Repository navigation
fix(training+eval): three defects found by the first real eval run - #26
Merged
Merged
Conversation
- _messages_dataset tolerates a 0-record split (tiny datasets emit them legitimately); the empty-TRAIN guard stays explicit. - task-eval batch builder: one single-message conversation per prompt; the flat list merged a whole batch into one sequence, so a 4-prompt batch produced a single 1965-token input and one answer for all. - load_student_model resolves Hub model ids (base-model side of task-eval) via snapshot_download; local absolute paths that do not exist still fail fast.
A malformed Hub id (extra path segments) raised a raw HFValidationError traceback through the CLI; model resolution failures now surface as clean BrainforgeError messages.
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
All three found by the first real GPU eval run (RTX 3080, Qwen3-1.7B base vs QLoRA student on the post-cutoff CTF holdout).
Issue for this PR
N/A
Checklist