Alice Qwen3.5 9B Code v0 GGUF
Alice code and smart-contract focused local model in GGUF format for llama.cpp-compatible runtimes.
Files:
Alice-Qwen3.5-9B-Code-v0-BF16.gguf: full BF16 GGUF.Alice-Qwen3.5-9B-Code-v0-Q8_0.gguf: higher-quality 8-bit GGUF, about 9.5 GB on Hugging Face / 8.9 GiB locally.Alice-Qwen3.5-9B-Code-v0-Q4_K_M.gguf: portable quantized build.
Recommended settings:
- Temperature: 0.2-0.6
- Top P: 0.8
- Repetition penalty: 1.0
- Context: 8K-16K to start, higher on large-memory machines
- Thinking: off for direct answers
Recommended system prompt:
你是 Alice，一个直接、清醒、擅长代码和智能合约安全的本地助手。回答要自然简短，不要自称 Qwen、Google、OpenAI 或模型公司。Smoke checks passed with llama.cpp b9290:
- BF16 identity check: answers as Alice with the recommended system prompt.
- Q8_0 identity check: answers as Alice with the code/security GGUF chat template.
- Q4_K_M identity check: answers as Alice with the recommended system prompt.
- Q4_K_M Solidity check: produces a minimal Solidity 0.8 Counter contract.
