From 3252afb32363230a6e0a3356c75c526049913a89 Mon Sep 17 00:00:00 2001 From: 0cc4m Date: Sun, 21 Jul 2024 10:58:05 +0200 Subject: [PATCH] Fix Vulkan DeepSeek-Coder-V2-Lite MoE support --- ggml/src/ggml-vulkan.cpp | 4 +--- ggml/src/vulkan-shaders/mul_mm.comp | 2 +- 2 files changed, 2 insertions(+), 4 deletions(-) diff --git a/ggml/src/ggml-vulkan.cpp b/ggml/src/ggml-vulkan.cpp index ef728b56f..6bcd81a7b 100644 --- a/ggml/src/ggml-vulkan.cpp +++ b/ggml/src/ggml-vulkan.cpp @@ -3451,7 +3451,7 @@ static void ggml_vk_mul_mat_id_q_f16(ggml_backend_vk_context * ctx, vk_context * const uint64_t nei0 = ids->ne[0]; const uint64_t nei1 = ids->ne[1]; - GGML_ASSERT(nei0 * nei1 <= 2048); + GGML_ASSERT(nei0 * nei1 <= 3072); const uint32_t nbi1 = ids->nb[1]; const uint32_t nbi2 = ids->nb[2]; @@ -3463,8 +3463,6 @@ static void ggml_vk_mul_mat_id_q_f16(ggml_backend_vk_context * ctx, vk_context * const uint64_t n_as = ne02; - GGML_ASSERT(n_as <= 8); - ggml_tensor_extra_gpu * extra = (ggml_tensor_extra_gpu *) dst->extra; ggml_tensor_extra_gpu * extra_src0 = (ggml_tensor_extra_gpu *) src0->extra; ggml_tensor_extra_gpu * extra_src1 = (ggml_tensor_extra_gpu *) src1->extra; diff --git a/ggml/src/vulkan-shaders/mul_mm.comp b/ggml/src/vulkan-shaders/mul_mm.comp index 56e96a500..5fe9d5241 100644 --- a/ggml/src/vulkan-shaders/mul_mm.comp +++ b/ggml/src/vulkan-shaders/mul_mm.comp @@ -71,7 +71,7 @@ shared FLOAT_TYPE buf_a[BM * (BK+1)]; shared FLOAT_TYPE buf_b[BN * (BK+1)]; #ifdef MUL_MAT_ID -shared u16vec2 row_ids[2048]; +shared u16vec2 row_ids[3072]; #endif void main() {