What are the hardware requirements for Phi 4 Base?

Phi 4 Base requires approximately 29.5 GB of VRAM and supports a context window of 16K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Phi 4 Base and how large is it?

Phi 4 Base is developed by Sorawiz, a model with 14.7b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

Where can I download or evaluate Phi 4 Base?

Phi 4 Base is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

Phi 4 Base by Sorawiz — VRAM 29.5GB, 16K context

Name: Phi 4 Base
Author: Sorawiz

Phi 4 Base is an open-source language model by Sorawiz. Features: 14.7b LLM, VRAM: 29.5GB, Context: 16K, Merged, LLM Explorer Score: 0.16.

Merged Model Arxiv:2306.01708 Base model:ngxson/lora-phi-4-a... Base model:prithivmlmods/phi-4... Base model:prithivmlmods/phi-4... Base model:prithivmlmods/phi-4... Base model:prithivmlmods/phi-4... Conversational Endpoints compatible Llama Lora Region:us Safetensors Sharded Tensorflow

Model Card on HF 🤗: https://huggingface.co/Sorawiz/Phi-4-Base

Phi 4 Base Parameters and Internals

LLM Name	Phi 4 Base
Repository 🤗	https://huggingface.co/Sorawiz/Phi-4-Base
Base Model(s)	Phi 4 Math IO LoRA Phi 4 Abliterated Phi 4 O1 LoRA Phi 4 Abliterated Phi 4 QwQ LoRA Phi 4 Abliterated Phi 4 Empathetic LoRA Phi 4 Abliterated prithivMLmods/Phi-4-Math-IO ngxson/LoRA-phi-4-abliterated prithivMLmods/Phi-4-o1 ngxson/LoRA-phi-4-abliterated prithivMLmods/Phi-4-QwQ ngxson/LoRA-phi-4-abliterated prithivMLmods/Phi-4-Empathetic ngxson/LoRA-phi-4-abliterated
Merged Model	Yes
Model Size	14.7b
Required VRAM	29.5 GB
Updated	2026-05-16
Maintainer	Sorawiz
Model Type	llama
Model Files	5.0 GB: 1-of-6 5.0 GB: 2-of-6 5.0 GB: 3-of-6 4.9 GB: 4-of-6 5.0 GB: 5-of-6 4.6 GB: 6-of-6
Model Architecture	LlamaForCausalLM
Context Length	16384
Model Max Length	16384
Transformers Version	4.46.2
Tokenizer Class	GPT2Tokenizer
Padding Token	<\|dummy_87\|>
Vocabulary Size	100352
LoRA Model	Yes
Torch Data Type	float16

Rank the Phi 4 Base capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Best Alternatives to Phi 4 Base

Best Alternatives	Context / RAM	Downloads	Likes
...kpoints 511 Onlytrainex Merged	31K / 29.4 GB	25	0
...ruct Extended 3500steps Merged	31K / 29.4 GB	71	0
E2 Phi4 Instruct NTK Scaled	31K / 29.4 GB	20	0
...ct NTK Scaled 1000steps Merged	31K / 29.4 GB	5	0
Phi 4	16K / 29.4 GB	1539	91
Luminis Phi 4	16K / 29.5 GB	1517	13
Tea	16K / 29.4 GB	324	2
Phi 4 Sce Exp V0.1	16K / 29.5 GB	9	1
Phi 4 RR Shoup	16K / 29.5 GB	6	2
Phi 4 RStock V0.1	16K / 29.5 GB	8	0

Note: green Score (e.g. "73.2") means that the model is better than Sorawiz/Phi-4-Base.