mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 30 Packages Projects Releases Wiki Activity

spec : add self‑speculative decoding (no draft model required) + refactor (#18471)

Browse Source 
* server: introduce self-speculative decoding

* server: moved self-call into speculative.cpp

* can_speculate() includes self-speculation

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* server: can_speculate() tests self-spec

* server: replace can_speculate() with slot.can_speculate()

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

* common: use %zu format specifier for size_t in logging

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

* server: can_speculate() requires a task instance

* common: ngram map, config self-speculative decoding

* common: add enum common_speculative_type

* common: add vector of speculative states

* common: add option --spec-draftless

* server: cleanup (remove slot.batch_spec, rename)

* common: moved self-spec impl to ngram-map

* common: cleanup (use common_speculative_state_draft)

* spec : refactor

* cont : naming

* spec: remove --spec-config

* doc: (draftless) speculative decoding

* common: print performance in spec decoding

* minor : cleanup

* common : better names

* minor : cleanup + fix build

* minor: comments

* CODEOWNERS: add common/ngram-map.* (#18471)

* common : rename speculative.draftless_type -> speculative.type

* ngram-map : fix uninitialized values

* ngram-map : take into account the input can become shorter

* ngram-map : revert len check for now

* arg : change `--spec-draftless` -> `--spec-type`

* spec : add common_speculative_state::accept()

* spec : refactor + add common_speculative_begin()

* spec : fix begin() call with mtmd

* spec : additional refactor + remove common_speculative_params

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
This commit is contained in:
Sascha Rogmann
2026-01-28 18:42:42 +01:00
committed by GitHub
parent ebf5725870
commit 72d3b1898a
19 changed files with 1649 additions and 444 deletions
Download Patch File Download Diff File Expand all files Collapse all files
common/common.h
+46 -18
View File
@@ -164,6 +164,16 @@ enum common_params_sampling_config : uint64_t {
COMMON_PARAMS_SAMPLING_CONFIG_MIROSTAT_ETA = 1 << 11,
};
enum common_speculative_type {
COMMON_SPECULATIVE_TYPE_NONE, // no speculative decoding
COMMON_SPECULATIVE_TYPE_DRAFT, // draft model
COMMON_SPECULATIVE_TYPE_EAGLE3, // eagle draft model
COMMON_SPECULATIVE_TYPE_NGRAM_SIMPLE, // simple self-speculative decoding
COMMON_SPECULATIVE_TYPE_NGRAM_MAP_K, // self-speculative decoding with n-gram keys only
COMMON_SPECULATIVE_TYPE_NGRAM_MAP_K4V, // self-speculative decoding with n-gram keys and 4 m-gram values
COMMON_SPECULATIVE_TYPE_NGRAM_CACHE, // self-speculative decoding with 3-level n-gram cache
COMMON_SPECULATIVE_TYPE_COUNT // number of types, unknown type
};
// sampling parameters
struct common_params_sampling {
@@ -243,16 +253,35 @@ struct common_params_model {
};
struct common_params_speculative {
std::vector<ggml_backend_dev_t> devices; // devices to use for offloading
common_speculative_type type = COMMON_SPECULATIVE_TYPE_NONE; // type of speculative decoding
int32_t n_ctx = 0; // draft context size
int32_t n_max = 16; // maximum number of tokens to draft during speculative decoding
int32_t n_min = 0; // minimum number of draft tokens to use for speculative decoding
int32_t n_gpu_layers = -1; // number of layers to store in VRAM for the draft model (-1 - use default)
float p_split = 0.1f; // speculative decoding split probability
float p_min = 0.75f; // minimum speculative decoding probability (greedy)
std::vector<std::pair<std::string, std::string>> replacements; // main to speculative model replacements
std::vector<llama_model_tensor_buft_override> tensor_buft_overrides;
// general-purpose speculative decoding parameters
int32_t n_max = 16; // maximum number of tokens to draft during speculative decoding
int32_t n_min = 0; // minimum number of draft tokens to use for speculative decoding
float p_split = 0.1f; // speculative decoding split probability
float p_min = 0.75f; // minimum speculative decoding probability (greedy)
// ngram-based speculative decoding
uint16_t ngram_size_n = 12; // ngram size for lookup
uint16_t ngram_size_m = 48; // mgram size for speculative tokens
uint16_t ngram_check_rate = 1; // check rate for ngram lookup
uint16_t ngram_min_hits = 1; // minimum hits at ngram/mgram lookup for mgram to be proposed
std::string lookup_cache_static; // path of static ngram cache file for lookup decoding // NOLINT
std::string lookup_cache_dynamic; // path of dynamic ngram cache file for lookup decoding // NOLINT
// draft-model speculative decoding
struct common_params_model mparams_dft;
llama_model * model_dft = nullptr; // a llama_model that can be shared by multiple speculative contexts
llama_context_params cparams_dft; // these are the parameters for the draft llama_context
int32_t n_ctx = 0; // draft context size
int32_t n_gpu_layers = -1; // number of layers to store in VRAM for the draft model (-1 - use default)
ggml_type cache_type_k = GGML_TYPE_F16; // KV cache data type for the K
ggml_type cache_type_v = GGML_TYPE_F16; // KV cache data type for the V
@@ -260,7 +289,14 @@ struct common_params_speculative {
struct cpu_params cpuparams;
struct cpu_params cpuparams_batch;
struct common_params_model model;
std::vector<ggml_backend_dev_t> devices; // devices to use for offloading
std::vector<std::pair<std::string, std::string>> replacements; // main to speculative model replacements
std::vector<llama_model_tensor_buft_override> tensor_buft_overrides;
bool has_dft() const {
return !mparams_dft.path.empty() || !mparams_dft.hf_repo.empty();
}
};
struct common_params_vocoder {
@@ -378,8 +414,6 @@ struct common_params {
std::string path_prompt_cache = ""; // path to file for saving/loading prompt eval state // NOLINT
std::string input_prefix = ""; // string to prefix user inputs with // NOLINT
std::string input_suffix = ""; // string to suffix user inputs with // NOLINT
std::string lookup_cache_static = ""; // path of static ngram cache file for lookup decoding // NOLINT
std::string lookup_cache_dynamic = ""; // path of dynamic ngram cache file for lookup decoding // NOLINT
std::string logits_file = ""; // file for saving *all* logits // NOLINT
// llama-debug specific options
@@ -575,10 +609,6 @@ struct common_params {
// return false from callback to abort model loading or true to continue
llama_progress_callback load_progress_callback = NULL;
void * load_progress_callback_user_data = NULL;
bool has_speculative() const {
return !speculative.model.path.empty() || !speculative.model.hf_repo.empty();
}
};
// call once at the start of a program if it uses libcommon
@@ -714,8 +744,6 @@ struct common_init_result {
std::vector<llama_adapter_lora_ptr> & lora();
void free_context();
private:
struct impl;
std::unique_ptr<impl> pimpl;