Phi-4-mini-instruct · unsloth

Phi-4-mini-instruct Q2_K_L

This file is exactly 1,682,635,744 bytes — 1.57 GiB / 1.68 GB at an effective 3.509 bits per weight. The nominal rate for Q2_K_L is lower; mixed-precision tensors make the real figure higher, always.

From the file· 1 file(s), summed

Get it

1.57 GiB · 1 file
llama.cpp
llama-cli -hf unsloth/Phi-4-mini-instruct-GGUF:Q2_K_L

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

Hugging Face CLI
hf download unsloth/Phi-4-mini-instruct-GGUF Phi-4-mini-instruct-Q2_K_L.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
1.57 GiB
1.68 GB
Effective bpw
3.509
from real bytes ÷ params
Tensors
—
Header
—
GGUF metadata

Fit by accelerator

AcceleratorMemory○Total @4K◐Total @32K◐Fits◐tok/s @4K◐
Arc A310 4GB4 GB2.88 GiB6.38 GiByes4K only31±30%
Arc A350M 4GB4 GB2.88 GiB6.38 GiByes4K only29±30%
Arc A370M 4GB4 GB2.88 GiB6.38 GiByes4K only29±30%
Arc A530M 4GB4 GB2.88 GiB6.38 GiByes4K only54±30%
Arc Pro A30M 4GB4 GB2.88 GiB6.38 GiByes4K only29±30%
Radeon Pro W64004 GB2.98 GiB6.48 GiByes4K only38±26.5%
Radeon RX 64004 GB2.98 GiB6.48 GiByes4K only38±26.5%
Radeon RX 6500 XT4 GB2.98 GiB6.48 GiByes4K only43±26.5%
RTX A4004 GB3.08 GiB6.58 GiByes4K only26±22%
Arc A380 6GB6 GB2.88 GiB6.38 GiByes4K only46±30%
Arc Pro A40 6GB6 GB2.88 GiB6.38 GiByes4K only47±30%
Arc Pro A50 6GB6 GB2.88 GiB6.38 GiByes4K only47±30%
GeForce RTX 20606 GB2.88 GiB6.38 GiByes4K only110±12.9%
GeForce RTX 30506 GB2.88 GiB6.38 GiByes4K only57±12.9%
GeForce RTX 3060 OEM6 GB2.88 GiB6.38 GiByes4K only110±12.9%
RTX A20006 GB3.08 GiB6.58 GiByes4K only77±22%
Apple M18 GB2.63 GiB6.13 GiByes4K only24±8.3%
Apple M28 GB2.63 GiB6.13 GiByes4K only34±8.3%
Apple M38 GB2.63 GiB6.13 GiByes4K only34±8.3%
Apple M48 GB2.63 GiB6.13 GiByes4K only38±8.3%
Arc A530M 8GB8 GB2.88 GiB6.38 GiByes54±30%
Arc A550M 8GB8 GB2.88 GiB6.38 GiByes54±30%
Arc A570M 8GB8 GB2.88 GiB6.38 GiByes54±30%
Arc A580 8GB8 GB2.88 GiB6.38 GiByes108±30%
○Spec sheet◐Predictedwhat these mean

Memory at context

ContextWeights●KV (f16)●KV (q8_0)●Working buffer◐Total (f16)◐
4,0961.57 GiB0.50 GiB0.27 GiB0.31 GiB2.38 GiB
8,1921.57 GiB1.00 GiB0.53 GiB0.31 GiB2.88 GiB
16,3841.57 GiB2.00 GiB1.06 GiB0.31 GiB3.88 GiB
32,7681.57 GiB4.00 GiB2.13 GiB0.31 GiB5.88 GiB
65,5361.57 GiB8.00 GiB4.25 GiB0.31 GiB9.88 GiB
131,0721.57 GiB16.00 GiB8.50 GiB0.31 GiB17.88 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.