PH

Phi-4-Reasoning

PhiGGUF by unslothUpdated 2025-05-01
Download on Hugging Face
Parameters14.7B
Context32K
Layers40
Architecturephi3
Smallest quant3.6 GB
Quick answer

Phi-4-Reasoning needs about 11.2 GB of memory at Q4_K_M (a 9.1 GB file) with an 8K context window, so it fits on a 12 GB GPU. This repository has 24 quants, from 3.6 GB (IQ1_S) to 29.3 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 6.7 GB.

Phi-4-Reasoning: 14.7B Phi3 model focused on coding and reasoning

Phi-4-Reasoning is a 14.7B parameter Phi3 architecture model optimized for coding and reasoning tasks. It supports a 32768 token context window and is released under the MIT license. The model is available in multiple GGUF quantizations.

Good at

  • Coding tasks
  • Reasoning problems
  • Long context handling

Limits

  • No vision support
  • Requires temperature=0.8, top_k=50, top_p=0.95 for best results
  • Large memory for higher quant levels

Best for

  • Competitive coding contests
  • Math olympiad problem solving
  • Complex chain‑of‑thought reasoning

Who it’s for. Developers and researchers needing a large Phi model for code generation and advanced reasoning. The model is suitable for applications that can supply the recommended inference parameters and have sufficient GPU memory.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
IQ1_S3.6 GB 5.8 GB
Fits Get file →
IQ1_M3.8 GB 6.0 GB
Fits Get file →
IQ2_XXS4.3 GB 6.4 GB
Fits Get file →
IQ2_M5.3 GB 7.4 GB
Tight Get file →
Q2_K5.5 GB 7.7 GB
Tight Get file →
Q2_K_L5.7 GB 7.8 GB
Tight Get file →
Q2_K_XL5.8 GB 8.0 GB
Tight Get file →
IQ3_XXS5.9 GB 8.1 GB
Too big Get file →
Q3_K_S6.5 GB 8.7 GB
Too big Get file →
Q3_K_M7.4 GB 9.5 GB
Too big Get file →
Q3_K_XL7.4 GB 9.6 GB
Too big Get file →
IQ4_XS8.0 GB 10.1 GB
Too big Get file →
IQ4_NL8.4 GB 10.6 GB
Too big Get file →
Q4_08.4 GB 10.6 GB
Too big Get file →
Q4_K_XL8.9 GB 11.1 GB
Too big Get file →
Q4_K_M9.1 GB 11.2 GB
Too big Get file →
Q4_19.3 GB 11.4 GB
Too big Get file →
Q5_K_XL10.4 GB 12.6 GB
Too big Get file →
Q5_K_M10.6 GB 12.8 GB
Too big Get file →
Q6_K12.0 GB 14.2 GB
Too big Get file →
Q6_K_XL13.1 GB 15.3 GB
Too big Get file →
Q8_015.6 GB 17.8 GB
Too big Get file →
Q8_K_XL18.0 GB 20.2 GB
Too big Get file →
BF1629.3 GB 31.5 GB
Too big Get file →

For 8 GB VRAM, use IQ1_S (3.58 GB) or IQ1_M (3.83 GB). For 12 GB VRAM, IQ2_M (5.26 GB) or Q2_K (5.55 GB) are good choices. For 24 GB VRAM, Q5_K_XL (10.39 GB) or Q6_K (12.03 GB) provide higher precision. Memory estimates are approximate.

What GPU do you need for Phi-4-Reasoning?

These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.

  • 4 GB VRAM: no quant fits at an 8K context.
  • 6 GB VRAM: the best quant that fits is IQ1_M (3.8 GB file, about 6.0 GB in use).
  • 8 GB VRAM: the best quant that fits is Q2_K_XL (5.8 GB file, about 8.0 GB in use).
  • 12 GB VRAM: the best quant that fits is Q4_1 (9.3 GB file, about 11.4 GB in use).
  • 16 GB VRAM: the best quant that fits is Q6_K_XL (13.1 GB file, about 15.3 GB in use).
  • 24 GB VRAM: the best quant that fits is Q8_K_XL (18.0 GB file, about 20.2 GB in use).
  • 32 GB VRAM: the best quant that fits is BF16 (29.3 GB file, about 31.5 GB in use).

Which Phi-4-Reasoning quant should you choose?

Q4_K_M is the usual starting point for Phi-4-Reasoning: it balances file size and quality well. Here it is a 9.1 GB file.

The Q8_0 file is 15.6 GB, about 1.7 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.

If memory is tight, the smallest file is IQ1_S at 3.6 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architecturephi3
Parameters14.7B
Layers40
Trained context32,768 tokens
KV cache (fp16)200 KB per token

Source

Quantized byunsloth
FamilyPhi
Licensemit
Downloads (HF)2,184
Last updated2025-05-01

Frequently asked questions about Phi-4-Reasoning

How much VRAM does Phi-4-Reasoning need?

About 11.2 GB at Q4_K_M with an 8K context window. That is the 9.1 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.

Can I run Phi-4-Reasoning on an 8 GB GPU?

Yes, with Q2_K_XL: the 5.8 GB file needs about 8.0 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of Phi-4-Reasoning GGUF?

It depends on the quant: IQ1_S is 3.6 GB, IQ1_M is 3.8 GB, IQ2_XXS is 4.3 GB, IQ2_M is 5.3 GB, Q2_K is 5.5 GB, Q2_K_L is 5.7 GB and 18 more.

What context length does Phi-4-Reasoning support?

The GGUF header lists a trained context length of 32,768 tokens (32K). The KV cache grows with the context you use: about 1.7 GB at 8K and 6.7 GB at 32K.

Where can I download Phi-4-Reasoning GGUF?

The files are on Hugging Face in the unsloth/Phi-4-reasoning-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.

What license does Phi-4-Reasoning use?

Hugging Face lists the license as mit. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.

Can I run Phi-4-Reasoning on 8 GB VRAM?

Yes, use the IQ1_S or IQ1_M quantizations. They require about 5.8–6 GB VRAM at 8k context.

What inference parameters should I use?

Set temperature=0.8, top_k=50, top_p=0.95, do_sample=True, and max_new_tokens=32768 for best performance.

Does the model support vision?

No, Phi-4-Reasoning does not have vision capabilities.

Other GGUFs of Phi-4-reasoning

Similar models

Browse all →

More from unsloth

All by unsloth →

More Phi models

All Phi →

Other medium models

All medium models →