GLM-4.7-Flash-heretic · DavidAU

GLM-4.7-Flash-heretic Q4_K_M

This file is exactly 25,837,564,960 bytes — 24.06 GiB / 25.84 GB at an effective 6.903 bits per weight. The nominal rate for Q4_K_M is lower; mixed-precision tensors make the real figure higher, always.

From the file· 1 file(s), summedFrom the file· 1186 tensors parsed

Get it

24.06 GiB · 1 file
llama.cpp
llama-cli -hf DavidAU/GLM-4.7-Flash-Grande-Heretic-UNCENSORED-42B-A3B-GGUF:Q4_K_M

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

Hugging Face CLI
hf download DavidAU/GLM-4.7-Flash-Grande-Heretic-UNCENSORED-42B-A3B-GGUF GLM-4.7-30B-A3B-20-2-Heretic-30B-A3B-Q4_K_M.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
24.06 GiB
25.84 GB
Effective bpw
6.903
from real bytes ÷ params
Tensors
1186
Header
9.50 MB
GGUF metadata

What this quantization actually contains

per-tensor types, parsed from the GGUF header
Q4_K
659
F32
395
Q5_0
66
Q6_K
65
BF16
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 GB25.08 GiB26.52 GiBno—
Arc A350M 4GB4 GB25.08 GiB26.52 GiBno—
Arc A370M 4GB4 GB25.08 GiB26.52 GiBno—
Arc A530M 4GB4 GB25.08 GiB26.52 GiBno—
Arc Pro A30M 4GB4 GB25.08 GiB26.52 GiBno—
Radeon Pro W64004 GB25.18 GiB26.62 GiBno—
Radeon RX 64004 GB25.18 GiB26.62 GiBno—
Radeon RX 6500 XT4 GB25.18 GiB26.62 GiBno—
RTX A4004 GB25.28 GiB26.72 GiBno—
Arc A380 6GB6 GB25.08 GiB26.52 GiBno—
Arc Pro A40 6GB6 GB25.08 GiB26.52 GiBno—
Arc Pro A50 6GB6 GB25.08 GiB26.52 GiBno—
GeForce RTX 20606 GB25.08 GiB26.52 GiBno—
GeForce RTX 30506 GB25.08 GiB26.52 GiBno—
GeForce RTX 3060 OEM6 GB25.08 GiB26.52 GiBno—
RTX A20006 GB25.28 GiB26.72 GiBno—
Apple M18 GB24.83 GiB26.27 GiBno—
Apple M28 GB24.83 GiB26.27 GiBno—
Apple M38 GB24.83 GiB26.27 GiBno—
Apple M48 GB24.83 GiB26.27 GiBno—
Arc A530M 8GB8 GB25.08 GiB26.52 GiBno—
Arc A550M 8GB8 GB25.08 GiB26.52 GiBno—
Arc A570M 8GB8 GB25.08 GiB26.52 GiBno—
Arc A580 8GB8 GB25.08 GiB26.52 GiBno—
○Spec sheet◐Predictedwhat these mean

Memory at context

ContextWeights●KV (f16)●KV (q8_0)●Working buffer◐Total (f16)◐
4,09624.06 GiB0.21 GiB0.11 GiB0.31 GiB24.58 GiB
8,19224.06 GiB0.41 GiB0.22 GiB0.31 GiB24.79 GiB
16,38424.06 GiB0.83 GiB0.44 GiB0.31 GiB25.20 GiB
32,76824.06 GiB1.65 GiB0.88 GiB0.31 GiB26.02 GiB
65,53624.06 GiB3.30 GiB1.76 GiB0.31 GiB27.68 GiB
131,07224.06 GiB6.61 GiB3.51 GiB0.31 GiB30.98 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.