Bug fix for issue 184#189
Open
marcromeyn wants to merge 1 commit into
Open
Conversation
Signed-off-by: Marc Romeyn <marcromeyn@gmail.com>
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fixes: #184
SFT data preparation was filtering out a large fraction of rows because some chat templates emit a trailing newline after the generation prompt that does not appear at the matching position in the full template, causing the strict prefix check in split_template_into_messages to raise ValueError and the row to be dropped (issue #184). The fix replaces the strict-only check with a strict-then-fallback pattern: the original equality check still runs first, so any row the upstream materialize.py processed cleanly produces byte-identical chunks; only when that check fails do we try template_up_to_here.rstrip() and accept the row if the stripped form matches. While in the file, we also harden find_last_user_message_end and split_template_into_messages against conversations with no user message or a trailing user turn — both previously crashed with bare ValueError/IndexError from internal indexing — by raising typed, informative ValueErrors instead. New regression tests cover the issue #184 recovery, byte-identical output on well-behaved templates, and the new input guards.