Can I run Qwen2.5-Coder-7B-Instruct on a Mac with 16 GB?

Yes, with a GGUF version

The published weights are too big, but the Q8_0 file from Qwen/Qwen2.5-Coder-7B-Instruct-GGUF (8.10 GB) fits, with room for ≈ 44K tokens of context.

Full analysis of Qwen2.5-Coder-7B-Instruct Everything that runs on Mac with 16 GB Get the Q8_0 file on Hugging Face

llama-server -hf Qwen/Qwen2.5-Coder-7B-Instruct-GGUF:Q8_0

Files and context on Mac with 16 GB

Each file of Qwen2.5-Coder-7B-Instruct (and of GGUF versions the site has analyzed), and how much context fits next to it, with the context cache in 16-bit or 8-bit.

File	Size	On Mac with 16 GB	Context (16-bit)	Context (8-bit)
As published	15.2 GB	Doesn't fit	—	—
qwen2.5-coder-7b-instruct-fp16Qwen/Qwen2.5-Coder-7B-Instruct-GGUF	30.5 GB	Doesn't fit	—	—
Q8_0Qwen/Qwen2.5-Coder-7B-Instruct-GGUF	8.10 GB	Fits	44K	87K
qwen2.5-coder-7b-instruct-q8_0Qwen/Qwen2.5-Coder-7B-Instruct-GGUF	8.10 GB	Fits	44K	87K
Q6_KQwen/Qwen2.5-Coder-7B-Instruct-GGUF	6.25 GB	Fits	75K	128K
qwen2.5-coder-7b-instruct-q6_kQwen/Qwen2.5-Coder-7B-Instruct-GGUF	6.25 GB	Fits	75K	128K
Q5_K_MQwen/Qwen2.5-Coder-7B-Instruct-GGUF	5.44 GB	Fits	89K	128K
qwen2.5-coder-7b-instruct-q5_k_mQwen/Qwen2.5-Coder-7B-Instruct-GGUF	5.44 GB	Fits	89K	128K
Q5_0Qwen/Qwen2.5-Coder-7B-Instruct-GGUF	5.32 GB	Fits	91K	128K
Q4_K_MQwen/Qwen2.5-Coder-7B-Instruct-GGUF	4.68 GB	Fits	102K	128K
qwen2.5-coder-7b-instruct-q4_k_mQwen/Qwen2.5-Coder-7B-Instruct-GGUF	4.68 GB	Fits	102K	128K
Q4_0Qwen/Qwen2.5-Coder-7B-Instruct-GGUF	4.43 GB	Fits	106K	128K
qwen2.5-coder-7b-instruct-q4_0Qwen/Qwen2.5-Coder-7B-Instruct-GGUF	4.43 GB	Fits	106K	128K
Q3_K_MQwen/Qwen2.5-Coder-7B-Instruct-GGUF	3.81 GB	Fits	117K	128K
Q2_KQwen/Qwen2.5-Coder-7B-Instruct-GGUF	3.02 GB	Fits	128K	128K

Qwen2.5-Coder-7B-Instruct on other hardware

Popular models that run on Mac with 16 GB

Assumes 67% of the Mac's memory is usable (macOS's default limit for the GPU; it can be raised) and at least 4K tokens of context for one conversation, with the context cache in 16-bit unless noted. Fits leaves at least 20% of that memory for context; Tight leaves less.