Alibaba · 15B · Q4_K_M · 8.37 GiB

Qwen2.5 14B Instruct at Q4_K_M needs about 10.9 GiB at 8K context, and the smallest listed machine that fits it is GeForce RTX 4060 Ti 16GB.

Dense Qwen2.5 instruct model. use_sliding_window is false in config.json, so every layer keeps a full KV cache.

ollama run qwen2.5:14b

Weights: Qwen2.5-14B-Instruct-Q4_K_M.gguf. Architecture: config source. Hugging Face repo created 2024-09-16. License field: apache-2.0.

Quants with a measured file size

QuantFileSize
Q4_K_M · recommendedQwen2.5-14B-Instruct-Q4_K_M.gguf8.37 GiB
Q5_K_MQwen2.5-14B-Instruct-Q5_K_M.gguf9.79 GiB
Q6_KQwen2.5-14B-Instruct-Q6_K.gguf11.3 GiB
Q8_0Qwen2.5-14B-Instruct-Q8_0.gguf14.6 GiB

Machines

Same quant and 8,192 tokens on every row.
MachineMemoryTotalVerdictSpeed
Apple M1, 8 GB8 GB unified10.9 GiBNo—
Apple M2, 8 GB8 GB unified10.9 GiBNo—
Apple M3, 8 GB8 GB unified10.9 GiBNo—
GeForce RTX 3060 Ti8 GB GDDR10.9 GiBOffload4–6 tok/s
GeForce RTX 30708 GB GDDR10.9 GiBOffload4–6 tok/s
GeForce RTX 40608 GB GDDR10.9 GiBOffload4–6 tok/s
GeForce RTX 4060 Ti 8GB8 GB GDDR10.9 GiBOffload4–6 tok/s
GeForce RTX 50608 GB GDDR10.9 GiBOffload8–11 tok/s
GeForce RTX 5060 Ti 8GB8 GB GDDR10.9 GiBOffload8–11 tok/s
Radeon RX 66008 GB GDDR610.9 GiBOffload4–5 tok/s
Radeon RX 76008 GB GDDR610.9 GiBOffload4–6 tok/s
GeForce RTX 3080 10GB10 GB GDDR10.00 GiBTight43–68 tok/s
GeForce RTX 3060 12GB12 GB GDDR10.9 GiBTight19–32 tok/s
GeForce RTX 407012 GB GDDR10.9 GiBTight26–45 tok/s
GeForce RTX 4070 Super12 GB GDDR10.9 GiBTight26–45 tok/s
GeForce RTX 4070 Ti12 GB GDDR10.9 GiBTight26–45 tok/s
GeForce RTX 507012 GB GDDR10.9 GiBTight35–60 tok/s
Intel Arc B58012 GB GDDR610.9 GiBTight24–41 tok/s
Radeon RX 6700 XT12 GB GDDR610.9 GiBTight20–34 tok/s
Radeon RX 7700 XT12 GB GDDR610.9 GiBTight22–38 tok/s
Apple M1 Pro, 16 GB16 GB unified10.9 GiBTight10–18 tok/s
Apple M1, 16 GB16 GB unified10.9 GiBTight4–6 tok/s
Apple M2 Pro, 16 GB16 GB unified10.9 GiBTight10–18 tok/s
Apple M2, 16 GB16 GB unified10.9 GiBTight5–9 tok/s
Apple M3, 16 GB16 GB unified10.9 GiBTight5–9 tok/s
Apple M4, 16 GB16 GB unified10.9 GiBTight6–11 tok/s
GeForce RTX 4060 Ti 16GB16 GB GDDR10.9 GiBFits15–26 tok/s
GeForce RTX 4070 Ti Super16 GB GDDR10.9 GiBFits35–60 tok/s
GeForce RTX 408016 GB GDDR10.9 GiBFits37–64 tok/s
GeForce RTX 4080 Super16 GB GDDR10.9 GiBFits38–66 tok/s
GeForce RTX 5060 Ti 16GB16 GB GDDR10.9 GiBFits23–40 tok/s
GeForce RTX 5070 Ti16 GB GDDR10.9 GiBFits46–80 tok/s
GeForce RTX 508016 GB GDDR10.9 GiBFits50–85 tok/s
Intel Arc A770 16GB16 GB GDDR610.9 GiBFits27–46 tok/s
Legion Go16 GB LPDDR5X-750010.9 GiBTight6–11 tok/s
Radeon RX 6800 XT16 GB GDDR610.9 GiBFits27–46 tok/s
Radeon RX 7800 XT16 GB GDDR610.9 GiBFits32–56 tok/s
Radeon RX 9060 XT 16GB16 GB GDDR610.9 GiBFits17–28 tok/s
Radeon RX 9070 XT16 GB GDDR610.9 GiBFits33–57 tok/s
ROG Ally16 GB LPDDR5-640010.9 GiBTight5–9 tok/s
Steam Deck LCD16 GB LPDDR5-550010.9 GiBTight5–8 tok/s
Steam Deck OLED16 GB LPDDR5-640010.9 GiBTight5–9 tok/s
Apple M3 Pro, 18 GB18 GB unified10.9 GiBFits8–13 tok/s
Radeon RX 7900 XT20 GB GDDR610.9 GiBFits42–71 tok/s
Apple M2, 24 GB24 GB unified10.9 GiBFits5–9 tok/s
Apple M3, 24 GB24 GB unified10.9 GiBFits5–9 tok/s
Apple M4 Pro, 24 GB24 GB unified10.9 GiBFits14–24 tok/s
Apple M4, 24 GB24 GB unified10.9 GiBFits6–11 tok/s
GeForce RTX 309024 GB GDDR10.9 GiBFits49–83 tok/s
GeForce RTX 409024 GB GDDR10.9 GiBFits52–90 tok/s
Radeon RX 7900 XTX24 GB GDDR610.9 GiBFits50–85 tok/s
ROG Ally X24 GB LPDDR5X-750010.9 GiBFits6–11 tok/s
Apple M1 Max, 32 GB32 GB unified10.9 GiBFits21–36 tok/s
Apple M1 Pro, 32 GB32 GB unified10.9 GiBFits10–18 tok/s
Apple M2 Max, 32 GB32 GB unified10.9 GiBFits21–36 tok/s
Apple M2 Pro, 32 GB32 GB unified10.9 GiBFits10–18 tok/s
Apple M4, 32 GB32 GB unified10.9 GiBFits6–11 tok/s
GeForce RTX 509032 GB GDDR10.9 GiBFits93–159 tok/s
Apple M3 Pro, 36 GB36 GB unified10.9 GiBFits8–13 tok/s
Apple M4 Max 14-core, 36 GB36 GB unified10.9 GiBFits21–36 tok/s
Apple M3 Max 16-core, 48 GB48 GB unified10.9 GiBFits21–36 tok/s
Apple M4 Max 16-core, 48 GB48 GB unified10.9 GiBFits28–49 tok/s
Apple M4 Pro, 48 GB48 GB unified10.9 GiBFits14–24 tok/s
Apple M1 Max, 64 GB64 GB unified10.9 GiBFits21–36 tok/s
Apple M2 Max, 64 GB64 GB unified10.9 GiBFits21–36 tok/s
Apple M3 Max 16-core, 64 GB64 GB unified10.9 GiBFits21–36 tok/s
Apple M4 Max 16-core, 64 GB64 GB unified10.9 GiBFits28–49 tok/s
Apple M4 Pro, 64 GB64 GB unified10.9 GiBFits14–24 tok/s
Apple M2 Max, 96 GB96 GB unified10.9 GiBFits21–36 tok/s
Apple M3 Max 16-core, 128 GB128 GB unified10.9 GiBFits21–36 tok/s
Apple M4 Max 16-core, 128 GB128 GB unified10.9 GiBFits28–49 tok/s

FAQ

Why Q4_K_M?

Q4_K_M is the quant this page recommends. The file is Qwen2.5-14B-Instruct-Q4_K_M.gguf at 8.37 GiB. Other single-file quants are in the table when a real size was available.

Where do the architecture numbers come from?

Official config.json. head_dim = hidden_size / num_attention_heads. Source: https://huggingface.co/Qwen/Qwen2.5-14B-Instruct/raw/main/config.json

Does this include a long context?

The fit uses 8,192 tokens, or less when that is all the fast memory can hold. The model file allows up to 32,768 tokens, and KV cache grows as you use more of that.

What is the Ollama command?

ollama run qwen2.5:14b. The library tag is Q4-class or the vendor quant. Compare it with the GGUF linked on this page before you treat the sizes as identical.