diff --git a/sgemm.cpp b/sgemm.cpp index 3c768edc922108..df7419a5df66fb 100644 --- a/sgemm.cpp +++ b/sgemm.cpp @@ -1243,6 +1243,10 @@ class MixMul { assert(params->ith < params->nth); assert(plan->type == GGML_TYPE_I32); + // doesn't help for token generation + if (n < 2) + return false; + // supported types if (result->type != GGML_TYPE_F32) return false;