matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-10 15:57:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

convert gemma2

Browse Source
This commit is contained in:
Michael Yang
2024-06-28 13:27:05 -07:00
parent beb49eef65
commit 3546bbd08c
13 changed files with 132 additions and 46 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
convert/convert_gemma.go
View File

@@ -44,15 +44,14 @@ func (p *gemma) KV(t *Tokenizer) llm.KV {
}
func (p *gemma) Tensors(ts []Tensor) []llm.Tensor {
var out []llm.Tensor
out := make([]llm.Tensor, 0, len(ts))
for _, t := range ts {
name := p.tensorName(t.Name())
if strings.HasSuffix(name, "_norm.weight") {
if strings.HasSuffix(t.Name(), "_norm.weight") {
t.SetRepacker(p.addOne)
}
out = append(out, llm.Tensor{
Name: name,
Name: t.Name(),
Kind: t.Kind(),
Shape: t.Shape(),
WriterTo: t,
@@ -62,8 +61,8 @@ func (p *gemma) Tensors(ts []Tensor) []llm.Tensor {
return out
}
func (p *gemma) tensorName(n string) string {
return strings.NewReplacer(
func (p *gemma) Replacements() []string {
return []string{
"model.embed_tokens", "token_embd",
"model.norm", "output_norm",
"model.layers", "blk",
@@ -76,8 +75,7 @@ func (p *gemma) tensorName(n string) string {
"mlp.down_proj", "ffn_down",
"mlp.up_proj", "ffn_up",
"post_attention_layernorm", "ffn_norm",
"block_sparse_moe.gate", "ffn_inp",
).Replace(n)
}
}
func (*gemma) addOne(_ string, data []float32, shape []uint64) ([]float32, error) {