---
license: apache-2.0
library_name: llama.cpp
base_model:
- Qwen/Qwen3.6-35B-A3B
base_model_relation: quantized
pipeline_tag: text-generation
quantized_by: ROCmFPX
language:
- en
tags:
- gguf
- rocm
- amd
- strix-halo
- gfx1151
- rocmfpx
- quantized
- magicquant
---

# Qwen3.6-35B-A3B-ROCmFPX-GGUF

> ## ⚠️ These files do NOT load on standard llama.cpp
> They use AMD-native `*_ROCMFPX` tensor types from the experimental
> [ciru-ai/ROCmFPX](https://github.com/ciru-ai/ROCmFPX) llama.cpp fork (build from source).

Derivative of [Qwen3.6-35B-A3B](https://huggingface.co/Qwen/Qwen3.6-35B-A3B), quantized using MagicQuant hybrid evolutionary per-tensor search and quantized to AMD-native [ROCmFPX](https://github.com/ciru-ai/ROCmFPX) formats (fork-only) tuned for Strix Halo (gfx1151).

## Base Model

This is a derivative of [Qwen3.6-35B-A3B](https://huggingface.co/Qwen/Qwen3.6-35B-A3B).
All credit for the base model architecture and weights goes to the original authors.
The base model's license applies to this derivative.

## Quantization Method

Quantized using **[MagicQuant](https://github.com/lucasmcoleman/MagicQuant)** hybrid evolutionary per-tensor quantization,
based on the methodology by **[magiccodingman](https://github.com/magiccodingman/MagicQuant-Wiki)**:

- Tensors are classified into sensitivity groups (Embeddings, Head, Query, Key, Output, FFN Up/Down, MoE Experts, Router)
- An evolutionary search finds the optimal quantization type per group, balancing size vs. perplexity
- **Q4/Q5/Q6 tier targets** are produced with different size-quality tradeoffs
- Small-row tensors and sensitivity-critical layers (embeddings, output head, router) are kept at F32/F16/BF16
- This is NOT a uniform quantization -- each tensor group gets its own optimal type

## ROCmFPX (AMD-native, fork-only)

These GGUFs use AMD-native quantization schemes from the experimental
**[ciru-ai/ROCmFPX](https://github.com/ciru-ai/ROCmFPX)** llama.cpp fork,
tuned for and benchmarked on AMD Strix Halo (Radeon 8060S iGPU, gfx1151, unified memory):

- `ROCmFP3/4/6/8` tensor types with straight and "agent" presets (agent presets keep
  tool-calling / JSON-structured output reliable at low bit-widths)
- Files load **only** on the fork -- it is an experimental upstream research
  build, so build from the pinned commit that produced these files (the
  default branch may have moved on since):

```bash
git clone https://github.com/ciru-ai/ROCmFPX.git ROCmFPX
cd ROCmFPX
git checkout 68f23f34c12d7e61177a034b0d8d3fea2129565e
# then build per the fork's own README
```

## GGUF Files

| File | Size | Quant |
|------|------|-------|
| [Qwen3.6-35B-A3B-ROCMFPX-MQ-Q4.gguf](./Qwen3.6-35B-A3B-ROCMFPX-MQ-Q4.gguf) | 22.7 GB | MagicQuant Q4 layout in ROCmFPX types (hybrid, fork-only) |

## Usage

Requires a from-source build of the [ROCmFPX fork](https://github.com/ciru-ai/ROCmFPX)
(stock llama.cpp, LM Studio, and Ollama cannot load these files):

```bash
# Interactive chat (--jinja uses the model's embedded chat template)
llama-cli -m Qwen3.6-35B-A3B-ROCMFPX-MQ-Q4.gguf -c 8192 --jinja -cnv

# Server mode
llama-server -m Qwen3.6-35B-A3B-ROCMFPX-MQ-Q4.gguf -c 8192 --port 8080 -ngl 99 -fa on --jinja
```

## Serving: MTP Speculative Decoding

This model includes **MTP ("nextn") draft tensors**, enabling self-speculative
decoding -- measured **~1.6-1.9x faster generation** with a ~95% first-token
accept rate (no separate draft model needed; it drafts from itself):

```bash
llama-server -m Qwen3.6-35B-A3B-ROCMFPX-MQ-Q4.gguf -c 8192 --port 8080 --host 127.0.0.1 -ngl 99 -md Qwen3.6-35B-A3B-ROCMFPX-MQ-Q4.gguf --spec-type draft-mtp -ctk q8_0 -ctv q8_0 -fa on
```

**Memory cost:** MTP needs its own draft context alongside the main context,
so serving with it uses roughly **2x the model's memory** compared to serving
without ``-md``/``--spec-type draft-mtp``.

## Caveats

- The base model's license (apache-2.0) applies to all derivative files
- **Fork-only files**: stock llama.cpp, LM Studio, and Ollama cannot load these -- build [ciru-ai/ROCmFPX](https://github.com/ciru-ai/ROCmFPX) from source
- Quantization reduces precision -- verify outputs for your specific use case
- The hybrid quantization assigns different precision to different tensor groups, which means quality characteristics may differ from uniform quantizations

## Limitations

- Quantized models may exhibit subtle differences from the full-precision fine-tune
- This model inherits any limitations and biases present in the base model

---
*Generated with [MagicQuant](https://github.com/lucasmcoleman/MagicQuant)*