diff --git a/llm/ggml.go b/llm/ggml.go index 9b6da425..878800f3 100644 --- a/llm/ggml.go +++ b/llm/ggml.go @@ -125,9 +125,9 @@ type Tensor struct { func (t Tensor) blockSize() uint64 { switch t.Kind { - case 0, 1, 24, 25, 26, 27, 28, 31: // F32, F16, I8, I16, I32, I64, F64, BF16 + case 0, 1, 24, 25, 26, 27, 28, 30: // F32, F16, I8, I16, I32, I64, F64, BF16 return 1 - case 2, 3, 8, 9, 20: // Q4_0, Q4_1, Q8_0, Q8_1, IQ4_NL + case 2, 3, 4, 5, 6, 7, 8, 9, 20: // Q4_0, Q4_1, Q5_0, Q5_1, Q8_0, Q8_1, IQ4_NL return 32 default: // All others return 256