llm: limit generation to 10x context size to avoid run on generations (#3918)

* llm: limit generation to 10x context size to avoid run on generations

* add comment

* simplify condition statement

api/types.go

@@ -396,8 +396,10 @@ func (opts *Options) FromMap(m map[string]interface{}) error {
 func DefaultOptions() Options {
 	return Options{
 		// options set on request to runner
-		NumPredict:       -1,
-		NumKeep:          0,
+		NumPredict: -1,
+
+		// set a minimal num_keep to avoid issues on context shifts
+		NumKeep:          4,
 		Temperature:      0.8,
 		TopK:             40,
 		TopP:             0.9,