Add Q2_0 quantization: type definition and CPU backend (#24448)

gguf-py/gguf/constants.py

@@ -4533,6 +4533,7 @@ class GGMLQuantizationType(IntEnum):
     MXFP4   = 39
     NVFP4   = 40
     Q1_0    = 41
+    Q2_0    = 42
 
 
 class ExpertGatingFuncType(IntEnum):
@@ -4588,6 +4589,7 @@ class LlamaFileType(IntEnum):
     MOSTLY_MXFP4_MOE     = 38  # except 1d tensors
     MOSTLY_NVFP4         = 39  # except 1d tensors
     MOSTLY_Q1_0          = 40  # except 1d tensors
+    MOSTLY_Q2_0          = 41  # except 1d tensors
 
     GUESSED              = 1024  # not specified in the model file
 
@@ -4713,6 +4715,7 @@ GGML_QUANT_SIZES: dict[GGMLQuantizationType, tuple[int, int]] = {
     GGMLQuantizationType.MXFP4:   (32, 1 + 16),
     GGMLQuantizationType.NVFP4:   (64, 4 + 32),
     GGMLQuantizationType.Q1_0:    (128, 2 + 16),
+    GGMLQuantizationType.Q2_0:    (64, 2 + 16),
 }