github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-03-09 07:16:38 -05:00
Code Issues Packages Projects Releases Wiki Activity

convert(gptoss): mxfp4 to ggml layout to avoid jit conversion (#12018)

Browse Source 
* convert: return bytes written

* ggml flavor mxfp4

* simplify jit conversion

* comment
This commit is contained in:
Michael Yang
2025-08-26 16:41:02 -07:00
committed by GitHub
parent 86834a2797
commit 59412fbb43
6 changed files with 49 additions and 58 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
fs/ggml/ggml.go
View File

@@ -290,24 +290,24 @@ func (t Tensor) blockSize() uint64 {
func (t TensorType) BlockSize() uint64 {
switch t {
case
0, // F32
1, // F16
24, // I8
25, // I16
26, // I32
27, // I64
28, // F64
30: // BF16
TensorTypeF32,
TensorTypeF16,
TensorTypeI8,
TensorTypeI16,
TensorTypeI32,
TensorTypeI64,
TensorTypeF64,
TensorTypeBF16:
return 1
case
2, // Q4_0
3, // Q4_1
4, // MXFP4
6, // Q5_0
7, // Q5_1
8, // Q8_0
9, // Q8_1
20: // IQ4_NL
TensorTypeQ4_0,
TensorTypeQ4_1,
TensorTypeQ5_0,
TensorTypeQ5_1,
TensorTypeQ8_0,
TensorTypeQ8_1,
tensorTypeIQ4_NL,
4, TensorTypeMXFP4:
return 32
default:
return 256
@@ -330,8 +330,6 @@ func (t TensorType) TypeSize() uint64 {
return 2 + blockSize/2
case TensorTypeQ4_1:
return 2 + 2 + blockSize/2
case TensorTypeMXFP4, 39:
return 1 + blockSize/2
case TensorTypeQ5_0:
return 2 + 4 + blockSize/2
case TensorTypeQ5_1:
@@ -382,6 +380,8 @@ func (t TensorType) TypeSize() uint64 {
return blockSize/8 + blockSize/16 + blockSize/32
case TensorTypeBF16:
return 2
case 4, TensorTypeMXFP4:
return 1 + blockSize/2
default:
return 0
}

11
fs/ggml/type.go
View File

@@ -146,8 +146,6 @@ func (ftype FileType) ToTensorType() TensorType {
return TensorTypeQ4_0
case fileTypeQ4_1:
return TensorTypeQ4_1
case fileTypeMXFP4:
return TensorTypeMXFP4 // Formerly unused tensorTypeQ4_2
case FileTypeQ8_0:
return TensorTypeQ8_0
case fileTypeQ5_0:
@@ -176,6 +174,8 @@ func (ftype FileType) ToTensorType() TensorType {
return TensorTypeQ2_K
case FileTypeBF16:
return TensorTypeBF16
case fileTypeMXFP4:
return TensorTypeMXFP4
default:
slog.Warn("unsupported file type", "type", ftype)
return 0 // F32
@@ -191,8 +191,8 @@ const (
TensorTypeF16
TensorTypeQ4_0
TensorTypeQ4_1
TensorTypeMXFP4 // Formerly unused tensorTypeQ4_2
tensorTypeQ4_3 // unused by GGML
tensorTypeQ4_2
tensorTypeQ4_3 // unused by GGML
TensorTypeQ5_0
TensorTypeQ5_1
TensorTypeQ8_0
@@ -226,6 +226,7 @@ const (
tensorTypeIQ4_NL_4_4 // unused by GGML
tensorTypeIQ4_NL_4_8 // unused by GGML
tensorTypeIQ4_NL_8_8 // unused by GGML
TensorTypeMXFP4
)
// ParseFileType parses the provided GGUF file type
@@ -318,7 +319,7 @@ func (t TensorType) String() string {
return "F64"
case TensorTypeBF16:
return "BF16"
case TensorTypeMXFP4:
case 4, TensorTypeMXFP4:
return "MXFP4"
default:
return "unknown"