highperfocused/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2025-04-01 16:38:50 +02:00
Code Issues Packages Projects Releases Wiki Activity

default head_kv to 1

Browse Source
This commit is contained in:
Michael Yang 2024-04-02 16:37:59 -07:00
parent a039e383cd
commit 90f071c658
1 changed files with 6 additions and 6 deletions

12
llm/ggml.go

@ -148,15 +148,15 @@ func (kv KV) HeadCount() uint64 {
}
func (kv KV) HeadCountKV() uint64 {
return kv.u64(fmt.Sprintf("%s.attention.head_count_kv", kv.Architecture()))
if headCountKV := kv.u64(fmt.Sprintf("%s.attention.head_count_kv", kv.Architecture())); headCountKV > 0 {
return headCountKV
}
return 1
}
func (kv KV) GQA() uint64 {
if headCountKV := kv.HeadCountKV(); headCountKV > 0 {
return kv.HeadCount() / headCountKV
}
return 0
return kv.HeadCount() / kv.HeadCountKV()
}
func (kv KV) EmbeddingLength() uint64 {