Skip to content

Get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.

llm: normalise kvct parameter handling (#7926)

Sam539be43640ed
diff --git a/llm/memory.go b/llm/memory.go
index c5d861b6a..384e2dc60 100644
--- a/llm/memory.go
+++ b/llm/memory.go
@@ -129,7 +129,7 @@ func EstimateGPULayers(gpus []discover.GpuInfo, ggml *GGML, projectors []string,
var kvct string
if fa {
- requested := envconfig.KvCacheType()
+ requested := strings.ToLower(envconfig.KvCacheType())
if requested != "" && ggml.SupportsKVCacheType(requested) {
kvct = requested
}
diff --git a/llm/server.go b/llm/server.go
index 23caa9a0a..debdd35e8 100644
--- a/llm/server.go
+++ b/llm/server.go
@@ -225,7 +225,7 @@ func NewLlamaServer(gpus discover.GpuInfoList, model string, ggml *GGML, adapter
fa = false
}
- kvct := envconfig.KvCacheType()
+ kvct := strings.ToLower(envconfig.KvCacheType())
if fa {
slog.Info("enabling flash attention")