on

Can the NVIDIA L4 run Phi-4 Multimodal Instruct?

43Fair quality (C) · 43 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking.

Yes, it runs great

Runs great. At Q6_K it needs about 7.4 GB of 24.0 GB usable, with an estimated speed of 45.9 tok/s.

Best quant
Q6_K
Generation speed
45.9 tok/s
Prompt processing
3,917 tok/s
First token (1K prompt)
0.3 s
Largest safe context
128K

Best app to run it on the NVIDIA L4

On Linux

vLLM

Inference server · runs on vLLM

  • The full model fits in GPU memory at 8-bit, so it can batch many requests: best for serving a team or an app.
vllm serve microsoft/Phi-4-multimodal-instruct
  • SGLang The full model fits in GPU memory at 8-bit, so it can batch many requests: best for serving a team or an app.
  • Docker Model Runner Handy if you already work in Docker.

On Windows

vLLM

Inference server · runs on vLLM

  • The full model fits in GPU memory at 8-bit, so it can batch many requests: best for serving a team or an app.
  • On Windows it runs through WSL2.
vllm serve microsoft/Phi-4-multimodal-instruct
  • SGLang The full model fits in GPU memory at 8-bit, so it can batch many requests: best for serving a team or an app.
  • Docker Model Runner Handy if you already work in Docker.

How we choose: the files published for this model, your operating system and GPU, whether it fits fully in memory, and published speed tests (MLX vs llama.cpp on Apple silicon, vLLM vs llama.cpp vs Ollama on NVIDIA). Details.

Download Phi-4 Multimodal Instruct

Original weights · Hugging Face

Full-precision files from Microsoft, for fine-tuning, vLLM, Transformers.

huggingface.co/microsoft/Phi-4-multimodal-instruct (opens in a new tab)

No sign-in needed.

Links checked 3 days ago against Hugging Face, Ollama and GitHub.

How it’s rated

Quality 43Fair quality (C) · 43 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking.
How capable the model itself is at Q6_K, from its editorial quality tier scaled by the quantization or precision. It’s the same on any hardware that runs this variant. Scale: A+ 85+ · A 70+ · B 55+ · C 40+ · D under 40.
Runs Runs great
How well it runs on the NVIDIA L4: speed, capped by how well it fits in memory. Runs great: 30+ tokens/s (or near the image, video or decision speed target) · Runs well: 10+ tokens/s · Slow: 3+ tokens/s · Very slow: under 3 · Won’t run: doesn’t fit even with offload. A tight fit or RAM offload caps it at Runs well; heavy offload caps it at Slow.

Match score

To order lists, each result also gets a match score out of 100 from four parts. It favors the most capable model that still runs well.

Model qualityEditorial quality tier of the model, scaled by the quantization or precision used.19.6 / 45
Memory fitFull marks when everything sits in GPU memory; fewer when part spills to system RAM.25.0 / 25
SpeedGeneration speed against a 40 tokens/s target for LLMs, or render time against a target for image and video.15.0 / 15
Memory useRewards using the memory you have, so a bigger, better variant beats a tiny one that leaves most memory idle.7.2 / 15
Match score66.8 / 100

The match score orders results; it is not shown as a grade. Quality and Runs are the two ratings in the tables.

Every quantization on the NVIDIA L4

Quant Memory needed Available Speed Runs Quality Download
Q2_K 5.1 GB 24.0 GB 109 tok/s Runs great 30Basic quality (D) · 30 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. All files (opens in a new tab)
Q3_K_M 5.6 GB 24.0 GB 85.5 tok/s Runs great 36Basic quality (D) · 36 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. All files (opens in a new tab)
Q4_K_M 6.3 GB 24.0 GB 66.1 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. All files (opens in a new tab)
Q5_K_M 6.8 GB 24.0 GB 54.6 tok/s Runs great 42Fair quality (C) · 42 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. All files (opens in a new tab)
Q6_K 7.4 GB 24.0 GB 45.9 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. All files (opens in a new tab)
Q8_0 8.8 GB 24.0 GB 33.7 tok/s Runs great 44Fair quality (C) · 44 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. All files (opens in a new tab)
F16 14.0 GB 24.0 GB 15.5 tok/s Runs well 44Fair quality (C) · 44 of 100: how capable the model is at F16, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. Original weights (opens in a new tab)

Memory figures assume 8K of context. Longer conversations or documents grow the KV cache.

Other hardware that runs it

Quick answers

How much memory does Phi-4 Multimodal Instruct need on the NVIDIA L4 24GB?

At Q6_K it needs about 7.4 GB. The NVIDIA L4 24GB has 24.0 GB usable.

How fast is Phi-4 Multimodal Instruct on the NVIDIA L4 24GB?

About 45.9 tok/s at Q6_K, an estimate from the hardware’s memory bandwidth and the size of the model.

Which quantization of Phi-4 Multimodal Instruct should I use on the NVIDIA L4 24GB?

Q6_K. Of the versions that fit, it is the one that runs best; when two run equally well, the check prefers the higher-rated one.

Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.

Was this page helpful?