thek0tyara/llama-cpp-turboquant
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 4

hparams : add n_embd_inp() to support extended embed (#16928)

Browse source 
* add n_embd_full to support extended embed

* don't change output

* rename to n_embd_inp

* restore n_embd where applicable
This commit is contained in:
Sigbjørn Skjæret 2025-11-07 19:27:58 +01:00 • committed by GitHub
parent 16bcc1259d
commit 9008027aa3
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
9 changed files with 29 additions and 28 deletions
Download patch file Download diff file Expand all files Collapse all files

3
src/models/qwen3vl-moe.cpp
View file

@ -1,9 +1,8 @@
#include "models.h"
llm_build_qwen3vlmoe::llm_build_qwen3vlmoe(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
const int64_t n_embd_full = hparams.n_embd; // main embd + deepstack embds
const size_t n_deepstack_layers = hparams.n_deepstack_layers;
const int64_t n_embd = n_embd_full / (n_deepstack_layers + 1);
const int64_t n_embd = hparams.n_embd;
const int64_t n_embd_head = hparams.n_embd_head_v;
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);

5
src/models/qwen3vl.cpp
View file

@ -1,13 +1,10 @@
#include "models.h"
llm_build_qwen3vl::llm_build_qwen3vl(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
const int64_t n_embd_full = hparams.n_embd; // main embd + deepstack embds
const size_t n_deepstack_layers = hparams.n_deepstack_layers;
const int64_t n_embd = n_embd_full / (n_deepstack_layers + 1);
const int64_t n_embd = hparams.n_embd;
const int64_t n_embd_head = hparams.n_embd_head_v;
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
GGML_ASSERT(n_embd_head == hparams.n_rot);