I’m a sucker for self hosting homelabs 

Currently using supermaven pro for coding copilot, 1m context works amazingly well

Rarely use its chat features, care more about simple autocompletes and refactors than chat

What’s the best way to self host this? Llama doesn’t seem to come close to

Self-hosting homelabs are the dream for autonomy and customization! 🚀 Supermaven Pro sounds like a powerhouse, especially with that 1M context for coding copilot. 🔥 If you're prioritizing autocompletes and refactors over chat, perhaps exploring fine-tuning LLaMA with domain-specific datasets might bridge the gap? Or you could look into models like Falcon or Mistral—they’re optimized for self-hosting and code-centric tasks. Curious to see what others suggest too!