Skip to content

Actions: devnull37/dimba-lib-exp

Actions

All workflows

Actions

Loading...
Loading

Showing runs from all workflows
64 workflow runs
64 workflow runs

Filter by Workflow

⌥ + click/return to exclude

Filter by Event

⌥ + click/return to exclude

Filter by Status

⌥ + click/return to exclude

Filter by Branch

⌥ + click/return to exclude

Filter by Actor

⌥ + click/return to exclude
Reduce public repository to inference scope CI #64: Commit 53f3fab pushed by devnull37
1m 17s main
main
1m 17s
Fix CUDA graph capture at the denoiser boundary CI #63: Commit ad30664 pushed by devnull37
2m 39s main
main
2m 39s
Rewrite AGENTS.md as the full project-status guide; add CLAUDE.md poi… CI #62: Commit cccf5d8 pushed by devnull37
2m 36s main
main
2m 36s
General speedup pass: CUDA-graph sampler, early commits, chunked CE, … CI #61: Commit e243b98 pushed by devnull37
3m 2s main
main
3m 2s
Optimize training inference and multi-GPU stack CI #60: Commit a1a8e89 pushed by devnull37
2m 32s main
main
2m 32s
Clean up production-run slop before merge (multi-agent review, 17 con… CI #59: Commit b818f6f pushed by devnull37
2m 46s test-time-compute
test-time-compute
2m 46s
Clean up production-run slop before merge (multi-agent review, 17 con… CI #58: Commit b818f6f pushed by devnull37
2m 31s main
main
2m 31s
sync updated system card (288M correction + benchmark + slider + bidi… CI #57: Commit a1d6992 pushed by devnull37
2m 46s test-time-compute
test-time-compute
2m 46s
bidir A/B: double vs shared vs shared+per-direction LoRA (2k steps fr… CI #56: Commit cae7f59 pushed by devnull37
2m 46s test-time-compute
test-time-compute
2m 46s
slider curve: quality vs accuracy vs cost on the 40-question set CI #55: Commit 7b67e0d pushed by devnull37
3m 25s test-time-compute
test-time-compute
3m 25s
benchmark: DIMBA vs SmolLM/SmolLM-Instruct/GPT-2/Pythia (QA, degenera… CI #54: Commit 3318350 pushed by devnull37
2m 43s test-time-compute
test-time-compute
2m 43s
generate.py: accuracy/speed quality slider (steps 16-256 + best-of-N … CI #53: Commit 1d7c4a1 pushed by devnull37
2m 36s test-time-compute
test-time-compute
2m 36s
writeup: remove em dashes per author preference CI #52: Commit 035bcaf pushed by devnull37
2m 40s test-time-compute
test-time-compute
2m 40s
Test-time compute day: repair training, critic head, experiment scrip… CI #51: Commit dae19cb pushed by devnull37
2m 43s test-time-compute
test-time-compute
2m 43s
Make babysitter runbook fully self-contained (launch → monitor → reco… CI #50: Commit 9c53b72 pushed by devnull37
2m 48s main
main
2m 48s
Wire live monitoring + override into training; detailed babysit runbook CI #49: Commit d498512 pushed by devnull37
2m 44s main
main
2m 44s
Set full preset to 50B target + add babysit loop playbook CI #48: Commit f8b4516 pushed by devnull37
2m 44s main
main
2m 44s
Fix audit findings on next-run setup (streaming host-OOM blocker + 5 … CI #47: Commit c79e6eb pushed by devnull37
3m 2s main
main
3m 2s
add train_h100.py: H100 80 GB entry point with staged presets CI #46: Commit 98733a6 pushed by devnull37
2m 44s main
main
2m 44s
distill: add FFN frozen→unfrozen co-adaptation schedule + token-budge… CI #45: Commit a692122 pushed by devnull37
1m 59s main
main
1m 59s
SFT: shrink batch to 32 (OOM headroom) + make SFT/GRPO intermediates … CI #44: Commit afa3349 pushed by devnull37
2m 19s main
main
2m 19s
MLX backend: add flow-matching sampler + per-block FFN (block_ffn) CI #43: Commit d877bd3 pushed by devnull37
2m 47s main
main
2m 47s
add eval_vs_smollm.py: DIMBA vs SmolLM-135M-Instruct evaluation harness CI #42: Commit 72d7776 pushed by devnull37
2m 16s main
main
2m 16s
fix SFT/GRPO crash: drop torch.compile(reduce-overhead) — CUDA Graphs… CI #41: Commit e84b96a pushed by devnull37
2m 29s main
main
2m 29s
distill Stage 3: go to ~1B token-updates over more unique data CI #40: Commit 6d83cab pushed by devnull37
2m 27s main
main
2m 27s