Tags

Tag: local-llm

• A local 27B coding agent on the 780M: why it froze, and what unstuck it 2026-08-22

  Running Qwen3.8-27B (Q3_K_XL) through DeepSeek Harness as an autonomous coding agent on a Radeon 780M. A full spec in one shot burns the entire output budget on pure thinking and writes nothing, and manual task-splitting, an inline plan-then-act instruction, and the harness's own plan mode each only partly fix it before the model's own goal tool does.

• Minus Zero: running a spec-and-test codegen loop entirely on local models 2026-08-10

  A local codegen loop, escalating from qwen3:8b to qwen3.6:27b, reproduces its own past output, hits one real capability wall in a signed-zero bug, and ships eight tested functions into a working game.