deepseek-coder-6.7b-base · bartowski

deepseek-coder-6.7b-base IQ2_XS

This file is exactly 2,036,410,560 bytes — 1.90 GiB / 2.04 GB at an effective 2.417 bits per weight. The nominal rate for IQ2_XS is lower; mixed-precision tensors make the real figure higher, always.

From the file· 1 file(s), summed

Get it

1.90 GiB · 1 file
llama.cpp
llama-cli -hf bartowski/deepseek-coder-6.7B-kexer-GGUF:IQ2_XS

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

Hugging Face CLI
hf download bartowski/deepseek-coder-6.7B-kexer-GGUF deepseek-coder-6.7B-kexer-IQ2_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
1.90 GiB
2.04 GB
Effective bpw
2.417
from real bytes ÷ params
Tensors
—
Header
—
GGUF metadata

Fit by accelerator

AcceleratorMemory○Total @4K◐Total @32K◐Fits◐tok/s @4K◐
Arc A310 4GB4 GB4.72 GiB18.72 GiBno—
Arc A350M 4GB4 GB4.72 GiB18.72 GiBno—
Arc A370M 4GB4 GB4.72 GiB18.72 GiBno—
Arc A530M 4GB4 GB4.72 GiB18.72 GiBno—
Arc Pro A30M 4GB4 GB4.72 GiB18.72 GiBno—
Radeon Pro W64004 GB4.82 GiB18.82 GiBno—
Radeon RX 64004 GB4.82 GiB18.82 GiBno—
Radeon RX 6500 XT4 GB4.82 GiB18.82 GiBno—
RTX A4004 GB4.92 GiB18.92 GiBno—
Arc A380 6GB6 GB4.72 GiB18.72 GiByes4K only25±30%
Arc Pro A40 6GB6 GB4.72 GiB18.72 GiByes4K only26±30%
Arc Pro A50 6GB6 GB4.72 GiB18.72 GiByes4K only26±30%
GeForce RTX 20606 GB4.72 GiB18.72 GiByes4K only61±12.9%
GeForce RTX 30506 GB4.72 GiB18.72 GiByes4K only31±12.9%
GeForce RTX 3060 OEM6 GB4.72 GiB18.72 GiByes4K only61±12.9%
RTX A20006 GB4.92 GiB18.92 GiByes4K only42±22%
Apple M18 GB4.47 GiB18.47 GiByes4K only14±8.3%
Apple M28 GB4.47 GiB18.47 GiByes4K only20±8.3%
Apple M38 GB4.47 GiB18.47 GiByes4K only20±8.3%
Apple M48 GB4.47 GiB18.47 GiByes4K only23±8.3%
Arc A530M 8GB8 GB4.72 GiB18.72 GiByes4K only30±30%
Arc A550M 8GB8 GB4.72 GiB18.72 GiByes4K only30±30%
Arc A570M 8GB8 GB4.72 GiB18.72 GiByes4K only30±30%
Arc A580 8GB8 GB4.72 GiB18.72 GiByes4K only64±30%
○Spec sheet◐Predictedwhat these mean

Memory at context

ContextWeights●KV (f16)●KV (q8_0)●Working buffer◐Total (f16)◐
4,0961.90 GiB2.00 GiB1.06 GiB0.32 GiB4.22 GiB
8,1921.90 GiB4.00 GiB2.13 GiB0.32 GiB6.22 GiB
16,3841.90 GiB8.00 GiB4.25 GiB0.32 GiB10.22 GiB
32,7681.90 GiB16.00 GiB8.50 GiB0.32 GiB18.22 GiB
65,5361.90 GiB32.00 GiB17.00 GiB0.32 GiB34.22 GiB
131,0721.90 GiB64.00 GiB34.00 GiB0.32 GiB66.22 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.