Qwen3.8-27B in EXL3 at six sizes, 2.5 to 6.0 bits per weight, and the EXL3 drafters TandemLLM uses for speculative decoding.
-
0xBakeer/TandemLLM-Qwen3.8-27B-EXL3-2.5
Image-Text-to-Text • 6B • Updated • 2 -
0xBakeer/TandemLLM-Qwen3.8-27B-EXL3-3.0
Image-Text-to-Text • 7B • Updated • 2 -
0xBakeer/TandemLLM-Qwen3.8-27B-EXL3-3.5
Image-Text-to-Text • 8B • Updated • 2 -
0xBakeer/TandemLLM-Qwen3.8-27B-EXL3-4.0
Image-Text-to-Text • 8B • Updated • 2