Tsinghua/Zhipu
CogVLM2 19B (19B parameters) requires approximately 15.5 GB of VRAM with Q4_K_M quantization. For the best balance of quality and speed, we recommend hardware with at least 18 GB of VRAM.
Get started
— copy & paste to run locallyCopy-paste commands to run CogVLM2 19B on your machine.
Run
docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
--hf-repo "THUDM/cogvlm2-llama3-chat-19B" \
--hf-file "cogvlm2-llama3-chat-19B-Q4_K_M.gguf" \
-c 4096 -ngl 99Quick specs
About this model
Quick picks
Best hardware
Run this model
Quantization options
No hardware detected — fit column shows raw VRAM estimates
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 7.4 GB | Low | — |
Q3_K_S | 3 | 9.3 GB | Low | — |
NVFP4 | 4 | 10.6 GB | Medium | — |
Q4_K_M | 4 | 11.6 GB | Medium | — |
Q5_K_M | 5 | 13.7 GB | High | — |
Q6_K | 6 | 15.6 GB | High | — |
Q8_0 | 8 | 20.3 GB | Very High | — |
F16 | 16 | 38.9 GB | Maximum | — |
Hardware compatibility
Computing compatibility...
Memory breakdown
Frequently asked questions
CogVLM2 19B (19B parameters) requires approximately 15.5 GB of VRAM with Q4_K_M quantization. Lower quantizations like Q4_K_M use less memory but may reduce quality.
Yes, Intel Arc Pro B60 24GB can run CogVLM2 19B with a compatibility score of 86/100. It provides 24 GB of memory and achieves approximately 22.8 tokens per second.
The recommended quantization for CogVLM2 19B is Q4_K_M, which offers the best balance between model quality and memory efficiency. Higher quantizations preserve more quality but require more VRAM.
The top recommended hardware for CogVLM2 19B: RTX 5090 Laptop 24GB (score: 89/100), NVIDIA A30 24GB (score: 89/100), RX 7900 XTX 24GB (score: 89/100). These provide the best combination of memory, bandwidth, and compute for running this model locally.
Yes, CogVLM2 19B is well-suited for chat as well as vision. It was designed with these use cases in mind.
See also
