Skip to content

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

scheduler: default parallel=1 for qwen3next/lfm (#14103)

Jeffrey Morgane36f389e8281
diff --git a/server/sched.go b/server/sched.go
index 3aa9969a0..67af5e8ce 100644
--- a/server/sched.go
+++ b/server/sched.go
@@ -417,9 +417,9 @@ func (s *Scheduler) load(req *LlmRequest, f *ggml.GGML, systemInfo ml.SystemInfo
numParallel = 1
}
- // `mllama`, `qwen3vl`, and `qwen3vlmoe` are snowflakes and uses an encoder cache which cannot be used with num_parallel > 1
+ // Some architectures are not safe with num_parallel > 1.
// ref: https://github.com/ollama/ollama/issues/4165
- if slices.Contains([]string{"mllama", "qwen3vl", "qwen3vlmoe"}, req.model.Config.ModelFamily) && numParallel != 1 {
+ if slices.Contains([]string{"mllama", "qwen3vl", "qwen3vlmoe", "qwen3next", "lfm2", "lfm2moe"}, req.model.Config.ModelFamily) && numParallel != 1 {
numParallel = 1
slog.Warn("model architecture does not currently support parallel requests", "architecture", req.model.Config.ModelFamily)
}