Qwopus3.6-27B-Coder-MTP-ROCmFPX

Qwopus3.6-27B-Coder · ROCmFPX

Stock Q6_K quality, ~30% faster prompt-processing on AMD Strix Halo (gfx1151)

ROCmFPX 3→8-bit quants of Jackrong/Qwopus3.6-27B-Coder-MTP-GGUF — 27B, MTP speculative decoding + Qwen3-VL vision, agent/tool-use tuned.

⚠️ Requires the ROCmFPX fork (build main — the FP* types are merged in) — custom AMD quant types (enum IDs 110–115), not upstream-stable. Won't load in stock llama.cpp / LM Studio / Ollama. HF's precision badge is wrong — pick the file by name.

Pick a tier

Agent-routed _AGENT tiers + the full enum/bpw table are in the details below and the Files tab. All filenames prefixed Qwopus3.6-27B-Coder-MTP-STRIX-.

Quick start

# build the fork once — main already has the ROCmFPX quant types
git clone https://github.com/charlie12345/ROCmFPX.git && cd ROCmFPX
JOBS=16 scripts/build-strix-rocmfp4-mtp.sh

# serve the flagship — MTP + vision
HSA_OVERRIDE_GFX_VERSION=11.5.1 build-strix-rocmfp4/bin/llama-server \
  -m Qwopus3.6-27B-Coder-MTP-STRIX-embF16-headQ6-Q6_0_ROCMFPX_AGENT.gguf \
  -dev ROCm0 -ngl 999 -fa on -c 32768 \
  --spec-type draft-mtp --spec-draft-ngl all --spec-draft-n-max 2 \
  --jinja --mmproj mmproj/mmproj-F32.gguf --host 0.0.0.0 --port 8080

Tool calls: point your client at the qwen3_coder parser, or the model narrates code instead of emitting structured calls.

Better tool-calling / agent compatibility (optional). These ROCmFPX quants carry the exact weights of Jackrong/Qwopus3.6-27B-Coder, so the interoperability-hardened chat template from Jackrong's Compat-MTP edition (JSON-string + mapping/list tool args; llama.cpp/minja ↔ HF-Jinja parity) drops in unchanged. If your agent runtime hits flaky function-calling, grab chat_template_compat.jinja from this repo and serve with --chat-template-file chat_template_compat.jinja (overrides the bundled --jinja template). Same weights, just a more portable template.

<details> <summary><b>All tiers · recipe · benchmarks</b></summary>

All tiers

f16 token embeddings throughout; _AGENT presets keep attention/FFN routing at higher precision for tool-call/JSON coherence. (HF labels Q4/Q8 but not Q6/Q3 — the latter aren't standard llama.cpp quant names.)

Verification (Strix Halo gfx1151)

Performance — prompt-processing throughput (t/s) vs Q6_K

The gfx1151-tuned kernels win the compute-bound prefill; the edge is largest at short context and narrows toward +22% at 64k as O(n²) attention takes over. Decode is bandwidth-bound (≈ Q6_K raw), and MTP (--spec-type draft-mtp) ~doubles it in serving. Q4_0_ROCMFP4 is the decode king (~13 tok/s raw). Single-rep llama-bench; treat absolutes as ±a few %.

</details>

Credits & license

Apache-2.0 (inherited). Jackrong + Kyle Hessling (fine-tune) → Qwen3.6-27B (base) → charlie12345 / ROCmFPX (quant fork). ROCmFPX quantization by this repo's author.