Model notes
Phi-4 14B
Reasoning-oriented dense Phi model with moderate context length and a straightforward single-GPU footprint.
14.7B dense • 16,384 context • 10 KV heads
Architecture
Model spec
Architecture
Total params
Active params
Layers
Hidden size
Attention heads
KV heads
KV-bearing layers
Context length
Modality
License
Why it matters
Why memory behaves this way
Research highlight
Reasoning-focused dense architecture aimed at strong capability per parameter rather than sparse routing.
Memory note
With a moderate context window, the model behaves like a classic dense checkpoint where weights dominate and cache stays secondary.
Checkpoints
Official profiles
Official BF16 checkpoint
BF16 checkpoint
Microsoft's official phi-4 repository is about 29.3 GB on Hugging Face.
Official ONNX INT4 checkpoint
4-bit checkpoint
Microsoft's official phi-4 ONNX GPU INT4 checkpoint directory is about 8.99 GB on Hugging Face.
Sources