mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2025-11-13 10:57:15 +00:00
mtmd : fix embedding size for image input (#17123)
This commit is contained in:
@@ -182,7 +182,7 @@ int32_t mtmd_helper_decode_image_chunk(
|
|||||||
}
|
}
|
||||||
|
|
||||||
const llama_model * model = llama_get_model(lctx);
|
const llama_model * model = llama_get_model(lctx);
|
||||||
int n_mmproj_embd = llama_model_n_embd(model);
|
int n_mmproj_embd = llama_model_n_embd_inp(model);
|
||||||
int n_pos_per_embd = mtmd_decode_use_mrope(ctx) ? 4 : 1;
|
int n_pos_per_embd = mtmd_decode_use_mrope(ctx) ? 4 : 1;
|
||||||
|
|
||||||
int32_t n_tokens = mtmd_input_chunk_get_n_tokens(chunk);
|
int32_t n_tokens = mtmd_input_chunk_get_n_tokens(chunk);
|
||||||
|
|||||||
Reference in New Issue
Block a user