Workflow runs · TUES-AI/Code-JEPA · GitHub

Actions

All workflows
Actions
Loading...
Loading

Showing runs from all workflows

27 workflow runs

27 workflow runs

Prepare Discoverer JAX Siamese multi-GPU run Build & push Code-JEPA training image #27: Commit 7083c7c pushed by DebelToni

10m 28s main

10m 28s

Add multi-GPU Siamese JAX trainer and profiler Build & push Code-JEPA training image #26: Commit e60eadc pushed by DebelToni

10m 18s main

10m 18s

Merge remote-tracking branch 'origin/main' Build & push Code-JEPA training image #25: Commit be36753 pushed by DebelToni

5m 13s main

5m 13s

Fix gradient clipping to include projector parameters Build & push Code-JEPA training image #24: Commit ab9519b pushed by SookX

10m 4s main

10m 4s

Add multi-language DDP support to raw UniXcoder MLM+contrastive pretr… Build & push Code-JEPA training image #23: Commit a12ca90 pushed by SookX

11m 48s main

11m 48s

Merge branch 'main' of github.com:TUES-AI/Code-JEPA Build & push Code-JEPA training image #22: Commit 02a1e2c pushed by DebelToni

10m 14s main

10m 14s

Add projection head so rank/JEPA/SIGReg losses train on z, not h Build & push Code-JEPA training image #21: Commit 45f88b9 pushed by SookX

10m 39s main

10m 39s

Add raw CodeSearchNet UniXcoder baseline Build & push Code-JEPA training image #20: Commit 83fded1 pushed by SookX

11m 17s main

11m 17s

Use Hugging Face BigCloneBench staging Build & push Code-JEPA training image #19: Commit 160e21a pushed by SookX

10m 48s main

10m 48s

Load POJ-104 from HuggingFace when local jsonl files are missing Build & push Code-JEPA training image #18: Commit 3856519 pushed by SookX

9m 49s main

9m 49s

Add DDP to fine-tuning: torchrun --nproc_per_node=2 per task, sequent… Build & push Code-JEPA training image #17: Commit 0fa4a99 pushed by SookX

10m 51s main

10m 51s

Fix DDP: find_unused_parameters=True instead of static_graph=True Build & push Code-JEPA training image #16: Commit d668ccf pushed by SookX

11m 48s main

11m 48s

Fix DDP crash: static_graph=True, skip gradient_checkpointing in dist… Build & push Code-JEPA training image #15: Commit 69a8df0 pushed by SookX

10m 2s main

10m 2s

Add DDP support to pretraining; pretrain scripts use torchrun --nproc… Build & push Code-JEPA training image #14: Commit fd694dd pushed by SookX

11m 1s main

11m 1s

v0-only runs, separate SLURM jobs, better gradient logging Build & push Code-JEPA training image #13: Commit 386866b pushed by SookX

10m 57s main

10m 57s

Fix sampler: load all views globally instead of per-shard pairing Build & push Code-JEPA training image #12: Commit 19602b6 pushed by SookX

9m 42s main

9m 42s

Add eta_h to training logs, fix GradScaler deprecation warning Build & push Code-JEPA training image #11: Commit 4cb3b29 pushed by SookX

9m 54s main

9m 54s

Fix SmallUniXcoder attention mask for transformers >=5.x Build & push Code-JEPA training image #10: Commit 11b1913 pushed by SookX

11m 14s main

11m 14s

Fix SDPA 5D causal mask crash on newer transformers Build & push Code-JEPA training image #9: Commit 5c415d2 pushed by SookX

11m 16s main

11m 16s

Set up small UniXcoder offline benchmark workflow Build & push Code-JEPA training image #8: Commit 9e194ed pushed by SookX

10m 41s main

10m 41s

Record full-corpus CodeSearchNet BPE tokenizers Build & push Code-JEPA training image #7: Commit 7ed261a pushed by DebelToni

13m 35s main

13m 35s

docker cicd better, peromance jax trianing, docs around ablations Build & push Code-JEPA training image #6: Commit 1b9fe57 pushed by DebelToni

1m 41s main

1m 41s

Add production JAX tokenized training path Build & push Code-JEPA training image #5: Commit 821a430 pushed by DebelToni

12m 50s main

12m 50s

Always refresh latest training artifacts on S3 Build & push Code-JEPA training image #4: Commit d117a8f pushed by DebelToni

15m 40s main

15m 40s

Fix s5cmd sync wildcard handling Build & push Code-JEPA training image #3: Commit 32af173 pushed by DebelToni

14m 33s main

14m 33s