What are the hardware requirements for Phi 3.5 Mini Instruct?

Phi 3.5 Mini Instruct requires approximately 7.7 GB of VRAM and supports a context window of 128K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Phi 3.5 Mini Instruct and how large is it?

Phi 3.5 Mini Instruct is developed by microsoft, a model with 3.8b 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 3.5 Mini Instruct?

Phi 3.5 Mini Instruct is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

Phi 3.5 Mini Instruct by microsoft — VRAM 7.7GB, 128K context

Name: Phi 3.5 Mini Instruct
Rating: 3.57 (7 reviews)
Author: microsoft

Phi 3.5 Mini Instruct is an open-source language model by microsoft. Features: 3.8b LLM, VRAM: 7.7GB, Context: 128K, License: mit, Instruction-Based, LLM Explorer Score: 0.36.

Arxiv:2403.06412 Arxiv:2404.14219 Arxiv:2407.13833 Code Conversational Custom code Deploy:sagemaker Endpoints compatible Eval-results Instruct Multilingual Phi3 Region:us Safetensors Sharded Tensorflow

Model Card on HF 🤗: https://huggingface.co/microsoft/Phi-3.5-mini-instruct

Phi 3.5 Mini Instruct Parameters and Internals

Model Type

text-generation

Use Cases

Areas:

research, commercial applications

Applications:

AI systems, natural language processing

Primary Use Cases:

memory/compute constrained environments, latency-bound scenarios, strong reasoning tasks

Limitations:

languages other than English may have worse performance

Considerations:

Developers should consider model limitations and adhere to safety and regulatory guidelines.

Additional Notes

None

Supported Languages

Arabic (supported), Chinese (supported), Czech (supported), Danish (supported), Dutch (supported), English (supported), Finnish (supported), French (supported), German (supported), Hebrew (supported), Hungarian (supported), Italian (supported), Japanese (supported), Korean (supported), Norwegian (supported), Polish (supported), Portuguese (supported), Russian (supported), Spanish (supported), Swedish (supported), Thai (supported), Turkish (supported), Ukrainian (supported)

Training Details

Data Sources:

publicly available documents, textbook-like synthetic data

Data Volume:

3.4T tokens

Methodology:

supervised fine-tuning, proximal policy optimization, and direct preference optimization

Context Length:

128000

Training Time:

10 days

Hardware Used:

512 H100-80G GPUs

Model Architecture:

dense decoder-only Transformer

Safety Evaluation

Methodologies:

red-teaming, adversarial conversation simulations

Findings:

models may refuse undesirable outputs in English across multiple languages

Risk Categories:

misinformation, bias

Ethical Considerations:

Industry-wide investment in high-quality safety evaluation datasets is needed.

Responsible Ai Considerations

Fairness:

Models may over- or under-represent groups of people and need fine-tuning for diversity.

Transparency:

Model operation and biases should be understood and communicated to users.

Accountability:

Microsoft accountable for model's outputs.

Mitigation Strategies:

Utilize safety classifiers and fine-tuning based on deployment scenarios.

Input Output

Input Format:

Text inputs with chat format expected

Accepted Modalities:

text

Output Format:

Generated text

Performance Tips:

Use in-memory or latency-bound scenarios.

Release Notes

Version:

June 2024

Date:

2024-06

Notes:

Updated with feedback, improved conversation quality in multilingual settings.

LLM Name	Phi 3.5 Mini Instruct
Repository 🤗	https://huggingface.co/microsoft/Phi-3.5-mini-instruct
Model Size	3.8b
Required VRAM	7.7 GB
Updated	2026-08-03
Maintainer	microsoft
Model Type	phi3
Instruction-Based	Yes
Model Files	5.0 GB: 1-of-2 2.7 GB: 2-of-2
Model Architecture	Phi3ForCausalLM
License	mit
Context Length	131072
Model Max Length	131072
Transformers Version	4.43.3
Tokenizer Class	LlamaTokenizer
Padding Token	<\|endoftext\|>
Vocabulary Size	32064
Torch Data Type	bfloat16

Rank the Phi 3.5 Mini Instruct 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

Quantized Models of the Phi 3.5 Mini Instruct

Model	Likes	Downloads	VRAM
Phi 3.5 Mini Instruct GGUF	33	142279	0 GB
Sign Language 20250823 190451	0	18	7 GB
Phi 3.5 Mini Instruct Bnb 4bit	13	8758	2 GB
Flow Judge V0.1 AWQ	6	1605	2 GB
...nt4 Asym Bf16 Onnx Ryzen Strix	3	11	0 GB

Best Alternatives to Phi 3.5 Mini Instruct

Best Alternatives	Context / RAM	Downloads	Likes
Phi 4 Mini Instruct	128K / 7.7 GB	514603	791
Phi 3 Mini 128K Instruct	128K / 7.7 GB	233663	1700
Phi 3 Mini 128K Instruct	128K / 7.7 GB	7	0
Phi 4 Mini Abliterated	128K / 7.7 GB	33	1
Phi 3 Mini 128K Instruct	128K / 7.7 GB	11	0
Phi 4 Mini Instruct Heretic	128K / 7.7 GB	13	0
Phi 3.5 Mini TitanFusion 0.1	128K / 7.7 GB	7	0
NuExtract V1.5	128K / 7.7 GB	108511	89
MediPhi Instruct	128K / 7.7 GB	1671	69
ECE EIFFEL 3Bv2	128K / 7.7 GB	5	0

Note: green Score (e.g. "73.2") means that the model is better than microsoft/Phi-3.5-mini-instruct.