jeffbolznv commited on
Commit
271c7e4
·
1 Parent(s): 4be146e

vulkan: fix assertion when qy_needs_dequant (llama/12068)

Browse files
ggml/src/ggml-vulkan/ggml-vulkan.cpp CHANGED
@@ -4183,7 +4183,7 @@ static void ggml_vk_mul_mat_q_f16(ggml_backend_vk_context * ctx, vk_context& sub
4183
  }
4184
  if (qy_needs_dequant) {
4185
  d_Y = ctx->prealloc_y;
4186
- GGML_ASSERT(d_Y->size >= y_sz * ne02 * ne03);
4187
  } else {
4188
  d_Y = d_Qy;
4189
  y_buf_offset = qy_buf_offset;
@@ -4760,7 +4760,7 @@ static void ggml_vk_mul_mat_id_q_f16(ggml_backend_vk_context * ctx, vk_context&
4760
  }
4761
  if (qy_needs_dequant) {
4762
  d_Y = ctx->prealloc_y;
4763
- GGML_ASSERT(d_Y->size >= y_sz * ne02 * ne03);
4764
  } else {
4765
  d_Y = d_Qy;
4766
  y_buf_offset = qy_buf_offset;
 
4183
  }
4184
  if (qy_needs_dequant) {
4185
  d_Y = ctx->prealloc_y;
4186
+ GGML_ASSERT(d_Y->size >= y_sz * ne12 * ne13);
4187
  } else {
4188
  d_Y = d_Qy;
4189
  y_buf_offset = qy_buf_offset;
 
4760
  }
4761
  if (qy_needs_dequant) {
4762
  d_Y = ctx->prealloc_y;
4763
+ GGML_ASSERT(d_Y->size >= y_sz * ne12 * ne13);
4764
  } else {
4765
  d_Y = d_Qy;
4766
  y_buf_offset = qy_buf_offset;