[throwdbaaway]

For 27B, just get a used 3090 and hop on to r/LocalLLaMA. You can run a 4bpw quant at full context with Q8 KV cache.