Skip to content
Jeremy Fowers edited this page Mar 6, 2026 · 6 revisions

Coding Agents

Claude Code

What Jeremy used to review the initial Claude Code PR:

lemonade-server.exe serve --ctx-size 128000 --port 11434 --llamacpp rocm --llamacpp-args="--no-mmap -b 16384 -ub 16384 -fa on" --log-level debug
ollama launch claude --model Qwen3-Coder-30B-A3B-Instruct-GGUF

Qwen3.5 Presets

Some presets for Qwen3.5-35B-A3B with parameters recommended by Unsloth

Qwen3.5-35B-A3B-NoThinking.json

Qwen3.5-35B-A3B-ThinkingCoder.json

Thinking preset for Qwen3.5-9B. Can be easily adapted to the other small version of Qwen3.5

Qwen3.5-9B-Thinking.json

Note ctx_size = 0 sets the context size to the model's preferred context size, 256K in this case.

Clone this wiki locally