Qwen3.5-0.8B · bartowski

Qwen3.5-0.8B Q4_0

This file is exactly 537,517,152 bytes — 0.50 GiB / 0.54 GB at an effective 4.923 bits per weight. The nominal rate for Q4_0 is lower; mixed-precision tensors make the real figure higher, always.

From the file· 1 file(s), summedFrom the file· 335 tensors parsed

Get it

0.50 GiB · 1 file
llama.cpp
llama-cli -hf bartowski/Qwen_Qwen3.5-0.8B-GGUF:Q4_0

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

Hugging Face CLI
hf download bartowski/Qwen_Qwen3.5-0.8B-GGUF Qwen_Qwen3.5-0.8B-Q4_0.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
0.50 GiB
0.54 GB
Effective bpw
4.923
from real bytes ÷ params
Tensors
335
Header
10.96 MB
GGUF metadata

What this quantization actually contains

per-tensor types, parsed from the GGUF header
F32
176
Q4_0
136
Q8_0
17
Q4_1
3
Q6_K
3

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 GB1.33 GiB1.65 GiByes101±30%
Arc A350M 4GB4 GB1.33 GiB1.65 GiByes93±30%
Arc A370M 4GB4 GB1.33 GiB1.65 GiByes93±30%
Arc A530M 4GB4 GB1.33 GiB1.65 GiByes157±30%
Arc Pro A30M 4GB4 GB1.33 GiB1.65 GiByes93±30%
Radeon Pro W64004 GB1.43 GiB1.75 GiByes124±26.5%
Radeon RX 64004 GB1.43 GiB1.75 GiByes124±26.5%
Radeon RX 6500 XT4 GB1.43 GiB1.75 GiByes136±26.5%
RTX A4004 GB1.53 GiB1.85 GiByes95±22%
Arc A380 6GB6 GB1.33 GiB1.65 GiByes138±30%
Arc Pro A40 6GB6 GB1.33 GiB1.65 GiByes141±30%
Arc Pro A50 6GB6 GB1.33 GiB1.65 GiByes141±30%
GeForce RTX 20606 GB1.33 GiB1.65 GiByes338±12.9%
GeForce RTX 30506 GB1.33 GiB1.65 GiByes193±12.9%
GeForce RTX 3060 OEM6 GB1.33 GiB1.65 GiByes338±12.9%
RTX A20006 GB1.53 GiB1.85 GiByes257±22%
Apple M18 GB1.08 GiB1.40 GiByes65±8.3%
Apple M28 GB1.08 GiB1.40 GiByes81±8.3%
Apple M38 GB1.08 GiB1.40 GiByes81±8.3%
Apple M48 GB1.08 GiB1.40 GiByes87±8.3%
Arc A530M 8GB8 GB1.33 GiB1.65 GiByes157±30%
Arc A550M 8GB8 GB1.33 GiB1.65 GiByes157±30%
Arc A570M 8GB8 GB1.33 GiB1.65 GiByes157±30%
Arc A580 8GB8 GB1.33 GiB1.65 GiByes256±30%
○Spec sheet◐Predictedwhat these mean

Memory at context

ContextWeights●KV (f16)●KV (q8_0)●Working buffer◐Total (f16)◐
4,0960.50 GiB0.05 GiB0.02 GiB0.28 GiB0.83 GiB
8,1920.50 GiB0.09 GiB0.05 GiB0.28 GiB0.87 GiB
16,3840.50 GiB0.19 GiB0.10 GiB0.28 GiB0.97 GiB
32,7680.50 GiB0.38 GiB0.20 GiB0.28 GiB1.15 GiB
65,5360.50 GiB0.75 GiB0.40 GiB0.28 GiB1.53 GiB
131,0720.50 GiB1.50 GiB0.80 GiB0.28 GiB2.28 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.