Skip to content

common : allow raw byte in SPM vocabs; don't crash if newline token is not found #5478

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Feb 13, 2024
Merged

common : allow raw byte in SPM vocabs; don't crash if newline token is not found #5478

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
93aed75
common : don't crash if newline token is not found
akx Feb 13, 2024
72b353f
common : llama_byte_to_token: allow falling back to finding just the …
akx Feb 13, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 13 additions & 2 deletions llama.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3314,7 +3314,12 @@ static void llm_load_vocab(

// determine the newline token: LLaMA "<0x0A>" == 10 == '\n', Falcon 193 == '\n'
if (vocab.type == LLAMA_VOCAB_TYPE_SPM) {
vocab.linefeed_id = llama_byte_to_token(vocab, '\n');
try {
vocab.linefeed_id = llama_byte_to_token(vocab, '\n');
} catch (const std::exception & e) {
LLAMA_LOG_WARN("%s: SPM vocabulary, but newline token not found: %s! Using special_pad_id instead.", __func__, e.what());
vocab.linefeed_id = vocab.special_pad_id;
}
} else if (vocab.type == LLAMA_VOCAB_TYPE_WPM) {
vocab.linefeed_id = vocab.special_pad_id;
} else {
Expand Down Expand Up @@ -7746,7 +7751,13 @@ static llama_token llama_byte_to_token(const llama_vocab & vocab, uint8_t ch) {
switch (llama_vocab_get_type(vocab)) {
case LLAMA_VOCAB_TYPE_SPM: {
const char buf[7] = { '<', '0', 'x', hex[ch >> 4], hex[ch & 15], '>', 0 };
return vocab.token_to_id.at(buf);
auto token = vocab.token_to_id.find(buf);
if (token != vocab.token_to_id.end()) {
return (*token).second;
}
// Try to fall back to just the byte as a string
const char buf2[2] = { (char)ch, 0 };
return vocab.token_to_id.at(buf2);
Comment on lines +7758 to +7760
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Does this fallback work for the Finnish model that you are trying?

Copy link
Contributor Author

@akx akx Feb 13, 2024 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yep! @R4ZZ3, who's been involved with training the model, pointed out that tokenizer.json has

    {
      "id": 64261,
      "content": "\n",
      "single_word": false,
      "lstrip": false,
      "rstrip": false,
      "normalized": false,
      "special": true
    }

}
case LLAMA_VOCAB_TYPE_WPM:
case LLAMA_VOCAB_TYPE_BPE: {
Expand Down