BeeLlama v0.3.0 and v0.3.1 are here! Big architectural update to align the fork with upstream llama.cpp and integrate all its additions like MTP and Gemma 4 12B support, while also updating DFlash to handle complex configurations like multi-slot and multi-GPU. Now also recommended by club-3090 ! Thanks to noonghunna for inviting Bee to the club and for their help with testing v0.3.0 on a multi-GPU setup. Not quite a pegasus, but close enough. GitHub | Qwen 3.6 27B Quick Sta...