Repository navigation
Conversation
- Port paper-code to current PyTorch and Slurm: per-rank GPU selection, env:// init, pure-PyTorch sign packing (replaces bit2byte), torchtext-free WikiText-2 pipeline. - Add run_experiment.py with the paper's hyperparameter presets for Tables 1-7. - Add analyze_results.py comparing runs with the paper's reported numbers. - Add perlmutter/ setup, job and submission scripts. - Add smoke-test results from Perlmutter. Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
…ogging-only trainer patch First GPU results (job 60622124): DDP vs PowerSGD r1/r2/r4 on NanoGPT, 1 node, seed 1. Adds results/, results/RESULTS.md and figures/ (tools/make_results.py), docs/RUNS.md, and the two sbatch files submitted on Perlmutter (pending, no results). train_ddp.py: logging only (best_step, lr_eff, resolved batch/lr in config.json).
… organizer tool results/organized/ holds copies of the raw scp results grouped by Slurm job; MANIFEST.md lists every run with validity (valid, smoke test, partial, suspect collision), best val loss computed from metrics.jsonl, and GPU utilization per CSV. gpt_ddp_2node holds two runs in one metrics.jsonl; each block is attributed by exact match to the runs.jsonl record or the batch .out log. tools/organize_results.py is copy-only and re-runnable. Raw copies (powersgd_results/) are git-ignored.
…files; add real sbatch scripts Re-organized from the final copies (results_batch, logs_batch). The three 2-node runs written by both job 60627319 and job 60623329 are split per job by exact match to runs.jsonl records and the batch .out log. MANIFEST adds long-run, 2-node, GPU-utilization and hours sections. scripts/nanogpt_2node.sbatch and scripts/nanogpt_1node_long.sbatch are the exact files the batch jobs ran. Job 60623330 elapsed and the interactive-CSV utilization are marked provisional.
Brings in the PowerSGD paper reproduction port from paper-reproduction: paper-code fixes for current PyTorch, run_experiment.py, analyze_results.py, perlmutter/ scripts and the smoke-test summary. Also ignores the raw powersgd_results/ copies.
… seeds) Co-authored-by: Cursor <cursoragent@cursor.com>
NanoGPT 1-node results (DDP vs PowerSGD), results generator, queued sbatch files
Merge paper-code reproduction work (Nayan)
… running-average column Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
…n deck Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com> # Conflicts: # .gitignore
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.
No description provided.