Deepseek harness
retained text
im considering using deepseek harness as it claims that everything is a plugin Im using a qwen3.6 35b a3b tiel coder local ai model rx6700xt 12gb vram, 32gb ddr4 3200 and r5 5600x component i tested qwen3.8 27b iq3xxs but im averaging around 10tps unlike the moe model (35tps+) i've tested pi dev and opencode but i kinda hit the deadend because im not sure on what other things to try. Plus i haven't found my actual purpose for local ai models yet, i don't consider switching to cloud models. But i do hope my local ai model can **produce useful work**, could be anything rather than just testing stuff based on hype without making any progress I hope dsh allows me to find one, if not ill look into other suggestion, (useful task that doesn't take hours to complete. )
document fetched 2026-10-29 00:44 · canonical https://reddit.com/r/LocalLLaMA/comments/1xhq8bb/deepseek_harness