sleepy/llama.cpp
1
0
Fork 0
Code Issues 11 Pull Requests Actions 108 Packages Projects Releases Wiki Activity

spec : fix vocab compat checks (#22358)

Browse Source
This commit is contained in:
Georgi Gerganov
2026-04-25 20:11:35 +03:00
committed by GitHub
parent 9725a313be
commit 98dc1418ea
1 changed files with 17 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files
common/speculative.cpp
+17 -9
View File
@@ -61,18 +61,26 @@ static bool common_speculative_are_compatible(
LOG_DBG("%s: vocab_type dft: %d\n", __func__, vocab_type_dft); LOG_DBG("%s: vocab_type dft: %d\n", __func__, vocab_type_dft);
if (vocab_type_tgt != vocab_type_dft) { if (vocab_type_tgt != vocab_type_dft) {
LOG_DBG("%s: draft model vocab type must match target model to use speculation but ", __func__); LOG_WRN("%s: draft model vocab type must match target model to use speculation but "
LOG_DBG("vocab_type_dft = %d while vocab_type_tgt = %d\n", vocab_type_dft, vocab_type_tgt); "vocab_type_dft = %d while vocab_type_tgt = %d\n", __func__, vocab_type_dft, vocab_type_tgt);
return false; return false;
} }
if ( if (llama_vocab_get_add_bos(vocab_tgt) != llama_vocab_get_add_bos(vocab_dft) ||
llama_vocab_get_add_bos(vocab_tgt) != llama_vocab_get_add_bos(vocab_dft) || (llama_vocab_get_add_bos(vocab_tgt) && llama_vocab_bos(vocab_tgt) != llama_vocab_bos(vocab_dft))) {
llama_vocab_get_add_eos(vocab_tgt) != llama_vocab_get_add_eos(vocab_dft) || LOG_WRN("%s: draft model bos tokens must match target model to use speculation. add: %d - %d, id: %d - %d)\n",
llama_vocab_bos(vocab_tgt) != llama_vocab_bos(vocab_dft) || __func__,
llama_vocab_eos(vocab_tgt) != llama_vocab_eos(vocab_dft) llama_vocab_get_add_bos(vocab_tgt), llama_vocab_get_add_bos(vocab_dft),
) { llama_vocab_bos(vocab_tgt), llama_vocab_bos(vocab_dft));
LOG_DBG("%s: draft model special tokens must match target model to use speculation\n", __func__); return false;
}
if (llama_vocab_get_add_eos(vocab_tgt) != llama_vocab_get_add_eos(vocab_dft) ||
(llama_vocab_get_add_eos(vocab_tgt) && llama_vocab_eos(vocab_tgt) != llama_vocab_eos(vocab_dft))) {
LOG_WRN("%s: draft model eos tokens must match target model to use speculation. add: %d - %d, id: %d - %d)\n",
__func__,
llama_vocab_get_add_eos(vocab_tgt), llama_vocab_get_add_eos(vocab_dft),
llama_vocab_eos(vocab_tgt), llama_vocab_eos(vocab_dft));
return false; return false;
} }