OSSForks/ollama
1
0
Fork 0
mirror of https://github.com/zebrajr/ollama.git synced 2025-12-06 00:19:51 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

comment split

Browse Source
This commit is contained in:
Michael Yang 2025-10-09 16:13:03 -07:00 committed by Michael Yang
parent e0cd511661
commit 629db9dc43
1 changed files with 4 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
convert/convert_gptoss.go
View File

@ -86,6 +86,8 @@ func (m *gptossModel) Tensors(ts []Tensor) []*ggml.Tensor {
mxfp4s[name].scales = t
}
} else if strings.HasSuffix(t.Name(), "gate_up_exps.bias") {
// gate_up_exps is interleaved, need to split into gate_exps and up_exps
// e.g. gate_exps, up_exps = gate_up_exps[:, 0::2, ...], gate_up_exps[:, 1::2, ...]
out = append(out, slices.Collect(splitDim(t, 1,
split{
Replacer: strings.NewReplacer("gate_up_exps", "gate_exps"),
@ -116,6 +118,8 @@ func (m *gptossModel) Tensors(ts []Tensor) []*ggml.Tensor {
WriterTo: mxfp4,
})
} else if strings.Contains(name, "ffn_gate_up_exps") {
// gate_up_exps is interleaved, need to split into gate_exps and up_exps
// e.g. gate_exps, up_exps = gate_up_exps[:, 0::2, ...], gate_up_exps[:, 1::2, ...]
out = append(out, &ggml.Tensor{
Name: strings.Replace(name, "gate_up", "gate", 1) + ".weight",
Kind: uint32(ggml.TensorTypeMXFP4),