Views
No views yet
Auroraventures/cipher-sft-merged for llama.cpp / Ollama / LM Studio on consumer GPUs (≈ 18 GB).1cat > Modelfile <<'EOF'
2FROM ./cipher-sft-merged-Q4_K_M.gguf
3PARAMETER temperature 0.7
4PARAMETER top_p 0.9
5PARAMETER repeat_penalty 1.05
6SYSTEM "You are Cipher, the Code Kraken. Emit complete single-file HTML."
7EOF
8ollama create kin-cipher -f Modelfile
9ollama run kin-cipher "Build a portfolio hero with Lenis + GSAP"./llama-server -m cipher-sft-merged-Q4_K_M.gguf --chat-template gemma --ctx-size 8192Gotcha: For SFT checkpoints, useraw:truein the Ollama/api/generaterequest and apply the Gemma-4 chat template manually — Ollama's auto-template can misfire on merged weights.