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

convert safetensor adapters into GGUF (#6327)

Browse Source
This commit is contained in:
Patrick Devine
2024-08-23 11:29:56 -07:00
committed by GitHub
parent 7a1e1c1caf
commit 0c819e167b
16 changed files with 697 additions and 101 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
convert/convert_gemma2.go
View File

@@ -4,15 +4,15 @@ import (
"github.com/ollama/ollama/llm"
)
type gemma2 struct {
gemma
type gemma2Model struct {
gemmaModel
SlidingWindow uint32 `json:"sliding_window"`
AttentionLogitSoftcap float32 `json:"attn_logit_softcapping"`
FinalLogitSoftcap float32 `json:"final_logit_softcapping"`
}
func (p *gemma2) KV(t *Tokenizer) llm.KV {
kv := p.Parameters.KV(t)
func (p *gemma2Model) KV(t *Tokenizer) llm.KV {
kv := p.ModelParameters.KV(t)
kv["general.architecture"] = "gemma2"
kv["gemma2.context_length"] = p.MaxPositionEmbeddings
kv["gemma2.embedding_length"] = p.HiddenSize
@@ -33,9 +33,9 @@ func (p *gemma2) KV(t *Tokenizer) llm.KV {
return kv
}
func (p *gemma2) Replacements() []string {
func (p *gemma2Model) Replacements() []string {
return append(
p.gemma.Replacements(),
p.gemmaModel.Replacements(),
"post_attention_layernorm", "post_attention_norm",
"pre_feedforward_layernorm", "ffn_norm",
"post_feedforward_layernorm", "post_ffw_norm",