Repository navigation
One-command GPU run for the transformer grids - #33
Merged
Merged
Conversation
scripts/gpu/run_neural.sh builds a CUDA environment, rebuilds and checks the pinned corpus, runs the DistilBERT and DeBERTa grids, predicts with every run and packs the results; scripts/gpu/score_neural.sh scores them here. train and experiment take --dev-gold, so early stopping uses only the model-drafted dev batches and the hand-labelled dev documents stay out of model selection.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
scripts/gpu/run_neural.sh(Kaggle, Colab or any CUDA box):TORCH_INDEXfallback for older drivers;scripts/gpu/score_neural.sh RESULTS.tgz(development machine): agreement and hand-labelled reports for every run, plus paired comparisons of each grid's best run againstcrf-batches-01-02. Tested end to end with a stand-in run.train/experiment --dev-gold: early stopping and the experiment table use only the given files' dev records. The runbook passes the model-drafted batches 03 and 06, so the 10 hand-labelled dev documents never influence model selection. New test.docs/training_on_gpu.mdnow opens with the one-command run (Kaggle steps); the manual Colab notes are kept. README usage updated.The test split is not touched. Not yet run on a GPU.
Checks
ruff check,ruff format --check,mypy src/,pytest -m "not slow"(234 passed);shellcheckclean on both scripts; both grids dry-run to 6 runs each; Linux dependency resolution gives torch 2.14.1 with the CUDA 13 runtime.