OpenCode setup with Local models

Notes on this, will publish more as I go: opencode.json { “$schema”: “https://opencode.ai/config.json”, “provider”: { “llama.cpp”: { “npm”: “@ai-sdk/openai-compatible”, “name”: “llama.cpp (local)”, “options”: { “baseURL”: “http://localhost:8000/v1” }, “models”: { “unsloth/Qwen3.5-27B-GGUF:Q6_K”: { “name”: “unsloth/Qwen3.5-27B-GGUF:Q6_K”, “tool_call”: true, “reasoning”: true, “limit”: { “context”: 262144, “output”: 16384 }, “modalities”: { “input”: [ “text”, “image” ], “output”: [ “text” ]…

ROCm with two different GPUs? How hard can it be?

I spent a couple of days this week trying to get vllm to work with more than one GPU.. not the easy way, but two different architecture GPU’s. One gfx1100 (Radeon 7900 XTX) and one gfx908 (Radeon Instinct mi100) – that’s one RDNA3 and one CDNA GPU. Yeah, I’m pushing water uphill here. So now,…

Cooling the mi100 in a desktop case

I’ve had this AMD Instinct mi100 sitting around for waay too long unused.. I tried a couple of different cooling options a while ago, but neither worked satisfactorily, and so it went back on the shelf pending a bit more free time. Of course, “free time” is a myth, I should be more honest and…