From b97bc3966e852adb626c90be64fd48282800f504 Mon Sep 17 00:00:00 2001 From: Pedro Cuenca Date: Sun, 21 Apr 2024 13:50:41 +0200 Subject: llama : support Llama 3 HF conversion (#6745) * Support Llama 3 conversion The tokenizer is BPE. * style * Accept suggestion Co-authored-by: Sourab Mangrulkar <13534540+pacman100@users.noreply.github.com> * llama : add llama_token_is_eog() ggml-ci * llama : auto-detect more EOT tokens when missing in KV data * convert : replacing EOS token is a hack * llama : fix codegemma EOT token + add TODOs * llama : fix model type string for 8B model --------- Co-authored-by: Sourab Mangrulkar <13534540+pacman100@users.noreply.github.com> Co-authored-by: Georgi Gerganov --- examples/beam-search/beam-search.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'examples/beam-search') diff --git a/examples/beam-search/beam-search.cpp b/examples/beam-search/beam-search.cpp index 866c6d7a..3d34378a 100644 --- a/examples/beam-search/beam-search.cpp +++ b/examples/beam-search/beam-search.cpp @@ -47,7 +47,7 @@ struct beam_search_callback_data { // In this case, end-of-beam (eob) is equivalent to end-of-sentence (eos) but this need not always be the same. // For example, eob can be flagged due to maximum token length, stop words, etc. static bool is_at_eob(const beam_search_callback_data & callback_data, const llama_token * tokens, size_t n_tokens) { - return n_tokens && tokens[n_tokens-1] == llama_token_eos(llama_get_model(callback_data.ctx)); + return n_tokens && llama_token_is_eog(llama_get_model(callback_data.ctx), tokens[n_tokens-1]); } // Function matching type llama_beam_search_callback_fn_t. -- cgit v1.2.3