Use unsloth BF16 GGUF to quantize IQ1_M/S. Blk.46 is not being used in llama.cpp therefore the weights of blk.46 are quantized to TQ1_0 to have minimum memory allocation.


Added MXFP4 version:

  1. MXFP4: Embedding, Output are kept with Q6_K. The attn layers use IQ4_XS. All ffn expert layers including shared experts are quantized to SOTA MXFP4.
  2. MXFP4 Max: Embedding, Output and attn layers are kept with Q6_K. First layer uses full precision. The rest of ffn expert layers are quantized to SOTA MXFP4. The shared experts weights keep BF16.
Downloads last month
2,792
GGUF
Model size
110B params
Architecture
glm4moe
Hardware compatibility
Log In to view the estimation

1-bit

2-bit

3-bit

4-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for lovedheart/GLM-4.5-Air-GGUF-IQ1_M

Quantized
(38)
this model