adventists-ai/Sophon-Realtime-32B-Tag
Any-to-Any • Updated • 1.01k • 14
Open full-duplex voice models: Tag, Flash and E2E builds in 32B (one 80 GB GPU) and 4B (one 24/32 GB GPU), plus Flash-4bit for Mac.
Note Recommended: expressive voice, one 80 GB GPU
Note Fastest, most users per GPU
Note Fastest first audio, expressive
Note New in 1.4: one 24 GB GPU
Note New in 1.4: one 32 GB GPU
Note New in 1.4: one 32 GB GPU
Note Offline on a Mac (MLX)
Note Mac (MLX), 16 GB+
Note one 24 GB GPU (4-bit brain)
Note Unquantized BF16 thinker for hardware without FP8