Skip to content
Local AI Wizard
Your machine
Theme

Phi-4-multimodal-instruct

Microsoft · microsoft/Phi-4-multimodal-instruct

Weights license: MIT · code and components may use other terms

Reported by the publisher20 resultsMicrosoft's model card · read 2026-10-25 · not ranked · see them
Memory needed~5 GBQ4 weights + working room, before the machine's reserve
Parameters5.6 BHub safetensors total
Repository updated2025-12-10A commit is not a release date

Quality

No comparable score

We track no public, comparable benchmark that covers this model yet. No score is inferred.

Reported by the publisher

20 results · Microsoft's model card · read 2026-10-25
s_AI2D
68.9Vision-Speech tasks
s_ChartQA
69.0Vision-Speech tasks
s_DocVQA
87.3Vision-Speech tasks
s_InfoVQA
63.7Vision-Speech tasks
Average
72.2Vision-Speech tasks
Art Style
86.3Visual Perception
Counting
60.0Visual Perception
Forensic Detection
90.2Visual Perception
Every reported result20
Functional Correspondence
30.0Visual Perception
IQ Test
22.7Visual Perception
Jigsaw
68.7Visual Perception
Multi-View Reasoning
76.7Visual Perception
Object Localization
52.5Visual Perception
Relative Depth
69.4Visual Perception
Relative Reflectance
26.9Visual Perception
Semantic Correspondence
52.5Visual Perception
Spatial Relation
72.7Visual Perception
Visual Correspondence
67.4Visual Perception
Visual Similarity
86.7Visual Perception
Overall
61.6Visual Perception

Values as printed on the card, with the publisher's own settings (card section: Model Quality › Vision › Vision-Speech tasks; Model Quality › Vision tasks › Visual Perception); column “Phi-4-multimodal-instruct”. Not comparable across publishers and not used in any ranking. Model card

Runs on

Fast memory holds it 114

The planning estimate fits the memory the model can use quickly: graphics card VRAM, or unified memory within the macOS GPU share.

Show all 114

Within memory, outside the fast part 5

Within the machine's memory, outside its fast part: on a graphics card the layers that do not fit run on the CPU from PC RAM (the listed RAM, or 32 GB assumed); on a Mac, above the macOS default GPU share. Not a speed figure.

Show all 5

Hardware profiles grouped by the Q4 planning estimate on each machine (dedicated GPUs reserve 2 GB; unified and system memory 25%, 2 to 8 GB): the fast memory holds it, it is within memory but outside the fast part (layers offloaded to PC RAM, or above the macOS GPU share), or the machine is one memory tier short. Not a measured fit or speed.

Consider it for

Microsoft released Phi-4-multimodal-instruct on 2025-02: a vision-language model that answers about images, with 5.6B parameters on the Hugging Face count and the MIT license. Languages the publisher declares: ar, zh, cs, da, nl, en, fi, fr, de, he, hu, it, ja, ko, no, pl, pt, ru, es, sv, th, tr, uk.

Check first

License: commercial use allowed. At Q4 the weights load about 4.2 GB before each machine's reserve; image resolution, video frames and context add memory, and the vision encoder needs support in the runtime you pick.

Measured speed

No published measurement for this checkpoint

Tokens per second are shown only when a source publishes them with hardware, build and setup. They are never estimated.

How to run

Runtime compatibility and tokens per second have not been measured for this entry. Published metadata and a memory estimate do not prove a successful local run.

Library in source: transformers

Documented ways to run it

Copied as printed in the linked source (read 2026-10-18); not run by this site.

Transformers · model card · Loading the model locally
import requests
import torch
import os
import io
from PIL import Image
import soundfile as sf
from transformers import AutoModelForCausalLM, AutoProcessor, GenerationConfig
from urllib.request import urlopen


# Define model path
model_path = "microsoft/Phi-4-multimodal-instruct"

# Load model and processor
processor = AutoProcessor.from_pretrained(model_path, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_path, 
    device_map="cuda", 
    torch_dtype="auto", 
    trust_remote_code=True,
    # if you do not use Ampere or later GPUs, change attention to "eager"
    _attn_implementation='flash_attention_2',
).cuda()

# Load generation config
generation_config = GenerationConfig.from_pretrained(model_path)

# Define prompt structure
user_prompt = '<|user|>'
assistant_prompt = '<|assistant|>'
prompt_suffix = '<|end|>'

# Part 1: Image Processing
print("\n--- IMAGE PROCESSING ---")
image_url = 'https://www.ilankelman.org/stopsigns/australia.jpg'
prompt = f'{user_prompt}<|image_1|>What is shown in this image?{prompt_suffix}{assistant_prompt}'
print(f'>>> Prompt\n{prompt}')

# Download and open image
image = Image.open(requests.get(image_url, stream=True).raw)
inputs = processor(text=prompt, images=image, return_tensors='pt').to('cuda:0')

# Generate response
generate_ids = model.generate(
    **inputs,
    max_new_tokens=1000,
    generation_config=generation_config,
)
generate_ids = generate_ids[:, inputs['input_ids'].shape[1]:]
response = processor.batch_decode(
    generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False
)[0]
print(f'>>> Response\n{response}')

# Part 2: Audio Processing
print("\n--- AUDIO PROCESSING ---")
audio_url = "https://upload.wikimedia.org/wikipedia/commons/b/b0/Barbara_Sahakian_BBC_Radio4_The_Life_Scientific_29_May_2012_b01j5j24.flac"
speech_prompt = "Transcribe the audio to text, and then translate the audio to French. Use <sep> as a separator between the original transcript and the translation."
prompt = f'{user_prompt}<|audio_1|>{speech_prompt}{prompt_suffix}{assistant_prompt}'
print(f'>>> Prompt\n{prompt}')

# Downlowd and open audio file
audio, samplerate = sf.read(io.BytesIO(urlopen(audio_url).read()))

# Process with the model
inputs = processor(text=prompt, audios=[(audio, samplerate)], return_tensors='pt').to('cuda:0')

generate_ids = model.generate(
    **inputs,
    max_new_tokens=1000,
    generation_config=generation_config,
)
generate_ids = generate_ids[:, inputs['input_ids'].shape[1]:]
response = processor.batch_decode(
    generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False
)[0]
print(f'>>> Response\n{response}')

Choose my computer

Technical details

What the source and the catalog store for this checkpoint. Missing fields are left out, never zero.

Task
speech recognition (speech to text)
Architecture
Phi4MMForCausalLM
Library in source
transformers
Weight license
MITmicrosoft/Phi-4-multimodal-instruct model card (Hugging Face) · read 2026-10-18source
Commercial use
Allowed MIT license text (SPDX) · read 2026-10-18license text
Languages
ar, zh, cs, da, nl, en, fi, fr, de, he, hu, it, ja, ko, no, pl, pt, ru, es, sv, th, tr, ukModel card metadata · read 2026-10-25source
Released
Phi-4-multimodal-instruct model card, 'Release date: February 2025' (Microsoft, Hugging Face) · read 2026-10-18source
Every stored field11 fields
Revision checked
93f923e1a7727d1c4f446756212d9d3e8fcc5d81
Architecture
Phi4MMForCausalLM
Library
transformers
Hub task tag
automatic-speech-recognition
Formats
transformers, safetensors
Access
Open download
Publisher kind
Official
Repository created
2025-02-24creation is not the model's release
Repository updated
2025-12-10a commit is not a release date
Hub downloads (30 days) · likes
111,473 · 1,626this repository only · read 2026-10-25
Catalog state
Hub config or weights found; no inventory benchmark link

Sources

Identity and published specifications come from the checkpoint page. Benchmark scores belong to the named release and do not measure speed on your computer.

Curated facts checked: 2026-10-18. The license and release date cite the publisher's sources; parameters, files and task come from the Hugging Face record, and memory is a labelled estimate from its parameter count. A value no source states stays empty; a repository commit is not a model release date.