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

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
+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,