Qwen2.5-Coder-32B-Instruct · bartowski

Qwen2.5-Coder-32B-Instruct IQ4_XS

This file is exactly 17,693,154,272 bytes — 16.48 GiB / 17.69 GB at an effective 4.320 bits per weight. The nominal rate for IQ4_XS is lower; mixed-precision tensors make the real figure higher, always.

From the file· 1 file(s), summedFrom the file· 771 tensors parsed

Get it

16.48 GiB · 1 file
llama.cpp
llama-cli -hf bartowski/Qwen2.5-Coder-32B-Instruct-GGUF:IQ4_XS

Downloads and runs in one step, resolving the quantization by name.

Hugging Face CLI
hf download bartowski/Qwen2.5-Coder-32B-Instruct-GGUF Qwen2.5-Coder-32B-Instruct-IQ4_XS.gguf
Direct download

Straight from the Hugging Face CDN — we host nothing and earn nothing from this. Verify what you received against the exact byte count above; a size mismatch is the usual cause of a file that will not load.

Repository →
Size
16.48 GiB
17.69 GB
Effective bpw
4.320
from real bytes ÷ params
Tensors
771
Header
5.98 MB
GGUF metadata

What this quantization actually contains

per-tensor types, parsed from the GGUF header
IQ4_XS
385
F32
321
Q5_K
64
Q6_K
1

A quantization label names a mixture, not a uniform precision. Attention and output tensors are routinely kept at higher precision than the label implies, which is exactly why the effective bits-per-weight above exceeds the nominal rate.

Fit by accelerator

AcceleratorMemory○Total @4K◐Total @32K◐Fits◐tok/s @4K◐
Arc A310 4GB4 GB18.38 GiB25.38 GiBno—
Arc A350M 4GB4 GB18.38 GiB25.38 GiBno—
Arc A370M 4GB4 GB18.38 GiB25.38 GiBno—
Arc A530M 4GB4 GB18.38 GiB25.38 GiBno—
Arc Pro A30M 4GB4 GB18.38 GiB25.38 GiBno—
Radeon Pro W64004 GB18.48 GiB25.48 GiBno—
Radeon RX 64004 GB18.48 GiB25.48 GiBno—
Radeon RX 6500 XT4 GB18.48 GiB25.48 GiBno—
RTX A4004 GB18.58 GiB25.58 GiBno—
Arc A380 6GB6 GB18.38 GiB25.38 GiBno—
Arc Pro A40 6GB6 GB18.38 GiB25.38 GiBno—
Arc Pro A50 6GB6 GB18.38 GiB25.38 GiBno—
GeForce RTX 20606 GB18.38 GiB25.38 GiBno—
GeForce RTX 30506 GB18.38 GiB25.38 GiBno—
GeForce RTX 3060 OEM6 GB18.38 GiB25.38 GiBno—
RTX A20006 GB18.58 GiB25.58 GiBno—
Apple M18 GB18.13 GiB25.13 GiBno—
Apple M28 GB18.13 GiB25.13 GiBno—
Apple M38 GB18.13 GiB25.13 GiBno—
Apple M48 GB18.13 GiB25.13 GiBno—
Arc A530M 8GB8 GB18.38 GiB25.38 GiBno—
Arc A550M 8GB8 GB18.38 GiB25.38 GiBno—
Arc A570M 8GB8 GB18.38 GiB25.38 GiBno—
Arc A580 8GB8 GB18.38 GiB25.38 GiBno—
○Spec sheet◐Predictedwhat these mean

Memory at context

ContextWeights●KV (f16)●KV (q8_0)●Working buffer◐Total (f16)◐
4,09616.48 GiB1.00 GiB0.53 GiB0.40 GiB17.88 GiB
8,19216.48 GiB2.00 GiB1.06 GiB0.40 GiB18.88 GiB
16,38416.48 GiB4.00 GiB2.13 GiB0.40 GiB20.88 GiB
32,76816.48 GiB8.00 GiB4.25 GiB0.40 GiB24.88 GiB
65,53616.48 GiB16.00 GiB8.50 GiB0.40 GiB32.88 GiB
131,07216.48 GiB32.00 GiB17.00 GiB0.40 GiB48.88 GiB
●From the file◐Predictedwhat these mean

Quantizing the KV cache to q8_0 is a roughly 2× lever on the dominant term at long context, and it is the single most useful setting most local users never touch. Totals here exclude the allocator reserve your driver takes, which is hardware-specific — the per-accelerator table above includes it.