Skip to content

Promote reference-runs v1.0 backfill to main - #203

Merged
KraHsu merged 2 commits into
mainfrom
dev/docs-reference-numbers
Jun 1, 2026
Merged

KraHsu merged 2 commits into
mainfrom
dev/docs-reference-numbers

Conversation

@KraHsu

@KraHsu KraHsu commented Jun 1, 2026

Copy link
Copy Markdown
Owner

Summary

Promotes the reference-runs doc backfill (#202) from dev/docs-reference-numbers to main. All five bundled tasks now have their v1.0 cells filled from the actual 3-seed sweep (H200 for cartpoles + Franka, 4× RTX 4090 for G1).

See #202 for the per-seed numbers and v0.4.7 comparison.

Test plan

  • mkdocs --strict clean on en + zh
  • Diff is doc-only (no code changes)

KraHsu added 2 commits June 1, 2026 10:17
Fills the TBD cells in all five bundled-task tables (IP, DIP, G1-Velocity-Flat,
G1-Tracking-Flat, Franka-Pick-And-Place) with the converged v1.0 reference
numbers from a full 3-seed sweep:

* Cartpoles + Franka collected on a single NVIDIA H200 (cartpoles: parallel=3
  on GPU 0; Franka: one seed per GPU across GPUs 0-2). QD_GRAPH=0 enforced
  per the Hopper note already in the doc.
* G1-Velocity-Flat + G1-Tracking-Flat collected on a 4x RTX 4090 host. Vel
  ran one seed per GPU on GPUs 0-2 with Track seed 1 sharing the host on
  GPU 3 (Phase A); Track seeds 2/3 ran on GPUs 0/1 once Vel finished
  (Phase B).

Replaces the "Status: v1.0 sanity smoke, full sweep pending" admonition
with a hardware/driver/concurrency footer that mirrors the doc's own
warning: the G1 numbers below are the concurrent-run wall, not a solo-run
figure (rewards are deterministic and unaffected, but wall-clock is ~3-6x
longer than a solo run per rsl_rl's Time elapsed counter in train.log).

Highlights vs v0.4.7:
* G1-Velocity-Flat is meaningfully tighter across seeds (112.04 / 112.87 /
  113.16 vs v0.4.7's 112.42 / 93.42 / 92.03 -- v0.4.7 had two seeds drop
  ~17% below seed 1, v1.0 lands all three in the same band).
* G1-Tracking-Flat matches v0.4.7 within noise (138.44 / 137.98 / 138.06
  vs 137.80 / 138.05 / 138.12).
* Franka mean success_rate slightly higher and tighter (0.983 +/- 0.012
  vs 0.963 +/- 0.052) -- no outlier seed.
Backfill reference-runs docs with v1.0 numbers
@KraHsu
KraHsu merged commit 5acf1a3 into main Jun 1, 2026
4 checks passed
@KraHsu
KraHsu deleted the dev/docs-reference-numbers branch June 1, 2026 03:19
KraHsu added a commit that referenced this pull request Jun 5, 2026
Promote reference-runs v1.0 backfill to main
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant