light weight model to execute task while having a larger (more intelligent model to plan)
retained text
Below i start i have the following: rx6700xt 12gb vram (AMD) r5 5600x 32gb ddr4 3200 llama.cpp vulkan sdk qwen3.8 27b iq3xxs tielcoder qwen3.6 35b a3b ***Windows 11 pro*** since im out most of the time, i wanted to deploy my "local" agents to complete a series of task which ill get to in a second. The steps is as followed, \-at time x, the larger model is automatically deployed and run in llama.cpp \-somehow uses pi dev, to start completing a set of task assigned from the user (usually planned before hand and the model executes after) \-after it plans, its information, ideas are then saved somewhere in my desktop \-it thens **unloads** itself and have something to launch the smaller models \-the smaller models then completes the task, instructions all from the larger model \-after its completed it will offload itself this part is something that could be implemented but is optional if the process or load is too much \-the larger model then loads itself and checks the process, proceeds the information \-if the task is a success it will offload itself, if not it will provide new instructions and the smaller models will work on it after (agent loop i guess) "note" the smaller model shouldn't waste time rechecking or making sure it works well, the larger model should have planned before hand ***Im looking forward for any suggestions for models that satisfies these needs*** do not i can't run strata qwen3.8 flash next sadly... its 1 generation or model behind the supported one Maybe consider a fast model and great at completing task, and a larger model that plans or thinks well for the entire process ***For the task specifically, it could be anything*** but i do have some in mind researching, visual designing, file management, summarizing something I appreciate any suggestions and solutions. Im still at the planning process but currently im using qwen3.6 35b a3b tiel coder as my main model for chatting
document fetched 2026-11-01 00:38 · canonical https://reddit.com/r/LocalLLaMA/comments/1xk3d0f/light_weight_model_to_execute_task_while_having_a