highperfocused/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2025-09-01 01:02:59 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #3466 from ollama/mxyng/head-kv

Browse Source 
default head_kv to 1
This commit is contained in:
Michael Yang
2024-04-03 10:41:00 -07:00
committed by GitHub
parent 7d05a6ee8f 90f071c658
commit 4f895d633f
1 changed files with 6 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
llm/ggml.go
View File

@@ -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 {