ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-01-26 17:20:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

Port mdmd from mainline + Qwen2/2.5-VL support (#798)

Browse Source 
* Add mtmd: the beginning

* Add mtmd: mtmd.cpp compiles

* Add mtmd: clip initialization compiles

* Add mtmd: clip.cpp compiles

* Add mtmd: builds successfully

* Add CPU implementation for GGML_OP_GLU

* Add CUDA implementation for GGML_OP_GLU

* Add CPU implementation for GGML_OP_CONV_2D and GGML_OP_CONV_2D_DW

* Add CUDA implementation for GGML_OP_CONV_2D and GGML_OP_CONV_2D_DW

* Add mtmd: refresh CPU rope

* Add mtmd: refresh CUDA rope

* Add mtmd: add Qwen2-VL

* Add mtmd: Qwen2.5-VL text seems to work with this change

* Add mtmd: fix swiglu

* Add mtmd: use LOG_TEE so generated tokens show up in terminal

* Add mtmd: do not attempt to load a GPU backend if none are available

* GLU, not GPU

* Fix typo

* Fix new/free mismatch

* LOG stuff

* Add mtmd: this fixes gibberish on second image

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-09-27 08:45:29 +02:00
committed by GitHub
parent 7d8d232896
commit c1a0e15377
51 changed files with 115141 additions and 432 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
include/llama.h
View File

@@ -117,13 +117,12 @@ extern "C" {
// LLAMA_VOCAB_PRE_TYPE_KIMI_K2 = 38, //llama.cpp lists this as 37
//};
// note: these values should be synchronized with ggml_rope
// TODO: maybe move this enum to ggml.h (ggml_rope_type)
enum llama_rope_type {
LLAMA_ROPE_TYPE_NONE = -1,
LLAMA_ROPE_TYPE_NORM = 0,
LLAMA_ROPE_TYPE_NEOX = 2,
LLAMA_ROPE_TYPE_GLM = 4,
LLAMA_ROPE_TYPE_NONE = -1,
LLAMA_ROPE_TYPE_NORM = 0,
LLAMA_ROPE_TYPE_NEOX = GGML_ROPE_TYPE_NEOX,
LLAMA_ROPE_TYPE_MROPE = GGML_ROPE_TYPE_MROPE,
LLAMA_ROPE_TYPE_VISION = GGML_ROPE_TYPE_VISION,
};
enum llama_token_type { //TODO: remove, required until per token attributes are available from GGUF file
@@ -581,6 +580,14 @@ extern "C" {
LLAMA_API int32_t llama_n_embd (const struct llama_model * model);
LLAMA_API int32_t llama_n_layer (const struct llama_model * model);
// Compat
static int32_t llama_model_n_embd(const struct llama_model * model) { return llama_n_embd(model); }
LLAMA_API bool llama_vocab_get_add_bos(const struct llama_vocab * vocab);
LLAMA_API bool llama_vocab_get_add_eos(const struct llama_vocab * vocab);
LLAMA_API int32_t llama_vocab_n_tokens(const struct llama_vocab * vocab);
LLAMA_API llama_token llama_vocab_bos(const struct llama_vocab * vocab);
LLAMA_API llama_token llama_vocab_eos(const struct llama_vocab * vocab);
// Get the model's RoPE frequency scaling factor
LLAMA_API float llama_rope_freq_scale_train(const struct llama_model * model);
@@ -1018,6 +1025,7 @@ extern "C" {
// Check if the token is supposed to end generation (end-of-generation, eg. EOS, EOT, etc.)
LLAMA_API bool llama_token_is_eog(const struct llama_model * model, llama_token token);
LLAMA_API bool llama_vocab_is_eog(const struct llama_vocab * vocab, llama_token token);
// Identify if Token Id is a control token or a render-able token
LLAMA_API bool llama_token_is_control(const struct llama_model * model, llama_token token);
@@ -1061,6 +1069,14 @@ extern "C" {
int32_t n_tokens_max,
bool add_special,
bool parse_special);
LLAMA_API int32_t llama_vocab_tokenize(
const struct llama_vocab * vocab,
const char * text,
int32_t text_len,
llama_token * tokens,
int32_t n_tokens_max,
bool add_special,
bool parse_special);
// Token Id -> Piece.
// Uses the vocabulary in the provided context.
@@ -1074,6 +1090,13 @@ extern "C" {
int32_t length,
int32_t lstrip,
bool special);
LLAMA_API int32_t llama_vocab_token_to_piece(
const struct llama_vocab * vocab,
llama_token token,
char * buf,
int32_t length,
int32_t lstrip,
bool special);
/// @details Convert the provided tokens into text (inverse of llama_tokenize()).
/// @param text The char pointer must be large enough to hold the resulting text.