mirrors/ollama-for-amd
1
0
Fork 0
mirror of https://github.com/likelovewant/ollama-for-amd.git synced 2025-12-25 07:58:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix conversion

Browse Source
This commit is contained in:
Patrick Devine
2025-03-07 14:06:10 -08:00
committed by Michael Yang
parent 0df1800436
commit c62861f4fa
3 changed files with 57 additions and 42 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
convert/convert.go
View File

@@ -190,8 +190,8 @@ func ConvertModel(fsys fs.FS, ws io.WriteSeeker) error {
conv = &gemmaModel{}
case "Gemma2ForCausalLM":
conv = &gemma2Model{}
case "Gemma3ForConditionalGeneration":
conv = &gemma3Model{}
case "Gemma3ForCausalLM", "Gemma3ForConditionalGeneration":
conv = &gemma3Model{Architecture: p.Architectures[0]}
case "Phi3ForCausalLM":
conv = &phi3Model{}
case "Qwen2ForCausalLM":
@@ -226,6 +226,9 @@ func ConvertModel(fsys fs.FS, ws io.WriteSeeker) error {
}
switch {
case vocabSize == 0:
slog.Warn("vocabulary size was not explicitly set by the model", "default size", len(t.Vocabulary.Tokens))
vocabSize = len(t.Vocabulary.Tokens)
case vocabSize > len(t.Vocabulary.Tokens):
slog.Warn("vocabulary is smaller than expected, padding with dummy tokens", "expect", vocabSize, "actual", len(t.Vocabulary.Tokens))
for i := range vocabSize - len(t.Vocabulary.Tokens) {