mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 28 Packages Projects Releases Wiki Activity

llama: introduce support for model-embedded sampling parameters (#17120)

Browse Source
This commit is contained in:
Aaron Teo
2025-11-25 09:56:07 +08:00
committed by GitHub
parent 3d07caa99b
commit 877566d512
10 changed files with 293 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-arch.cpp
+25 -13
View File
@@ -114,19 +114,31 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
};
static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
{ LLM_KV_GENERAL_TYPE, "general.type" },
{ LLM_KV_GENERAL_ARCHITECTURE, "general.architecture" },
{ LLM_KV_GENERAL_QUANTIZATION_VERSION, "general.quantization_version" },
{ LLM_KV_GENERAL_ALIGNMENT, "general.alignment" },
{ LLM_KV_GENERAL_FILE_TYPE, "general.file_type" },
{ LLM_KV_GENERAL_NAME, "general.name" },
{ LLM_KV_GENERAL_AUTHOR, "general.author" },
{ LLM_KV_GENERAL_VERSION, "general.version" },
{ LLM_KV_GENERAL_URL, "general.url" },
{ LLM_KV_GENERAL_DESCRIPTION, "general.description" },
{ LLM_KV_GENERAL_LICENSE, "general.license" },
{ LLM_KV_GENERAL_SOURCE_URL, "general.source.url" },
{ LLM_KV_GENERAL_SOURCE_HF_REPO, "general.source.huggingface.repository" },
{ LLM_KV_GENERAL_TYPE, "general.type" },
{ LLM_KV_GENERAL_ARCHITECTURE, "general.architecture" },
{ LLM_KV_GENERAL_QUANTIZATION_VERSION, "general.quantization_version" },
{ LLM_KV_GENERAL_ALIGNMENT, "general.alignment" },
{ LLM_KV_GENERAL_FILE_TYPE, "general.file_type" },
{ LLM_KV_GENERAL_SAMPLING_SEQUENCE, "general.sampling.sequence" },
{ LLM_KV_GENERAL_SAMPLING_TOP_K, "general.sampling.top_k" },
{ LLM_KV_GENERAL_SAMPLING_TOP_P, "general.sampling.top_p" },
{ LLM_KV_GENERAL_SAMPLING_MIN_P, "general.sampling.min_p" },
{ LLM_KV_GENERAL_SAMPLING_XTC_PROBABILITY, "general.sampling.xtc_probability" },
{ LLM_KV_GENERAL_SAMPLING_XTC_THRESHOLD, "general.sampling.xtc_threshold" },
{ LLM_KV_GENERAL_SAMPLING_TEMP, "general.sampling.temp" },
{ LLM_KV_GENERAL_SAMPLING_PENALTY_LAST_N, "general.sampling.penalty_last_n" },
{ LLM_KV_GENERAL_SAMPLING_PENALTY_REPEAT, "general.sampling.penalty_repeat" },
{ LLM_KV_GENERAL_SAMPLING_MIROSTAT, "general.sampling.mirostat" },
{ LLM_KV_GENERAL_SAMPLING_MIROSTAT_TAU, "general.sampling.mirostat_tau" },
{ LLM_KV_GENERAL_SAMPLING_MIROSTAT_ETA, "general.sampling.mirostat_eta" },
{ LLM_KV_GENERAL_NAME, "general.name" },
{ LLM_KV_GENERAL_AUTHOR, "general.author" },
{ LLM_KV_GENERAL_VERSION, "general.version" },
{ LLM_KV_GENERAL_URL, "general.url" },
{ LLM_KV_GENERAL_DESCRIPTION, "general.description" },
{ LLM_KV_GENERAL_LICENSE, "general.license" },
{ LLM_KV_GENERAL_SOURCE_URL, "general.source.url" },
{ LLM_KV_GENERAL_SOURCE_HF_REPO, "general.source.huggingface.repository" },
{ LLM_KV_VOCAB_SIZE, "%s.vocab_size" },
{ LLM_KV_CONTEXT_LENGTH, "%s.context_length" },
src/llama-arch.h
+12
View File
@@ -123,6 +123,18 @@ enum llm_kv {
LLM_KV_GENERAL_QUANTIZATION_VERSION,
LLM_KV_GENERAL_ALIGNMENT,
LLM_KV_GENERAL_FILE_TYPE,
LLM_KV_GENERAL_SAMPLING_SEQUENCE,
LLM_KV_GENERAL_SAMPLING_TOP_K,
LLM_KV_GENERAL_SAMPLING_TOP_P,
LLM_KV_GENERAL_SAMPLING_MIN_P,
LLM_KV_GENERAL_SAMPLING_XTC_PROBABILITY,
LLM_KV_GENERAL_SAMPLING_XTC_THRESHOLD,
LLM_KV_GENERAL_SAMPLING_TEMP,
LLM_KV_GENERAL_SAMPLING_PENALTY_LAST_N,
LLM_KV_GENERAL_SAMPLING_PENALTY_REPEAT,
LLM_KV_GENERAL_SAMPLING_MIROSTAT,
LLM_KV_GENERAL_SAMPLING_MIROSTAT_TAU,
LLM_KV_GENERAL_SAMPLING_MIROSTAT_ETA,
LLM_KV_GENERAL_NAME,
LLM_KV_GENERAL_AUTHOR,
LLM_KV_GENERAL_VERSION,
src/llama-model.cpp
+18
View File
@@ -7687,6 +7687,24 @@ int32_t llama_model_meta_count(const llama_model * model) {
return (int)model->gguf_kv.size();
}
const char * llama_model_meta_key_str(llama_model_meta_key key) {
switch (key) {
case LLAMA_MODEL_META_KEY_SAMPLING_SEQUENCE: return "general.sampling.sequence";
case LLAMA_MODEL_META_KEY_SAMPLING_TOP_K: return "general.sampling.top_k";
case LLAMA_MODEL_META_KEY_SAMPLING_TOP_P: return "general.sampling.top_p";
case LLAMA_MODEL_META_KEY_SAMPLING_MIN_P: return "general.sampling.min_p";
case LLAMA_MODEL_META_KEY_SAMPLING_XTC_PROBABILITY: return "general.sampling.xtc_probability";
case LLAMA_MODEL_META_KEY_SAMPLING_XTC_THRESHOLD: return "general.sampling.xtc_threshold";
case LLAMA_MODEL_META_KEY_SAMPLING_TEMP: return "general.sampling.temp";
case LLAMA_MODEL_META_KEY_SAMPLING_PENALTY_LAST_N: return "general.sampling.penalty_last_n";
case LLAMA_MODEL_META_KEY_SAMPLING_PENALTY_REPEAT: return "general.sampling.penalty_repeat";
case LLAMA_MODEL_META_KEY_SAMPLING_MIROSTAT: return "general.sampling.mirostat";
case LLAMA_MODEL_META_KEY_SAMPLING_MIROSTAT_TAU: return "general.sampling.mirostat_tau";
case LLAMA_MODEL_META_KEY_SAMPLING_MIROSTAT_ETA: return "general.sampling.mirostat_eta";
default: return nullptr;
}
}
int32_t llama_model_meta_key_by_index(const llama_model * model, int i, char * buf, size_t buf_size) {
if (i < 0 || i >= (int)model->gguf_kv.size()) {
if (buf_size > 0) {