CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-30 08:42:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

qwen2vl: use llama_batch_ext_set_pos

Browse Source
This commit is contained in:
Xuan Son Nguyen
2025-03-14 11:25:36 +01:00
parent ba79369615
commit a363251fac
3 changed files with 18 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
examples/llava/qwen2vl-cli.cpp
View File

@@ -66,18 +66,11 @@ static bool qwen2vl_eval_image_embed(llama_context * ctx_llama, const struct lla
memcpy(&batch_mrope_pos[n_eval * 2], &mrope_pos[img_tokens * 2 + processed], n_eval * sizeof(llama_pos));
memcpy(&batch_mrope_pos[n_eval * 3], &mrope_pos[img_tokens * 3 + processed], n_eval * sizeof(llama_pos));
// TODO: move this to llama_batch_ext API
llama_batch batch = {
int32_t(n_eval), // n_tokens
nullptr, // token
(image_embed->embed+i*n_embd), // embed
batch_mrope_pos.data(), // pos
nullptr, // n_seq_id
nullptr, // seq_id
nullptr, // logits
};
float * batch_embd = image_embed->embed+i*n_embd;
llama_batch_ext_ptr batch(llama_batch_ext_init_from_embd(batch_embd, n_eval, n_embd, 0, 0));
llama_batch_ext_set_pos(batch.get(), batch_mrope_pos.data(), n_eval);
if (llama_decode(ctx_llama, batch)) {
if (llama_decode_ext(ctx_llama, batch.get())) {
LOG_ERR("%s : failed to eval\n", __func__);
return false;
}