Игорь Федосеев
5dedaf459f
Restore native inference, card-bound readings and paged dialogue. Provide manual, checksum-verified model setup instead of Git LFS so source checkouts work without uploading multi-gigabyte weights. The game stays offline; complete release archives still bundle the model.
961 B
Bundled Local Model
Model: Qwen3-4B-Instruct-2507, by the Qwen team.
Original: https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507
GGUF Q4_K_M conversion: Unsloth,
https://huggingface.co/unsloth/Qwen3-4B-Instruct-2507-GGUF
Pinned revision and file integrity: model.json.
Model license: Apache-2.0, reproduced in licenses/Qwen.txt.
Inference engine: llama.cpp contributors,
https://github.com/ggml-org/llama.cpp
Engine license: MIT, reproduced in licenses/llama.cpp.txt.
The local wrapper source is tools/fortune_ai/worker.cpp in this game repository.
This is a third-party local language model, not ChatGPT or a cloud service.
No runtime model downloads or external servers are used. The model file is part
of the complete release archive, not Git history. After cloning or copying source,
run the separate root download_fortune_model.bat to prepare the pinned weights.
This optional setup downloads and verifies the file before offline gameplay.