Implement non-greedy tokenizer that tries to maximize token lengths #242
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Although I haven't examined the code, I've tested it on several prompts and can already conclude that this patch allows Llama to write in French. |
thement
force-pushed
the
thement/fix-tokenizer
branch
from
e40d4e0
to
3838e51
Compare
j-f1
reviewed
Mar 17, 2023
Piezoid
reviewed
Mar 17, 2023
| @@ -846,6 +846,7 @@ int main(int argc, char ** argv) { | |||
| std::vector<float> logits; | |||
|
|
|||
| // tokenize the prompt | |||
| params.prompt.insert(0, 1, ' '); | |||
There was a problem hiding this comment.
Is the space meant to be a separate token? I noticed that it often get fused with the first user provided token.
There was a problem hiding this comment.
It should be fused to the first token! This is how original python llama code parses it.
I can dig out more details if you want.
ggerganov
approved these changes
Mar 17, 2023
- this is to match original llama tokenizer behavior
thement
force-pushed
the
thement/fix-tokenizer
branch
from
3838e51
to
7566d1a
Compare
mudler
pushed a commit
to go-skynet/llama
that referenced
this pull request
Mar 17, 2023
…gerganov#242) * Implement non-greedy tokenizer that tries to maximize token lengths * Insert single space in front of the prompt - this is to match original llama tokenizer behavior --------- Co-authored-by: Jakub Horak <jakub.horak@ibawizard.net>
antimatter15
pushed a commit
to antimatter15/alpaca.cpp
that referenced
this pull request
Mar 18, 2023
…gerganov#242) * Implement non-greedy tokenizer that tries to maximize token lengths * Insert single space in front of the prompt - this is to match original llama tokenizer behavior --------- Co-authored-by: Jakub Horak <jakub.horak@ibawizard.net>
gyohng
pushed a commit
to gyohng/alpaca.cpp
that referenced
this pull request
Mar 26, 2023
…gerganov#242) * Implement non-greedy tokenizer that tries to maximize token lengths * Insert single space in front of the prompt - this is to match original llama tokenizer behavior --------- Co-authored-by: Jakub Horak <jakub.horak@ibawizard.net>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.