From 3a848fc48c1fbe3d007f0122168fdce6085f14a8 Mon Sep 17 00:00:00 2001 From: Kawrakow Date: Mon, 12 Jan 2026 11:55:01 +0200 Subject: [PATCH] WIP - Qwen3-MoE (and hopefully all others) working But when I say here and in the previous commit "working", I mean PP is working. TG is still broken. --- ggml/src/ggml-cuda.cu | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-) diff --git a/ggml/src/ggml-cuda.cu b/ggml/src/ggml-cuda.cu index cdf4893b..cdda6113 100644 --- a/ggml/src/ggml-cuda.cu +++ b/ggml/src/ggml-cuda.cu @@ -2689,13 +2689,9 @@ static int ggml_cuda_moe_up_gate_unary(ggml_backend_cuda_context & ctx, ggml_ten ggml_swiglu_oai_cuda_f32((const float *)dst_up_gate_contiguous.get() + dst->ne[0], (const float *)dst_up_gate_contiguous.get(), (float *)dst->data, ggml_nelements(dst), dst->ne[0], src0_1->ne[1], src0_1->ne[1], 1.702f, 7.0f, stream); - //ggml_swiglu_oai_cuda_f32((const float *)dst_up_gate_contiguous.get(), (const float *)dst_up_gate_contiguous.get() + dst->ne[0], - // (float *)dst->data, ggml_nelements(dst), dst->ne[0], src0_1->ne[0], src0_1->ne[0], - // 1.702f, 7.0f, stream); } else { ggml_fused_mul_unary(ctx, (ggml_unary_op)dst->op_params[0], ggml_nelements(dst), dst->ne[0], - (const float *)dst_up_gate_contiguous.get(), - (float *)dst->data); + (const float *)dst_up_gate_contiguous.get(), (float *)dst->data); } } CUDA_CHECK(cudaGetLastError());