tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-08-17 21:51:27 -04:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'master' into cisc/jina-embeddings-v3

Browse Source
This commit is contained in:
Sigbjørn Skjæret
2025-07-20 20:16:05 +02:00
committed by GitHub
parent e092776f9b 2be60cbc27
commit 80094033fa
154 changed files with 41058 additions and 3977 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
common/common.h
View File

@@ -84,6 +84,7 @@ enum llama_example {
LLAMA_EXAMPLE_LOOKUP,
LLAMA_EXAMPLE_PARALLEL,
LLAMA_EXAMPLE_TTS,
LLAMA_EXAMPLE_DIFFUSION,
LLAMA_EXAMPLE_COUNT,
};
@@ -180,7 +181,8 @@ struct common_params_sampling {
std::vector<common_grammar_trigger> grammar_triggers; // optional triggers (for lazy grammars)
std::set<llama_token> preserved_tokens;
std::vector<llama_logit_bias> logit_bias; // logit biases to apply
std::vector<llama_logit_bias> logit_bias; // logit biases to apply
std::vector<llama_logit_bias> logit_bias_eog; // pre-calculated logit biases for EOG tokens
// print the parameters into a string
std::string print() const;
@@ -220,6 +222,14 @@ struct common_params_vocoder {
bool use_guide_tokens = false; // enable guide tokens to improve TTS accuracy // NOLINT
};
struct common_params_diffusion {
int32_t steps = 64; // number of diffusion steps
float eps = 1e-3f; // epsilon for timesteps
int32_t algorithm = 0; // diffusion algorithm (0=ORIGIN, 1=MASKGIT_PLUS, 2=TOPK_MARGIN, 3=ENTROPY)
float alg_temp = 0.0f; // algorithm temperature
bool visual_mode = false; // show progressive diffusion on screen
};
enum common_reasoning_format {
COMMON_REASONING_FORMAT_NONE,
COMMON_REASONING_FORMAT_DEEPSEEK_LEGACY, // Extract thinking tag contents and return as `message.reasoning_content`, or leave inline in <think> tags in stream mode
@@ -271,6 +281,7 @@ struct common_params {
struct common_params_sampling sampling;
struct common_params_speculative speculative;
struct common_params_vocoder vocoder;
struct common_params_diffusion diffusion;
struct common_params_model model;
@@ -333,6 +344,7 @@ struct common_params {
bool no_perf = false; // disable performance metrics
bool ctx_shift = true; // context shift on inifinite text generation
bool swa_full = false; // use full-size SWA cache (https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
bool kv_unified = false; // enable unified KV cache
bool input_prefix_bos = false; // prefix BOS to user inputs, preceding input_prefix
bool use_mmap = true; // use mmap for faster loads
@@ -373,6 +385,7 @@ struct common_params {
std::string hostname = "127.0.0.1";
std::string public_path = ""; // NOLINT
std::string api_prefix = ""; // NOLINT
std::string chat_template = ""; // NOLINT
bool use_jinja = false; // NOLINT
bool enable_chat_template = true;
@@ -524,6 +537,7 @@ static bool string_starts_with(const std::string & str,
// While we wait for C++20's std::string::ends_with...
bool string_ends_with(const std::string_view & str, const std::string_view & suffix);
bool string_remove_suffix(std::string & str, const std::string_view & suffix);
size_t string_find_partial_stop(const std::string_view & str, const std::string_view & stop);
bool string_parse_kv_override(const char * data, std::vector<llama_model_kv_override> & overrides);