root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-30 21:34:36 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

kompute : fix rope_f32 and scale ops (#5008)

Browse Source
This commit is contained in:
Georgi Gerganov 2024-01-18 18:49:39 +02:00 committed by GitHub
parent 02b9bafe29
commit 696faa8660
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 25 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
ggml-kompute.cpp
View File

@ -1540,7 +1540,8 @@ void ggml_vk_graph_compute(struct ggml_kompute_context * ctx, struct ggml_cgraph
} break; } break;
case GGML_OP_SCALE: case GGML_OP_SCALE:
{ {
const float scale = *(const float *) src1->data; float scale; memcpy(&scale, dst->op_params, sizeof(float));
ggml_vk_scale(seq, id_src0, id_dst, off_src0, off_dst, ggml_nelements(dst), scale); ggml_vk_scale(seq, id_src0, id_dst, off_src0, off_dst, ggml_nelements(dst), scale);
} break; } break;
case GGML_OP_UNARY: case GGML_OP_UNARY:

14
kompute-shaders/op_rope_f32.comp
View File

@ -40,16 +40,15 @@ void main() {
} }
} else { } else {
const float inv_ndims = -1.f/pcs.n_dims; const float inv_ndims = -1.f/pcs.n_dims;
for (uint ib = 0; ib < pcs.ne0/pcs.n_dims; ++ib) {
for (uint ic = 0; ic < pcs.n_dims; ic += 2) { for (uint ic = 0; ic < pcs.n_dims; ic += 2) {
const uint cur_rot = ib * pcs.n_dims + ic; const uint cur_rot = ic;
float cos_theta, sin_theta; float cos_theta, sin_theta;
rope_yarn(theta, pcs.freq_scale, corr_dims, cur_rot, pcs.ext_factor, pcs.attn_factor, cos_theta, sin_theta); rope_yarn(theta, pcs.freq_scale, corr_dims, cur_rot, pcs.ext_factor, pcs.attn_factor, cos_theta, sin_theta);
theta *= theta_scale; theta *= theta_scale;
const uint i0 = ib*pcs.n_dims + ic/2; const uint i0 = ic/2;
const uint src = uint((i3*pcs.nb03 + i2*pcs.nb02 + i1*pcs.nb01 + i0*pcs.nb00) / 4) + pcs.inAOff; // Based from in const uint src = uint((i3*pcs.nb03 + i2*pcs.nb02 + i1*pcs.nb01 + i0*pcs.nb00) / 4) + pcs.inAOff; // Based from in
const uint dst_data = uint((i3*pcs.nb3 + i2*pcs.nb2 + i1*pcs.nb1 + i0*pcs.nb0) / 4) + pcs.outOff; // Based from out_ const uint dst_data = uint((i3*pcs.nb3 + i2*pcs.nb2 + i1*pcs.nb1 + i0*pcs.nb0) / 4) + pcs.outOff; // Based from out_
@ -60,6 +59,15 @@ void main() {
out_[dst_data] = x0*cos_theta - x1*sin_theta; out_[dst_data] = x0*cos_theta - x1*sin_theta;
out_[dst_data+pcs.n_dims/2] = x0*sin_theta + x1*cos_theta; out_[dst_data+pcs.n_dims/2] = x0*sin_theta + x1*cos_theta;
} }
for (uint ic = pcs.n_dims; ic < pcs.ne0; ic += 2) {
const uint i0 = ic;
const uint src = uint((i3*pcs.nb03 + i2*pcs.nb02 + i1*pcs.nb01 + i0*pcs.nb00) / 4) + pcs.inAOff; // Based from in
const uint dst_data = uint((i3*pcs.nb3 + i2*pcs.nb2 + i1*pcs.nb1 + i0*pcs.nb0) / 4) + pcs.outOff; // Based from out_
out_[dst_data + 0] = inA[src + 0];
out_[dst_data + 1] = inA[src + 1];
} }
} }
} }