Microsoft

Phi-4 14B

14,7B

Parameters

8,4 Go

Q4_K_M file

≥ 9,9 Go

Recommended VRAM

16 384

Max context

Standard llama-bench run (prompt 512 / generation 128 tokens), model fully in VRAM. Click a card for details. Electricity cost at the default price (0,25 €/kWh).

Card · Q4_K_MVRAMGenerationPromptVRAM usedPower€ elec. / M tokPrice
RTX 5070 Ti16 Go84,9 tok/s3 99410,1 Go182 W0,15 €/M tok—

Technical sheet

Published
2024-12-11
Architecture
40 layers · 10 KV heads · dimension 128
Attention
Standard: every layer keeps the whole context in memory.
Context cache (f16)
≈ 195 MB per 1,000 tokens · 6,3 Go for 32,768 tokens
GGUF file
bartowski/phi-4-GGUF · phi-4-Q4_K_M.gguf Download (8,4 Go)
Official repo
microsoft/phi-4

Public Hugging Face data (official config.json, GGUF repo). The context cache only counts full-attention layers.