tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-07-27 03:33:46 -04:00
Code Issues Packages Projects Releases Wiki Activity

ggml : hide ggml_object, ggml_cgraph, ggml_hash_set

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2024-09-10 16:38:06 +03:00
parent 00ba2ff781
commit c8a3f291fe
13 changed files with 131 additions and 92 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
examples/llava/clip.cpp
View File

@ -2449,7 +2449,7 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
ggml_backend_graph_compute(ctx->backend, gf);
// the last node is the embedding tensor
struct ggml_tensor * embeddings = gf->nodes[gf->n_nodes - 1];
struct ggml_tensor * embeddings = ggml_graph_node(gf, -1);
// copy the embeddings to the location passed by the user
ggml_backend_tensor_get(embeddings, vec, 0, ggml_nbytes(embeddings));

2
examples/llava/llava.cpp
View File

@ -184,7 +184,7 @@ static bool clip_llava_handle_patches(clip_ctx * ctx_clip, std::vector<float *>
// ggml_tensor_printf(flatten,"flatten",__LINE__,false,false);
ggml_build_forward_expand(gf, flatten);
ggml_graph_compute_with_ctx(model.ctx, gf, 1);
struct ggml_tensor* result = gf->nodes[gf->n_nodes - 1];
struct ggml_tensor* result = ggml_graph_node(gf, -1);
memcpy(image_embd_out, image_embd_v[0], clip_embd_nbytes(ctx_clip)); // main image as global context
// append without newline tokens (default behavior in llava_arch when not using unpad ):