KookiesXy/Muse-Glimmer-30B-ternary-quants

Muse Glimmer 30B: extreme quantization experiments

Two artifacts built from the official BF16 weights:

• —gguf/Muse-Glimmer-30B-IQ1_S.gguf: usable llama.cpp IQ1_S (~1.5-bit) quantized with an importance matrix.
• —gguf/Muse-Glimmer-30B-TQ1_0.gguf: usable llama.cpp ternarized 1.69-bpw GGUF.
• —ternary/: experimental exact ternary {-1,0,+1} matrix-weight representation. It needs a custom runtime and has not been accuracy-tuned.