---

license: apache-2.0

base_model: Qwen/Qwen3.5-4B

base_model_relation: quantized

pipeline_tag: image-text-to-text

library_name: gguf

tags: [gguf, rocmfp4, rocmfpx, strix-halo, gfx1151, amd, rocm, qwen, multimodal, quantized]

---

Qwen3.5-4B — ROCmFP4 / ROCmFPX GGUF

First ROCmFP4/ROCmFPX quantization of Qwen/Qwen3.5-4B. Four quants, built from the Hub

BF16 GGUF (no re-conversion), each individually verified on real hardware.

Built for AMD Strix Halo (gfx1151) — Ryzen AI MAX+ 395, 128 GB unified — with the

ROCmFPX llama.cpp fork.

> ⚠️ Needs a ROCmFPX-capable llama.cpp build. These will not load in stock

> llama.cpp / Ollama / LM Studio.

Variants

| file | ftype | size | token_embd | decode | correctness |

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

| Qwen3.5-4B-Q4_0_ROCMFP4_COHERENT.gguf | 102 | 2.37 GiB | Q6_K | 55.57 t/s | 3/3 |

| Qwen3.5-4B-Q6_0_ROCMFPX_AGENT.gguf | 114 | 3.67 GiB | Q8_0 | 37.13 t/s | 3/3 |

| Qwen3.5-4B-Q8_0_ROCMFPX.gguf | 111 | 4.07 GiB | Q8_0 | 37.25 t/s | 3/3 |

| Qwen3.5-4B-Q8_0_ROCMFPX_AGENT.gguf | 115 | 4.12 GiB | Q8_0 | 36.92 t/s | 3/3 |

mmproj-BF16.gguf is included — required for image input (-fa off).

Which to pick: the 4-bit (102) — fastest and smallest, same answers as the 8-bits.

⚠️ Q6_0_ROCMFPX_AGENT (114) is the weakest choice here: larger than the 4-bit and

slower than the plain 8-bit. The AGENT recipe raises speculative-draft acceptance, and

Qwen3.5-4B ships no drafter, so that benefit cannot apply. It is included for completeness.

Measured — not estimated

Hardware: Ryzen AI MAX+ 395 (Strix Halo, gfx1151), 128 GB unified.

-ngl 999 -c 4096 -fa on -fit off. 300 tokens, warm-up discarded, median of 3.

| ftype | run 1 / 2 / 3 | median | spread |

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

| 102 | 55.58 55.57 55.56 | 55.57 | 1.0 |

| 114 | 37.15 37.13 37.13 | 37.13 | 1.001 |

| 111 | 37.2 37.25 37.41 | 37.25 | 1.006 |

| 115 | 36.85 36.92 36.97 | 36.92 | 1.003 |

No MTP/EAGLE drafter ships with this model, so these are the deployed numbers with

nothing left on the table.

Head protection

Qwen3.5-4B has tied embeddings — there is no output.weight tensor, so

--output-tensor-type is a silent no-op and --token-embedding-type is the only

flag that protects the head. Audited by exact tensor name on every artifact.

   2543899552  Qwen3.5-4B-Q4_0_ROCMFP4_COHERENT.gguf
   3935341472  Qwen3.5-4B-Q6_0_ROCMFPX_AGENT.gguf
   4370869152  Qwen3.5-4B-Q8_0_ROCMFPX.gguf
   4426410912  Qwen3.5-4B-Q8_0_ROCMFPX_AGENT.gguf

Credits

• Base model: Qwen/Qwen3.5-4B
• Quantization stack: ROCmFPX