Move beam search in case of chat scenario to sampler.cpp #1215
+171
−587
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
github-actions
bot
added
category: visual language
Wovchena
requested changes
Nov 15, 2024
Wovchena
reviewed
Nov 19, 2024
github-actions
bot
added
no-match-files
and removed
category: sampling
|
the tests are passed, but it still doesn't work for bigger max_token size for TinyLlama-1.1B-Chat-v1.0 . Tokenizer converts first token from |
- spread lenght penalty to multinomial
| } | ||
| m_templated_chat_history = new_templated_chat_history; | ||
| m_tokenized_chat_history = new_chat_tokens; | ||
| // TODO: Forbid LoRA config change if we are in the chat mode, because it requires regenerating the history with LoRA applied | ||
| } else { | ||
| encoded_input = m_tokenizer.encode(prompt); |
There was a problem hiding this comment.
I'm not sure that m_templated_chat_history.append(answer); is valid for all chat templates, because between current history and assistant answer we can have some tokens / words (e.g. ' ' in example below):
{% set content = message['content'] %}
{% endif %}
{% if message['role'] == 'user' %}{{ bos_token + '[INST] ' + content.strip() + ' [/INST]' }}
{% elif message['role'] == 'assistant' %}{{ ' ' + content.strip() + ' ' + eos_token }}
Let's discuss it on GenAI meeting.
sbalandi
force-pushed
the
beam
branch
2 times, most recently
from
815016b
to
f68e9db
Compare
|
|
||
| if (is_chat_conversation) { | ||
| m_history_available = true; | ||
| m_last_disappeared_token = result.tokens[0].back(); |
There was a problem hiding this comment.
I suppose EOS can also be added here, but it's OK for such token to disappear
can we extract from sampler the reason of sequence finishing and add last token only iff it's ended by not EOS condition ?
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.
Task CVS-156578