Add support for microsoft/Phi-4-reasoning-vision-15B.
This is reusing the existing Phi-3 text path for the decoder and adds a mmproj/mtmd path for Phi SigLIP2 vision.

I uploaded some converted weights for testing https://huggingface.co/dranger003/Phi-4-reasoning-vision-15B-GGUF.
The model generates coherent text and proper image descriptions, including OCR (see below).

DISCLAIMER: GPT-5.4 was used to help with this PR.

EDIT: More information about this model here
https://www.microsoft.com/en-us/research/blog/phi-4-reasoning-vision-and-the-lessons-of-training-a-multimodal-reasoning-model/

Several open-source multimodal language models have adapted their methodologies accordingly, e.g., Gemma3(opens in new tab) uses pan-and-scan and NVILA(opens in new tab) uses Dynamic S2. However, their trade-offs are difficult to understand across different datasets and hyperparameters. To this end, we conducted an ablation study of several techniques. We trained a smaller 5 billion parameter Phi-4 based proxy model on a dataset of 10 million image-text pairs, primarily composed of computer-use and GUI grounding data. We compared with Dynamic S2, which resizes images to a rectangular resolution that minimizes distortion while admitting a tiling by 384×384 squares; Multi-crop, which splits the image into potentially overlapping 384×384 squares and concatenates their encoded features on the token dimension; Multi-crop with S2, which broadens the receptive field by cropping into 1536×1536 squares before applying S2; and Dynamic resolution using the Naflex variant of SigLIP-2, a natively dynamic-resolution encoder with adjustable patch counts.

Our primary finding is that dynamic resolution vision encoders perform the best and especially well on high-resolution data. It is particularly interesting to compare dynamic resolution with 2048 vs 3600 maximum tokens: the latter roughly corresponds to native HD 720p resolution and enjoys a substantial boost on high-resolution benchmarks, particularly ScreenSpot-Pro. Reinforcing the high-resolution trend, we find that multi-crop with S2 outperforms standard multi-crop despite using fewer visual tokens (i.e., fewer crops overall). The dynamic resolution technique produces the most tokens on average; due to their tiling subroutine, S2-based methods are constrained by the original image resolution and often only use about half the maximum tokens. From these experiments we choose the SigLIP-2 Naflex variant as our vision encoder.