Views
No views yet
EXPERIMENTAL - REQUIRES CUSTOM BRANCHThese GGUF files will NOT work with mainline llama.cpp. You must use the branch linked below.
https://github.com/smpurkis/llama.cpp/tree/gpt-oss-puzzle-support1# Clone the required branch
2git clone --branch gpt-oss-puzzle-support https://github.com/smpurkis/llama.cpp.git
3cd llama.cpp
4
5# Build (example with Vulkan)
6cmake -B build -DGGML_VULKAN=1
7cmake --build build --config Release -j$(nproc)
8
9# Run
10./build/bin/llama-cli -m gpt-oss-puzzle-88B.MXFP4_MOE.gguf -ngl 99 -fa 1 -p "Hello"| File | Quant | Size | Description |
|---|---|---|---|
gpt-oss-puzzle-88B.f16.gguf | F16 | 47.0 GiB | Full precision (for requantisation) |
gpt-oss-puzzle-88B.MXFP4_MOE.gguf | MXFP4_MOE | 44.8 GiB | Native MXFP4 expert weights (matches original model precision) |
| Property | gpt-oss-120B | gpt-oss-puzzle-88B |
|---|---|---|
| Expert count | 128 per layer (uniform) | 128 or 64 per layer (heterogeneous) |
| Attention pattern | Interleaved global/SWA (single window) | Global + multiple SWA window sizes (128, 8192) |
| Total parameters | ~117B | ~88B |