ZAYA1-8B-ROCmFPX-Q8_0-AGENT-GGUF

🔧 Runtime: build the ROCmFPX fork below

Stock llama.cpp will not load this file. You need both the zaya architecture and the ROCmFP4 tensor types in one tree. Upstream charlie12345/ROCmFPX has the ROCmFP4 types but not zaya. Our fork has both:

kingjones30/ROCmFPX — a fork of charlie12345/ROCmFPX, branch main.

git clone https://github.com/kingjones30/ROCmFPX.git
cd ROCmFPX
cmake -B build -DGGML_HIP=ON -DGPU_TARGETS=gfx1151 -DGGML_NATIVE=ON -DCMAKE_BUILD_TYPE=Release
cmake --build build --target llama-server llama-quantize -j$(nproc)

Verified 2026-08-27 on gfx1151: clean clone → 0 build errors → llama-server loads a zaya ROCmFP4 GGUF from this family and generates coherent text.

⚠️ STOCK llama.cpp WILL NOT LOAD THIS MODEL

8.72 GiB · 21.02 tok/s on a Ryzen AI MAX+ 395.

ZAYA1-8B — 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.

⛔ tie_word_embeddings is TRUE, so output.weight does not exist — --output-tensor-type is a silent no-op here and --token-embedding-type is the flag that lands (262K vocab).

⛔ Requires a llama.cpp with the ROCmFPX quant types

Q8_0_ROCMFPX (ftype 111) and Q8_0_ROCMFPX_AGENT (ftype 115) exist only in charlie12345/ROCmFPX, not upstream llama.cpp. Stock llama.cpp reports invalid ggml type 103. Ignore the auto-generated "Use this model" commands above.

---

All quant variants

Three builds of this model, all measured in one session on one box with one binary (Ryzen AI MAX+ 395, gfx1151, ROCm 7.2.4, ROCmFPX-2809dc5) — so these rows are directly comparable. Median of 3, warm-up discarded, otherwise-idle box.

⚠️ Decode is ~88% weight-independent on this architecture (the CCA grouped conv is ~55% of decode). All three builds land within ~10% of each other; the 4-bit is smallest and marginally fastest. No 8-bit or 4-bit format will make this model meaningfully faster.

What AGENT actually changes: it keeps far more tensors at true Q8_0 instead of the packed 8-bit type — measured in these files, 154 tensors vs 1 tensor. On models with an MTP draft head that raises draft acceptance and wins ~6%; these two models have no MTP head, and here the two 8-bit builds are within noise of each other.

Correctness: All three builds answer correctly. On some prompts content is empty with finish_reason=length and the correct answer sits in reasoning_content — this model is verbose, give it ≥1024 tokens.

Per-tensor types (audited in this finished file)

token_embd Q8_0 · 247 packed TYPE_103 · 842 F32 · 40 BF16 · 154 Q8_0

---

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; a damaged model can pass them.
• No long-context testing. · No tool-calling evaluation.

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