~/can-i-run/qwen-2-5-coder-32b
Alibaba provider

Can I run Qwen 2.5 Coder 32B?

Short answer: yes, on a H100 80GB (80GB) at FP16/BF16. Long answer below.

qwen2.5-coder:32bLM StudiovLLMMLXoMLX

The math, in one paragraph.

$ ./vrambudget --explain qwen-2-5-coder-32b

Qwen 2.5 Coder 32B has 32.5B parameters. At FP16 that's 65 GB of raw weights. Quantization shrinks that, but you also need budget for the KV cache (definition), framework overhead, and safety headroom. The rule of thumb: real usable budget on a card is roughly its nameplate VRAM minus 25%. That's how the table below was computed.

What hardware actually fits.

$ grep "fits" gpus.json
FP16/BF16
65GB
13 GPUs fit
H100 80GB80GBM3 Max 9696GBRTX Pro 600096GBM4 Max 128128GB+ 9 more
Q8_0
35GB
21 GPUs fit
RTX A600048GBRTX 6000 Ada48GBL40S48GBRadeon Pro W790048GB+ 17 more
Q5_K_M
22GB
22 GPUs fit
RTX 509032GBRTX A600048GBRTX 6000 Ada48GBL40S48GB+ 18 more
Q4_K_M
18GB
27 GPUs fit
RTX 309024GBRTX 3090 Ti24GBRTX 409024GBRX 7900 XTX24GB+ 23 more
Q3_K_M
14GB
27 GPUs fit
RTX 309024GBRTX 3090 Ti24GBRTX 409024GBRX 7900 XTX24GB+ 23 more

Pick your path.

$ ls strategies/
Tightest budget

Smallest GPU that fits Qwen 2.5 Coder 32B at any quant: H100 80GB at FP16/BF16.

Reference quality (FP16)

Lossless inference needs 65 GB. Pick from 13 cards.

Best quality on a 24GB card

Q4_K_M fits comfortably (18 GB weights).

Tune the math yourself

Open the calculator pre-tuned for Qwen 2.5 Coder 32B: ↗ /calc?model=qwen-2-5-coder-32b

See the full model page.

$ ./open
$ openQwen 2.5 Coder 32B32.5BAlibaba · ctx 128K$ rankBest GPU for Qwen 2.5 Coder 32Bsee the ranking →ranked by fit quality

Discussion.

$ gh discussion list

// sign in with github to leave a comment. threads live in the repo's discussions tab.