diff options
| author | Jared Van Bortel <jared@nomic.ai> | 2024-04-09 13:44:08 -0400 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-04-09 13:44:08 -0400 |
| commit | 1b67731e184e27a465b8c5476061294a4af668ea (patch) | |
| tree | 15a2d877029fb509a34e462c227475bc7d6dc31e /examples/llava | |
| parent | c4a3a4ff47d62d2503ddf9bd91b58c21f04fe3c3 (diff) | |
BERT tokenizer fixes (#6498)
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
Diffstat (limited to 'examples/llava')
| -rw-r--r-- | examples/llava/llava-cli.cpp | 3 |
1 files changed, 1 insertions, 2 deletions
diff --git a/examples/llava/llava-cli.cpp b/examples/llava/llava-cli.cpp index e29da6cb..75948806 100644 --- a/examples/llava/llava-cli.cpp +++ b/examples/llava/llava-cli.cpp @@ -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); |