---

license: mit

base_model: microsoft/Phi-4-reasoning-vision-15B

base_model_relation: quantized

tags: [gguf, llama.cpp, rocm, gfx1151, strix-halo, amd, ryzen-ai-max-395, rocmfpx, phi, vision, multimodal, reasoning]

language: [en]

pipeline_tag: image-text-to-text

---

> ### ⚠️ STOCK llama.cpp WILL NOT LOAD THIS MODEL

>

> ⚠️ -fa off is required — flash attention breaks the vision path on gfx1151.

>

> 14.34 GiB · 14.08 tok/s on a Ryzen AI MAX+ 395.

Phi-4-Reasoning-Vision-15B — ROCmFPX 8-bit AGENT GGUF

An 8-bit ROCmFPX quantization for AMD gfx1151 (Ryzen AI MAX+ 395 / Strix Halo),

quantized from BF16 GGUF — a lossless source, not a requantization of a lower-bit build.

| | |

|---|---|

| File | Phi-4-reasoning-vision-15B-Q8_0_ROCMFPX_AGENT.gguf |

| Size | 14.34 GiB |

| BPW | 8.40 |

| ftype | Q8_0_ROCMFPX_AGENT (115) |

| mmproj | mmproj-phi-4-reasoning-vision-15b-bf16.gguf (BF16, 862 MiB, included — required for vision) |

⛔ Requires a llama.cpp with the ROCmFPX quant types

Q8_0_ROCMFPX (111) and Q8_0_ROCMFPX_AGENT (115) exist only in

charlie12345/ROCmFPX. Stock llama.cpp reports

invalid ggml type 103. Ignore the auto-generated "Use this model" commands above.

---

All quant variants

Three builds, all measured in one session on one box with one binary (Ryzen AI MAX+ 395,

gfx1151, ROCm 7.2.4, ROCmFPX-2809dc5, -fa off) — so these rows are directly comparable.

Median of 3, warm-up discarded, otherwise-idle box.

| variant | ftype | size | bpw | decode (median) | range | repo |

|---|---|---|---|---|---|---|

| 4-bit COHERENT | 102 | 7.93 GiB | 4.65 | 24.91 | 24.88 – 24.91 | link |

| 8-bit AGENT | 115 | 14.34 GiB | 8.40 | 14.08 | 14.06 – 14.09 | link |

| 8-bit plain | 111 | 14.11 GiB | 8.27 | 14.15 | 14.13 – 14.20 | link |

⚠️ The 4-bit build is ~1.7× faster and 43% smaller. The 8-bit builds exist for accuracy

headroom, not throughput. The two 8-bit builds are within noise of each other (14.08 vs

14.15, ranges touching) — this model has no MTP draft head, and AGENT's benefit shows up in

draft acceptance, so there is nothing here for it to win.

⚠️ -fa off is mandatory — flash attention breaks the vision path on gfx1151.

The BF16 mmproj (862 MiB) ships in every one of these repos and is required for vision.

Correctness: 17×23 ⇒ ✅ 391 · capital of Japan ⇒ ✅ Tokyo · days in 2024 ⇒ ✅ 366

> Give this model room to reason. On a curt "reply with only the number" prompt it can answer

> 365 for the 2024 question; allowed to reason it correctly derives leap year ⇒ 366.

---

What was NOT measured

• No perplexity run, and no quality A/B against the source. The checks above are

memorized-fact prompts — necessary but not sufficient.

• No long-context testing. · No tool-calling evaluation.
• Vision was smoke-tested only (the 4-bit build correctly describes an 8×8 red PNG as Red);

no vision benchmark was run on these 8-bit builds.

Base model licence inherited; credit for the model goes to its authors.