What are the hardware requirements for Phi 2?

Phi 2 requires approximately 5.6 GB of VRAM and supports a context window of 2K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Phi 2 and how large is it?

Phi 2 is developed by microsoft, a model with 2.8b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does Phi 2 perform on standard benchmarks?

Phi 2 has the following published scores: MMLU 57.92. Compare against reference models on this page or on the LLM Explorer leaderboards.

Phi 2 by microsoft — VRAM 5.6GB, 2K context

Name: Phi 2
Rating: 1 (2 reviews)
Author: microsoft

Phi 2 is an open-source language model by microsoft. Features: 2.8b LLM, VRAM: 5.6GB, Context: 2K, License: mit, HF Score: 61.1, LLM Explorer Score: 0.33, Arc: 61, HellaSwag: 74.9, MMLU: 57.9, TruthfulQA: 44.2, WinoGrande: 73.5, GSM8K: 55.

Autotrain compatible Code En Endpoints compatible Phi Region:us Safetensors Sharded Tensorflow

Model Card on HF 🤗: https://huggingface.co/microsoft/phi-2

Phi 2 Benchmarks

IFEval: 27.39 vs 88 (so35)^-68.9%

ARC: 61.01 vs 96.7 (so35)^-36.9%

HellaSwag: 74.92 vs 95.3 (gpt4)^-21.4%

MMLU: 57.92 vs 88.3 (so35)^-34.4%

TruthfulQA: 44.24 vs 59 (gpt4)^-25%

WinoGrande: 73.48 vs 87.5 (gpt4)^-16%

GSM8K: 54.97 vs 96.4 (so35)^-43%

MATH Lvl 5: 2.95

LLME Score: 0.33006

^nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").

What is the LLM Explorer Rank (Score)

🌟 Advertise your project 🚀

Phi 2 Parameters and Internals

Model Type

text generation, code generation

Use Cases

Areas:

research

Applications:

QA, chat, code format

Primary Use Cases:

QA format, chat format, code format

Limitations:

Generate Inaccurate Code and Facts, Limited Scope for code, Unreliable Responses to Instruction, Language Limitations, Potential Societal Biases, Toxicity, Verbosity

Additional Notes

Phi-2 is intended for QA, chat, and code purposes. Model-generated text/code should be treated as a starting point. Users should be cautious when employing these models in applications.

Supported Languages

en (Standard English)

Training Details

Data Sources:

Phi-1.5, NLP synthetic texts, filtered websites

Data Volume:

250B tokens

Methodology:

Transformer-based model with next-word prediction objective

Context Length:

2048

Training Time:

14 days

Hardware Used:

96xA100-80G GPUs

Model Architecture:

Transformer-based model with next-word prediction objective

Input Output

Accepted Modalities:

text

Performance Tips:

Phi-2 has an attention overflow issue (with FP16). If encountering this issue, enable/disable autocast on the PhiAttention.forward() function.

LLM Name	Phi 2
Repository 🤗	https://huggingface.co/microsoft/phi-2
Model Size	2.8b
Required VRAM	5.6 GB
Updated	2025-09-23
Maintainer	microsoft
Model Type	phi
Model Files	5.0 GB: 1-of-2 0.6 GB: 2-of-2
Supported Languages	en
Model Architecture	PhiForCausalLM
License	mit
Context Length	2048
Model Max Length	2048
Transformers Version	4.37.0
Tokenizer Class	CodeGenTokenizer
Vocabulary Size	51200
Torch Data Type	float16

Quantized Models of the Phi 2

Model	Likes	Downloads	VRAM
Phi 2 GGUF	232	20535	1 GB
Phi2 Bunny	3	673	5 GB
Damysus 2.7B Chat GGUF	0	67	1 GB
Phi 2 GGUF	6	240	1 GB
Phi 2 GGUF	3	207	1 GB
Phi 2 GPTQ	30	151	1 GB

Best Alternatives to Phi 2

Best Alternatives	Context / RAM	Downloads
MFANN3bv0.24	128K / 11.1 GB	5
MFANN3b	128K / 11.1 GB	168
MFANN3bv1.4	128K / 11.1 GB	8
MFANN Phigments Slerp V3.2	128K / 5.6 GB	10
MFANN3bv1.3	128K / 11.1 GB	11
MFANN3bv0.23	128K / 11.1 GB	6
MFANN3bv1.1	128K / 11.1 GB	5
MFANN3bv1.5	128K / 11.1 GB	5
MFANN3b Rebase	128K / 11.1 GB	7
MFANN Liminerity Slerp 4a	128K / 5.6 GB	6

Rank the Phi 2 Capabilities

🆘 Have you tried this model? Rate its performance. This feedback would greatly assist ML community in identifying the most suitable model for their needs. Your contribution really does make a difference! 🌟

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

What open-source LLMs or SLMs are you in search of? 53254 in total.

Email us: info@extractum.io. Our Privacy Policy | Terms and Conditions | Suggest an improvement.

Our Social Media →

Original data from HuggingFace, OpenCompass and various public git repos.

Check out Ag3ntum — our secure, self-hosted AI agent for server management.

Release v20260328a

Support LLM Explorer

Phi 2 by microsoft