5070 Ti16 GBgpuRuns~75 tok/s
NVIDIA GeForce RTX 508016 GBgpuRuns~80 tok/s
NVIDIA RTX A400016 GBgpuRuns~37 tok/s
iMac M1 16GB16 GBmacRuns~6 tok/s
iMac M4 16GB16 GBmacRuns~10 tok/s
Mac mini M1 16GB16 GBmacRuns~6 tok/s
Mac mini M4 16GB16 GBmacRuns~10 tok/s
MacBook Air M2 16GB16 GBmacRuns~8 tok/s
MacBook Air M4 16GB16 GBmacRuns~10 tok/s
MacBook Air M3 16GB16 GBmacRuns~8 tok/s
MacBook Air M5 16GB16 GBmacRuns~10 tok/s
MacBook Pro M2 Pro 16GB16 GBmacRuns~17 tok/s
MacBook Pro M1 16GB16 GBmacRuns~6 tok/s
MacBook Pro M5 16GB16 GBmacRuns~10 tok/s
AMD Radeon RX 6700 XT12 GBgpuCPU Offload~10 tok/s
AMD Radeon RX 7700 XT12 GBgpuCPU Offload~11 tok/s
Intel Arc B58012 GBgpuCPU Offload~11 tok/s
NVIDIA GeForce RTX 3060 12GB12 GBgpuCPU Offload~9 tok/s
NVIDIA GeForce RTX 3080 12GB12 GBgpuCPU Offload~23 tok/s
NVIDIA GeForce RTX 4070 Super12 GBgpuCPU Offload~13 tok/s
NVIDIA GeForce RTX 4070 Ti12 GBgpuCPU Offload~13 tok/s
NVIDIA GeForce RTX 407012 GBgpuCPU Offload~13 tok/s
NVIDIA GeForce RTX 507012 GBgpuCPU Offload~17 tok/s
NVIDIA GeForce GTX 1080 Ti11 GBgpuCPU Offload~12 tok/s
NVIDIA GeForce RTX 2080 Ti11 GBgpuCPU Offload~15 tok/s
Intel Arc B57010 GBgpuCPU Offload~10 tok/s
NVIDIA GeForce RTX 3080 10GB10 GBgpuCPU Offload~19 tok/s
AMD Radeon RX 6600 XT8 GBgpuCPU Offload~6 tok/s
AMD Radeon RX 76008 GBgpuCPU Offload~7 tok/s
AMD Radeon RX 9060 XT 8GB8 GBgpuCPU Offload~7 tok/s
Intel Arc A7508 GBgpuCPU Offload~13 tok/s
NVIDIA GeForce GTX 10708 GBgpuCPU Offload~6 tok/s
NVIDIA GeForce RTX 2060 Super8 GBgpuCPU Offload~11 tok/s
NVIDIA GeForce RTX 2070 Super8 GBgpuCPU Offload~11 tok/s
NVIDIA GeForce RTX 2080 Super8 GBgpuCPU Offload~12 tok/s
NVIDIA GeForce RTX 30508 GBgpuCPU Offload~6 tok/s
NVIDIA GeForce RTX 3060 Ti8 GBgpuCPU Offload~11 tok/s
NVIDIA GeForce RTX 30708 GBgpuCPU Offload~11 tok/s
NVIDIA GeForce RTX 4060 Ti 8GB8 GBgpuCPU Offload~7 tok/s
NVIDIA GeForce RTX 40608 GBgpuCPU Offload~7 tok/s
NVIDIA GeForce RTX 50508 GBgpuCPU Offload~6 tok/s
NVIDIA GeForce RTX 5060 Ti 8GB8 GBgpuCPU Offload~11 tok/s
NVIDIA GeForce RTX 50608 GBgpuCPU Offload~8 tok/s
MacBook Air M1 8GB8 GBmacCPU Offload~2 tok/s
MacBook Air M2 8GB8 GBmacCPU Offload~2 tok/s
2 hardware device(s) cannot run this model at Q4_K_M.

Benchmark Scores

78.0
mmlu