tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-27 12:05:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

BERT tokenizer fixes (#6498)

Browse Source 
Key changes:
* BERT conversion: fix abuse of LlamaHfVocab, do not set BOS or EOS
* Nomic Embed conversion: pad vocab instead of slicing embedding tensor
* llama_tokenize: handle added special tokens like HF does
This commit is contained in:
Jared Van Bortel
2024-04-09 13:44:08 -04:00
committed by GitHub
parent c4a3a4ff47
commit 1b67731e18
20 changed files with 221 additions and 194 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
examples/llava/llava-cli.cpp
View File

@ -146,7 +146,6 @@ static void process_prompt(struct llava_context * ctx_llava, struct llava_image_
int n_past = 0;
const int max_tgt_len = params->n_predict < 0 ? 256 : params->n_predict;
const bool add_bos = llama_should_add_bos_token(llama_get_model(ctx_llava->ctx_llama));
std::string system_prompt, user_prompt;
size_t image_pos = prompt.find("<image>");
@ -180,7 +179,7 @@ static void process_prompt(struct llava_context * ctx_llava, struct llava_image_
}
}
eval_string(ctx_llava->ctx_llama, system_prompt.c_str(), params->n_batch, &n_past, add_bos);
eval_string(ctx_llava->ctx_llama, system_prompt.c_str(), params->n_batch, &n_past, true);
llava_eval_image_embed(ctx_llava->ctx_llama, image_embed, params->n_batch, &n_past);
eval_string(ctx_llava->ctx_llama, user_prompt.c_str(), params->n_batch, &n_past, false);